CaltechTHESIS advisor: Monograph
https://feeds.library.caltech.edu/people/Burdick-J-W/combined_advisor.rss
A Caltech Library Repository Feedhttp://www.rssboard.org/rss-specificationpython-feedgenenWed, 26 Jun 2024 12:51:55 -0700Theory and Applications of Hyper-Redundant Robotic Manipulators
https://resolver.caltech.edu/CaltechETD:etd-11082006-132210
Year: 1992
DOI: 10.7907/F12D-0X25
The term "hyper-redundant" refers to robotic manipulators and mobile robots with a very large, possibly infinite, number of actuatable degrees of freedom. These robots are analogous in morphology and operation to snakes, worms, elephant trunks, and tentacles. This thesis presents a novel kinematic framework for hyper-redundant manipulator motion planning and task implementation. The basis of this formulation is the use of a "backbone reference set" which captures the essential macroscopic geometric features of hyper-redundant robots. In the analytical part of this work, the backbone representation is developed and used to solve problems in obstacle avoidance, locomotion, grasping, and "optimal" end effector placement. The latter part of this thesis deals with the design and implementation of a thirty-degree-of-freedom planar hyper-redundant manipulator which is used to demonstrate these novel kinematic and motion planning techniques. Design issues such as robustness with respect to mechanical failure, and design for easy assembly and repair are also addressed. The analytical and design concepts are combined to illustrate tasks for which hyper-redundant robotic mechanisms are well suited.
https://resolver.caltech.edu/CaltechETD:etd-11082006-132210TRUST : a new global optimization methodology, application to artificial neural networks, and analog VLSI implementation
https://resolver.caltech.edu/CaltechETD:etd-10192005-153248
Year: 1994
DOI: 10.7907/gf87-7y12
A new method for unconstrained global function optimization, acronymed TRUST, is introduced. This method formulates optimization as the solution of a deterministic dynamical system incorporating terminal repellers and a novel subenergy tunneling function. Benchmark tests comparing this method to other global optimization procedures are presented, and the TRUST algorithm is shown to be substantially faster.
This algorithm is provably convergent to the global minimum for objective functions of one variable. Theoretically, convergence to a global solution is not guaranteed in the multi-dimensional case. However, in practical applications, TRUST has found the global minimum in all multi-dimensional benchmark functions as a result of its global descent property. The TRUST formulation leads to a simple stopping criterion.
The algorithm is also applied to Backpropagation learning in artificial neural networks in order to overcome the susceptibility to local minima during training, which is associated with gradient descent. TRUST (or Global Descent in this context) was proposed as a candidate for replacing gradient descent in order to eliminate the local minima problem. We test the ability of the new dynamical system to overcome local minima with common benchmark examples and a pattern recognition example. The results demonstrate that the new method does indeed escape encountered local minima, and in most cases converges to the globally optimal solution of a specific problem.
The structure of the TRUST's equations enables an implementation of the algorithm in analog VLSI hardware for further substantial speed enhancement. We have designed, fabricated and tested a terminal repeller circuit and a gradient descent circuit, which constitute the main components of the TRUST's dynamics. Measured chip data, which confirmed the efficient performance of these circuits, are presented. We have also designed a novel global optimization circuit which incorporates the above circuits with additional control logic. This circuit implements the TRUST algorithm, and thus locates the global minimum of arbitrary one-dimensional objective functions. Simulated experiments of this circuit are thoroughly discussed. The convergence time required for the circuit to converge to the global minimum is remarkably at the order of micro-seconds.https://resolver.caltech.edu/CaltechETD:etd-10192005-153248Theory and Applications of Modular Reconfigurable Robotic Systems
https://resolver.caltech.edu/CaltechETD:etd-10202005-090745
Year: 1994
DOI: 10.7907/2AAA-RY45
A modular reconfigurable robotic system consists of various link and joint units with standardized connecting interfaces that can be easily separated and reassembled into different configurations. Compared to a fixed configuration robot, which is usually a compromised design for a limited set of tasks, a modular robot can accomplish a large class of tasks through reconfiguration of a small inventory of modules. This thesis studies how to find an optimal module assembly configuration constructed from a given inventory of module components for a specific task. A set of generalized module models that bear features found in many real implementations is introduced. The modular robot assembly configuration is represented by a novel Assembly Incidence Matrix (AIM). Equivalence relations based on module geometry symmetries and graph isomorphisms are defined on the AIMs. An enumeration algorithm to generate non-isomorphic assembly configurations based on this equivalence relation is proposed. Examples demonstrate that this method is a significant improvement over a brute force enumeration process. Configuration independent kinematic models for modular robots are developed, and they are essential for solving the task-optimal configuration problem. A task-oriented objective function is defined on the set of non-isomorphic module assembly configurations. Task requirements and kinematic constraints on the robot assembly are treated as parameters to this objective function. The task-optimal configuration problem is formulated as a combinatorial optimization problem to which genetic algorithms are employed for solutions. Examples of finding task-optimal serial revolute-jointed robot configurations are demonstrated. In addition, the applications of modular robots to planning multifinger grasping and manipulation are developed. Planning two-finger grasps is done through finding antipodal point grasps on smooth shaped objects. Planning n-finger grasps is achieved by defining a qualitative force-closure test function on the n-finger grasps on an object. Applications of this test function to manipulation task and finger gaiting are illustrated.https://resolver.caltech.edu/CaltechETD:etd-10202005-090745The mechanics and control of undulatory robotic locomotion
https://resolver.caltech.edu/CaltechETD:etd-10202005-153514
Year: 1996
DOI: 10.7907/Y1TF-RF86
In this dissertation, we examine a formulation of problems of undulatory robotic locomotion within the context of mechanical systems with nonholonomic constraints and symmetries. Using tools from geometric mechanics, we study the underlying structure found in general problems of locomotion. In doing so, we decompose locomotion into two basic components: internal shape changes and net changes in position and orientation. This decomposition has a natural mathematical interpretation in which the relationship between shape changes and locomotion can be described using a connection on a trivial principal fiber bundle.
We begin by reviewing the processes of Lagrangian reduction and reconstruction for unconstrained mechanical systems with Lie group symmetries, and present new formulations of this process which are easily adapted to accommodate external constraints. Additionally, important physical quantities such as the mechanical connection and reduced mass-inertia matrix can be trivially determined using this formulation. The presence of symmetries then allows us to reduce the necessary calculations to simple matrix manipulations.
The addition of constraints significantly complicates the reduction process; however, we show that for invariant constraints, a meaningful connection can be synthesized by defining a generalized momentum representing the momentum of the system in directions allowed by the constraints. We then prove that the generalized momentum and its governing equation possess certain invariances which allows for a reduction process similar to that found in the unconstrained case. The form of the reduced equations highlights the synthesized connection and the matrix quantities used to calculate these equations.
The use of connections naturally leads to methods for testing controllability and aids in developing intuition regarding the generation of various locomotive gaits. We present accessibility and controllability tests based on taking derivatives of the connection, and relate these tests to taking Lie brackets of the input vector fields.
The theory is illustrated using several examples, in particular the examples of the snakeboard and Hirose snake robot. We interpret each of these examples in light of the theory developed in this thesis, and examine the generation of locomotive gaits using sinusoidal inputs and their relationship to the controllability tests based on Lie brackets.
https://resolver.caltech.edu/CaltechETD:etd-10202005-153514Sensor Based Motion Planning: The Hierarchical Generalized Voronoi Graph
https://resolver.caltech.edu/CaltechETD:etd-12182007-090504
Year: 1996
DOI: 10.7907/49ee-a204
Sensor based motion planning incorporates sensor information reflecting the state of a robot's environment into its planning process, whereas traditional approaches assume complete prior knowledge of the robot's environment. Recent research has focused on the development and incremental construction of the hierarchical generalized Voronoi graph (HGVG), which is a concise representation of a robot's environment. The HGVG is advantageous in that it lends itself to sensor based construction in a rigorous and provably correct manner. With this approach, a robot can enter an unknown environment, incrementally construct the HGVG, and then use the HGVG for future excursions in the environment. Simulations and experiments validate this approach.https://resolver.caltech.edu/CaltechETD:etd-12182007-090504Robotic manipulation with flexible link fingers
https://resolver.caltech.edu/CaltechETD:etd-01172008-092801
Year: 1997
DOI: 10.7907/RVP5-Q254
Robots with structural flexibility provide an attractive alternative to rigid robots for many of the new and evolving applications in robotics. In certain applications their use is unavoidable. The increased complexity in modeling and control of such robots is offset by desirable performance enhancements in some respects. In this thesis we present a singular perturbation approach for modeling, analysis and control of robots with flexibility. As our singular perturbation approach does not treat the flexible manipulator as a perturbation of the rigid manipulator, it can treat significant flexibility, beyond the linear range. Analysis based on this approach leads to some provably stable control laws for the hybrid position and force control of flexible-link manipulators. The analysis is done in the framework of a single robot manipulator in a constrained motion task. Simulations and experimental results are presented for this case. To show applicability of the results to more general and complex systems with flexibilities we also present experimental data from a planar, two-fingered, reconfigurable grasping setup which allows rigid and flexible configurations. The aim of the experimentation is to show the applicability of the control laws and analysis developed, and to determine the performance enhancements resulting from the introduction of flexibility. Experimental data is analysed to show the tradeoffs between controller complexity and performance enhancement as we deal with greater flexibility. Various performance criteria are set up and experimental results are discussed within their framework. We conclude that large flexibility can be controlled without too much additional effort, has performance comparable to that of rigid robots, and possesses enhancing properties which make it appealing for use in certain types of applications.https://resolver.caltech.edu/CaltechETD:etd-01172008-092801Trajectory generation for nonlinear control systems
https://resolver.caltech.edu/CaltechETD:etd-01172008-085534
Year: 1997
DOI: 10.7907/9X7P-A431
This thesis explores the paradigm of two degree of freedom design for nonlinear control systems. In two degree of freedom design one generates an explicit trajectory for state and input around which the system is linearized. Linear techniques are then used to stabilize the system around the nominal trajectory and to deal with uncertainty. This approach allows the use of the wealth of tools in linear control theory to stabilize a system in the face of uncertainty, while exploiting the non-linearities to increase performance. Indeed, this thesis shows through simulations and experiments that the generation of a nominal trajectory allows more aggressive tracking in mechanical systems.
The generation of trajectories for general systems involves the solution of two point boundary value problems which are hard to solve numerically. For the special class of differentially flat systems there exists a unique correspondence between trajectories in the output space and the full state and input space. This allows us to generate trajectories in the lower dimensional output, space where we don't have differential constraints, and subsequently map these to the full state and input space through an algebraic procedure. No differential equations have to be solved in this process. This thesis gives a definition of differential flatness in terms of differential geometry, and proves some properties of flat systems. In particular, it is shown that differential flatness is equivalent to dynamic feedback linearizability in an open and dense set.
This dissertation focuses on differentially flat systems. We describe some interesting trajectory generation problems for these systems, and present software to solve them. We also present algorithms and software for real time trajectory generation, that allow a computational tradeoff between stability and performance. We prove convergence for a rather general class of desired trajectories. If a system is not differentially flat we can approximate it with a differentially flat system, and extend the techniques for flat systems. The various extensions for approximately flat systems are validated in simulation and experiments on a thrust vectored aircraft. A system may exhibit a two layer structure where the outer layer is a flat system, and the inner system is not. We call this structure outer flatness. We investigate trajectory generation for these systems and present theorems on the type of tracking we can achieve. We validate the outer flatness approach on a model helicopter in simulations and experiment.
https://resolver.caltech.edu/CaltechETD:etd-01172008-085534Control of stratified systems with robotic applications
https://resolver.caltech.edu/CaltechETD:etd-01232008-144001
Year: 1998
DOI: 10.7907/49h9-q898
Many interesting and important control systems evolve on stratified configuration spaces. Roughly speaking, a configuration manifold is called "stratified" if it contains subspaces (submanifolds) upon which the system had different equations of motion. Robotic systems, in particular, are of this nature. For example, a legged robot has discontinuous equations of motion near points in the configuration space where each of its "feet" comes into contact with the ground. In such a case, when the system moves from one submanifold to another, the equations of motion change in a non-smooth, or even discontinuous manner. In such cases, traditional nonlinear control methodologies are inapplicable because they generally rely upon some form of differentiation. Yet, it is precisely the discontinuous nature of such systems that is often their most important characteristic.
This dissertation presents methods which consider the intrinsic physical geometric structure present in such problems to address nonlinear controllability and motion planning for stratified systems. For both problems, by exploiting this geometric structure of stratified systems, we can extend standard nonlinear control results and methodologies to the stratified case. A related problem addressed by this dissertation is that of controllability of systems where some control inputs are constrained to be non-negative. This problem arises in stratified systems which arise by way of physical contact because the normal force between contacting systems must be nonnegative. For all the results, a basic goal is to generate results which are general. For example, for robotics applications, these results are independent of a particular robot's number of legs, fingers or morphology.
https://resolver.caltech.edu/CaltechETD:etd-01232008-144001Mechanics and Planning of Workpiece Fixturing and Robotic Grasping
https://resolver.caltech.edu/CaltechETD:etd-01302008-111854
Year: 1998
DOI: 10.7907/1d4m-j065
<p>This thesis addresses several key issues in mechanics and automated planning of workpiece fixturing and robotic grasping, including accurate and efficient modelling of compliance, well-defined and practically useful quality measures, and well-defined kinematic metric functions for rigid bodies.</p>
<p>The accurate and efficient modelling of compliant fixtures and grasps is considered. A stiffness matrix formula is derived using the overlap compliance representation for quasi-rigid bodies. In contrast to existing approaches using the linear contact model, this formula is well-suited to automated planning algorithms since it can incorporate realistic nonlinear contact models (e.g., the classical Hertz model), and can be directly computed from CAD data on basic geometric and material properties of the bodies. The formula is then used as a basis for a systematic analysis of local curvature effects on fixture stability. This analysis shows that destabilizing effects of local curvatures are practically negligible, and that curvature effects can be used to stabilize, sometimes significantly, an otherwise unstable fixture. The stiffness matrix formula is also used to show that stability analysis in general depends on the choice of contact models, which offers additional evidence for the importance of using realistic contact models.</p>
<p>The stiffness and deflection quality measures are defined for compliant fixtures and grasps, and are applied to optimal planning. Unlike existing quality measures that rely on heuristic rules or depend on reference frame choices, the stiffness and deflection quality measures are theoretically sound. Equally important is that these quality measures accurately characterize functional performances which are important to practical fixturing applications, such as fixture stiffness and workpiece deflections. The stiffness and deflection quality measures are applied to optimal fixture and grasp planning, resulting in maximum-stiffness and minimum-deflection fixtures and grasps. The qualitative properties of optimal fixtures are characterized with respect to each quality measure, and efficient techniques are developed for finding such optimal fixtures.</p>
<p>The final key issue is concerned with formal well-definedness conditions and practical development methods for rigid body kinematic metric functions, such as norms, inner products, and distance metrics. Based on an intrinsic definition of the configuration space of a rigid body, the notion of objectivity is proposed to formalize the natural requirement that metric measurements be indifferent to the observers who perform the measurements. This notion is then used to clarify the fundamental physical implications of left, right and bi-invariant functions on SE(3), and is further shown to be equivalent to the notion of frame-invariance. Based on these clarifications, several frame-invariant norms of rigid body velocities and wrenches, which have interesting physical interpretations, are defined.</p>https://resolver.caltech.edu/CaltechETD:etd-01302008-111854Computing with spiking neurons
https://resolver.caltech.edu/CaltechETD:etd-02042008-110206
Year: 1998
DOI: 10.7907/vf21-gw62
This thesis explores methods for computing with spikes. A spiking neuron model (SNM) is developed, which uses relatively few variables. A neuron's state is completely determined by the amount of neurotransmitter at its input synapses and the time since it last produced a spike. A spike is treated as a discrete event, which triggers the release of neurotransmitter at the neuron's output synapses. Neurotransmitter affects the voltage potentials of postsynaptic neurons.
The SNM is able to duplicate many of the properties of biological neurons, including: latency, efractory periods, and oscillatory spiking behavior, thus indicating that it is sufficiently complex for duplicating many of the computations performed by real neurons. Although the inspiration for the SNM comes from biology, the purpose of this research is to develop better computational devices.
Several single neuron building blocks are designed to perform useful functions, such as: a high gain response, a memory oscillator, a bounded threshold response, and an identity or inverse response. These single neuron building blocks are then used in larger networks to accomplish more complex tasks including: synchronizing input stimuli, recognizing spiking patterns, evaluating Boolean logic expressions, memorizing spike patterns, counting input spikes, multiplexing signals, comparing spike patterns, and recalling an associative memory.
When using the SNM, there are several possible methods for encoding information within a spike train. With synchronous spike patterns, each spike can encode a single bit. The strength of an input stimulus may be retained within the output phase of a spike or logarithmically encoded in the neurotransmitter released at a synapse. And when two sensory neurons receive the same input signal, the time duration of the stimulus can be linearly encoded within their phase differences, while the strength of the input signal is logarithmically encoded in their firing rates.
Learning may also be incorporated into an SNM network. A special feedforward network architecture is presented, in which each neuron has either an inhibitory or excitatory effect on all of the neurons to which it connects. A new learning rule is developed to train this network to respond to any combinations of input spike patterns.
https://resolver.caltech.edu/CaltechETD:etd-02042008-110206Modeling and Experiments for a Class of Robotic Endoscopes
https://resolver.caltech.edu/CaltechETD:etd-10112006-154843
Year: 1999
DOI: 10.7907/NG6V-TD44
<p>Current developments in minimally invasive medical practice motivated this study of self-propelled, robotic endoscopes for deep penetration into curved physiological lumens. The conceptual design of such devices is applicable to endoscopy within a variety of lumens in the human body, e.g., blood vessels, but the initial objective of this technology is to provide access to the interior of the entire small intestine without surgical incisions. The small intestine presents several challenges to endoscopic penetration: it is extremely compliant to applied loading, internally lubricated, easily injured, and contains many tight curves along its length of approximately eighteen feet.</p>
<p>This thesis reports the basic design and locomotion concepts for one class of endoscopic robots that are intended to provide safe and reliable traversal of the small intestine via worm-like actuation. Five generations of proof-of-concept prototype robots have been built to validate the fundamental concepts. Furthermore, these miniaturized robots have incorporated the following features: redundant actuation with computer control, tool-free modular assembly, and on-board videoimaging capability. The prototypes have been tested in rubber tubing, the small intestines of deceased pigs, and in the small intestines of live, anaesthetized pigs.</p>
<p>At the onset of this research, little regarding the elastic properties of small intestine existed in the biomechanics literature that would be applicable to the design of these mechanisms. However, accurate prediction of the small intestine's response to robotic loadings would dramatically improve the research and development process of these machines. Thus, an investigation of the elastic behavior of the small intestine commenced. Finite deformation, nonlinear, anisotropic, incompressible, viscoelastic behavior of the small intestine was studied. This soft tissue biomechanical analysis and experimentation (on living and dissected intestinal specimens) culminated with a numerical model that simulates intestinal response to the actions of a prototypical robotic component. Experiments on living specimens were performed to determine the levels of applied loadings and internal stresses that are likely to injure these fragile tissues, and the biomechanics computer modeling incorporates three distinct measures for injury potential.</p>https://resolver.caltech.edu/CaltechETD:etd-10112006-154843Theory and experiments in autonomous sensor-based motion planning with applications for flight planetary microrovers
https://resolver.caltech.edu/CaltechETD:etd-02202008-130626
Year: 1999
DOI: 10.7907/b1wv-hc78
With the success of Mars Pathfinder's Sojourner rover, a new era of planetary exploration has opened, with demand for highly capable mobile robots. These robots must be able to traverse long distances over rough, unknown terrain autonomously, under severe resource constraints. Much prior work in mobile robot path planning has been based on assumptions that are not truly applicable to navigation through planetary terrains. Based on the author's firsthand experience with the Mars Pathfinder mission, this work reviews issues which are critical for successful autonomous navigation of planetary rovers. No current methodology addresses all of these constraints. We next develop the sensor-based "Wedgebug" motion- planning algorithm. This algorithm is complete, correct, requires minimal memory for storage of its world model, and uses only on-board sensors, which are guided by the algorithm to efficiently sense only the data needed for motion planning, while avoiding unnecessary robot motion. The planner has the additional advantage of producing locally-optimal paths, and is suitable for robots with a field-of-view limited in both downrange and angular scope, for a variety of applications including planetary navigation. This work includes the proof of completeness and correctness of the Wedgebug algorithm, and in particular provides a corrected, detailed proof of a key result required for the proof of completeness of the Wedgebug algorithm (and for the TangentBug algorithm which inspired this approach). In addition, we extend this result to a broader class of environments. The implementation of a version of Wedgebug, called "RoverBug," on the Rocky7 Mars Rover prototype at the Jet Propulsion Laboratory (JPL) is described, and experimental results from operation in simulated martian terrain are presented.
https://resolver.caltech.edu/CaltechETD:etd-02202008-130626Control of Multiple Model Systems
https://resolver.caltech.edu/CaltechETD:etd-07312002-091923
Year: 2002
DOI: 10.7907/17Q7-Y019
This thesis considers the control of multiple model systems. These are systems for which only one model out of some finite set of models gives the system dynamics at any given time. In particular, the model that gives the system dynamics can change over time. This thesis covers some of the theoretical aspects of these systems, including controllability and stabilizability. As an application, ``overconstrained' mechanical systems are modeled as multiple model systems. Examples of such systems include distributed manipulation problems such as microelectromechanical systems and many wheeled vehicles such as the Sojourner vehicle of the Mars Pathfinder mission. Such systems are typified by having more Pfaffian constraints than degrees of freedom. Conventional classical motion planning and control theories do not directly apply to overconstrained systems. Control issues for two examples are specifically addressed. The first example is distributed manipulation. Distributed manipulation systems control an object's motion through contact with a high number of actuators. Stability results are shown for such systems and control schemes based on these results are implemented on a distributed manipulation test-bed. The second example is that of overconstrained vehicles, of which the Mars rover is an example. The nonlinear controllability test for multiple model systems is used to answer whether a kinematic model of the rover is or is not controllable.https://resolver.caltech.edu/CaltechETD:etd-07312002-091923Averaging and Control of Nonlinear Systems
https://resolver.caltech.edu/CaltechETD:etd-05282003-094253
Year: 2003
DOI: 10.7907/N7HH-PM67
<p>This dissertation investigates three principal areas regarding the dynamics and control of nonlinear systems: averaging theory, controllability of mechanical systems, and control of underactuated nonlinear systems. The most effective stabilizing controllers for underactuated nonlinear systems are time-periodic, which leads to the study of averaging theory for understanding the nonlinear effect generated by resonant oscillatory inputs.</p>
<p>The research on averaging theory generalizes averaging theory to arbitrary order by synthesizing series expansion methods for nonlinear time-varying vector fields and their flows with nonlinear Floquet theory. It is shown that classical averaging theory is the application of perturbation methods in conjunction with nonlinear Floquet theory. Many known properties and consequences of averaging theory are placed within a single framework.</p>
<p>The generalized averaging theory is merged with controllability analysis of underactuated nonlinear systems to derive exponentially stabilizing controllers. Although small-time local controllability (STLC) is easily demonstrated for driftless systems via the Lie algebra rank condition, STLC for systems with drift is more complicated. Furthermore, there exists a variety of techniques and canonical forms for determining STLC. This thesis exploits notions of geometric homogeneity to show that STLC results for a large class of mechanical systems with drift can be recovered by considering a class of nonlinear dynamical systems satisfying certain homogeneity conditions. These theorems generalize the controllability results for simple mechanical control systems found in Lewis and Murray [85]. Most nonlinear controllability results for classes of mechanical systems may be obtained using these methods.</p>
<p>The stabilizing controllers derived using the generalized averaging theory and STLC analysis can be used to stabilize both systems with and without drift. Furthermore, they result in a set of tunable gains and oscillatory parameters for modification and improvement of the feedback strategy. The procedure can not only derive known controllers from the literature, but can also be used to improve them. Examples demonstrate the diversity of controllers constructed using the generalized averaging theory.</p>
<p>This dissertation concludes with a chapter devoted to biomimetic and biomechanical locomotive control systems that have been stabilized using the generalized averaging theory and the controller construction procedure. The locomotive control systems roll, wriggle, swim, and walk, demonstrating the universal nature of the control strategy proposed.</p>https://resolver.caltech.edu/CaltechETD:etd-05282003-094253Symmetry, Reduction and Swimming in a Perfect Fluid
https://resolver.caltech.edu/CaltechETD:etd-06042003-181857
Year: 2003
DOI: 10.7907/CE65-XM80
This thesis presents a geometric picture of a deformable body in a perfect fluid and a way to approximate its dynamics and the motion, resulting from cyclic shape deformations, of the body and, interestingly, the fluid as well. Emphasis is placed on the group structure of the configuration space of the body fluid system and the resulting symmetry in their equations of motion. Symmetry is also used to reduce a series expansion for the flow of a time dependent vector field in order to obtain a novel expansion for the path-ordered exponential. This can be used to approximate holonomy, or geometric phase, in a principal bundle when its evolution is governed by a connection on the bundle and it is subject to periodic shape inputs. Simple models for swimming in and the stirring of a perfect fluid are proposed and examined.https://resolver.caltech.edu/CaltechETD:etd-06042003-181857Fluid Locomotion and Trajectory Planning for Shape-Changing Robots
https://resolver.caltech.edu/CaltechETD:etd-05292003-160843
Year: 2003
DOI: 10.7907/MFM1-0866
Motivated by considerations of shape changing propulsion of underwater robotic vehicles, I analyze the mechanics of deformable bodies operating in an ideal fluid. I give particular attention to fishlike robots which may be considered as one or more flexing or oscillating hydrofoils. I then describe methods of planning trajectories for a fishlike robot or any other sort of robot whose locomotion has a periodic or quasi-periodic nature.https://resolver.caltech.edu/CaltechETD:etd-05292003-160843Spike Train Characterization and Decoding for Neural Prosthetic Devices
https://resolver.caltech.edu/CaltechETD:etd-07232003-012018
Year: 2004
DOI: 10.7907/GK20-5W75
<p>Neural prosthetic device has the potential of benefiting millions of lock-in and spinal cord injury survivors. One branch of the ongoing research is to construct reach movement based prosthetic devices. This thesis proposes statistical methods based on applying the Haar wavelet packets to spike trains in order to answer some of the questions in this field.</p>
<p>Although spike train is the most frequently used data in the neural science community, its stochastic properties are not fully understood or characterized. This thesis suggests a formal spike train characterization method using the Haar wavelet packet. Because of the multi-scale property of the wavelet packet, Poisson characteristics at different scales can be assessed. Moreover, Poisson Scale-gram is proposed to help visualize the characteristics of the spike train at different scales.</p>
<p>Because some neurons display non-Poisson characteristics, it is necessary to extract the relevant features from spike trains in the context of decoding. The thesis suggests a feature extraction method that searches all the wavelet packet coefficients for the ones with the largest discriminability, quantified by mutual information. This technique returns the most informative feature(s) in the context of the Bayesian classifier. Decoding performance of this proposed method is compared against the one using mean firing rate only on both surrogate data and the actual data from PRR.</p>
<p>It is also crucial to decode cognitive states because they provide the extra control signals necessary for practical implementation of the prosthetic devices. This thesis proposes a simple finite state machine approach along with an interpreter that interprets the decoding results and to regulate when the transition should occur. It demonstrates that the finite state machine framework, when coupled with the interpreter, offers a simple autonomous control scheme for the neuron prosthetic system envisioned.</p>
<p>While the neural prosthetic system is in its infancy, many theoretical and experimental works lay the foundation for a bright future in this field. This thesis answers the spike train characterization and decoding questions in a theoretical manner while offering several novel techniques that bring new ideas and insights into the research field.</p>https://resolver.caltech.edu/CaltechETD:etd-07232003-012018A Control System for Positioning Recording Electrodes to Isolate Neurons in Extracellular Recordings
https://resolver.caltech.edu/CaltechETD:etd-06042006-160620
Year: 2006
DOI: 10.7907/6HHC-5456
<p>This thesis presents an algorithm that autonomously positions recording electrodes inside cortical tissue so as to isolate and then maintain optimal extracellular signal recording quality without human intervention. The algorithm is used to improve the quality and efficiency of acute (daily insertion) recordings that are needed for basic research in neurophysiology. It also offers the potential to increase the longevity and quality of chronic (long-term implant) recordings by controlling an emerging class of chronic arrays in which the electrodes can be continually repositioned after implantation.</p>
<p>The challenges encountered in attempting to isolate neurons are studied. A solution is proposed in which a finite state machine oversees a number of signal processing steps, computes various metrics of the recording quality and issues commands to move the electrode close to neurons without causing them damage. A number of metrics of the quality of neuron isolation are compared.</p>
<p>The algorithm has been used to control a number of commercial microdrive systems, including a single-electrode FHC microdrive and multielectrode microdrives from Thomas Recording and NAN, as well as a novel miniature microdrive. The autonomous positioning software is used by several neuroscientists to perform basic neurophysiology research. Analysis of the system's performance in isolating neurons is included.</p>https://resolver.caltech.edu/CaltechETD:etd-06042006-160620Multi-robot Systems: Modeling Swarm Dynamics and Designing Inspection Planning Algorithms
https://resolver.caltech.edu/CaltechETD:etd-05192006-063455
Year: 2006
DOI: 10.7907/G1T2-FB74
<p>For a variety of applications, the capability of simultaneous sensing and action in multiple locations that is inherent to multi-robot approaches offers potential advantages over single robot systems in robustness, efficiency, and application feasibility.</p>
<p>At the fully distributed and reactive end of the multi-robot system spectrum, I present mathematical modeling methodologies developed to predict and optimize a self-organized robotic swarm’s performance for several tasks. These models allow us to better understand the relationship between agent and group behavior by capturing the dynamics of these highly stochastic, nonlinear, asynchronous systems at various levels of abstraction, in some cases even achieving mathematical tractability. The models deliver qualitatively and quantitatively correct predictions several orders of magnitude more quickly than an embodied simulator can. Swarm modeling lays the foundation for more generalized SI system design methodology by saving time, enabling generalization to different robotic platforms, and estimating optimal design and control parameters.</p>
<p>In considering more complex target tasks and behaviors, efficiency and completeness of execution may be of concern, and a swarm approach may not be appropriate. In such cases a more deliberative approach may be warranted. In that context, I introduce the multi-robot boundary coverage problem, in which a group of robots is required to completely inspect the boundary of all two-dimensional objects in a specified environment. To make such a guarantee, I present a centralized planning approach that constructs a two-component abstraction of the problem: a graph representing the particular instance of the inspection task and a graph problem whose solution represents a complete plan for inspection. Using the building blocks of this approach, related inspection tasks that require the robotic system to adapt to a changes in team size and task assignment are also explored. The application of these planning methods to the case of long-term deployment for surveillance applications that require repetitive coverage is also discussed.</p>
<p>The recurring theme of this thesis is that we must look beyond implementation and validation of a particular system and ask how its design can contribute to the development of a more general design methodology.</p>https://resolver.caltech.edu/CaltechETD:etd-05192006-063455Tools and Algorithms for Mobile Robot Navigation with Uncertain Localization
https://resolver.caltech.edu/CaltechETD:etd-06012006-150109
Year: 2006
DOI: 10.7907/R6YB-NQ21
The ability for a mobile robot to localize itself is a basic requirement for reliable long range autonomous navigation. This thesis introduces new tools and algorithms to aid in robot localization and navigation. I introduce a new range scan matching method that incorporates realistic sensor noise models. This method can be thought of as an improved form of odometry. Results show an order of magnitude of improvement over typical mobile robot odometry. In addition, I have created a new sensor-based planning algorithm where the robot follows the locally optimal path to the goal without exception, regardless of whether or not the path moves towards or temporarily away from the goal. The cost of a path is defined as the path length. This new algorithm, which I call "Optim-Bug," is complete and correct. Finally, I developed a new on-line motion planning procedure that determines a path to a goal that optimally allows the robot to localize itself at the goal. This algorithm is called "Uncertain Bug." In particular, the covariance of the robot's pose estimate at the goal is minimized. This characteristic increases the likelihood that the robot will actually be able to reach the desired goal, even when uncertainty corrupts its localization during movement along the path. The robot's path is chosen so that it can use known features in the environment to improve its localization. This thesis is a first step towards bringing the tools of mobile robot localization and mapping together with ideas from sensor-based motion planning.https://resolver.caltech.edu/CaltechETD:etd-06012006-150109A Probabilistic Framework for Real-Time Mapping on an Unmanned Ground Vehicle
https://resolver.caltech.edu/CaltechThesis:01112018-111923389
Year: 2006
DOI: 10.7907/SXNY-XG55
In the course of preparing for the 2005 DARPA Grand Challenge, an off-road race for autonomous vehicles, a group of undergraduates from Caltech developed a set of deterministic algorithms for performing sensor fusion on maps generated by different range sensors on a mobile robot. That framework had serious limitations, however, including "disappearing" obstacles and lack of confidence data associated with features in the maps. In this thesis, we present a probabilistic framework that attempts to solve some of these problems by using error models of two typical types of range sensors, as well as by making use of Kalman filtering techniques from control theory to fuse the resulting measurements into an accurate digital elevation map. Our results indicate that this probabilistic framework has several advantages over the determinisic framework used by Team Caltech in the 2005 Grand Challenge.https://resolver.caltech.edu/CaltechThesis:01112018-111923389Algorithms for Mobile Robot Localization and Mapping, Incorporating Detailed Noise Modeling and Multi-scale Feature Extraction
https://resolver.caltech.edu/CaltechETD:etd-05262006-130209
Year: 2006
DOI: 10.7907/FN3J-M568
<p>Mobile robot localization and mapping in unknown environments is a fundamental requirement for effective autonomous navigation. Three different approaches to localization and mapping are presented. Each is based on data collected from a robot using a dense range scanner to generate a planar representation of the surrounding environment. This externally sensed range data is then overlayed and correlated to estimate the robot's position and build a map.</p>
<p>The three approaches differ in the choice of representation of the range data, but all achieve improvements over prior work using detailed sensor modeling and rigorous bookkeeping of the modeled uncertainty in the estimation processes. In the first approach, the raw range data points collected from two different positions are individually weighted and aligned to estimate the relative robot displacement. In the second approach, line segment features are extracted from the raw point data and are used as the basis for efficient and robust global map construction and localization. In the third approach, a new multi-scale data representation is introduced. New methods of localization and mapping are developed, taking advantage of this multi-scale representation to achieve significant improvements in computational complexity. A central focus of all three approaches is the determination of accurate and robust solutions to the data association problem, which is critical to the accuracy of any sensor-based localization and mapping method.</p>
<p>Experiments using data collected from a Sick LMS-200 laser scanner illustrate the effectiveness of the algorithms and improvements over prior work. All methods are capable of being run in real time on a mobile robot, and can be used to support fully autonomous navigation applications.</p>https://resolver.caltech.edu/CaltechETD:etd-05262006-130209Intelligent Information-Gathering: Using Control for Sensing and Decision-Making
https://resolver.caltech.edu/CaltechETD:etd-05312007-024822
Year: 2007
DOI: 10.7907/V5S4-4197
Information is everywhere and evolving, which necessitates both deliberate and efficient processing to acquire a good understanding of the dynamic situation, environment, or system of interest. Intelligent agents such as autonomous mobile sensors can control the way they gather information and thereby take advantage of feedback to improve the quality of that information. This approach reflects a shift from traditional "sensing for control" notions to "control for sensing" methods for addressing information-based objectives. This thesis presents several algorithms for distributed sensing tasks in the context of a team of mobile sensing agents. Applications of these types of mobile sensor networks include target tracking, dynamic environment monitoring, and distributed classification. These methods point beyond the use of sensory data for control and toward a framework for using control to improve information-based decisions made by intelligent agents. The sequential decision-theoretic framework presented herein has relevant applications in engineered systems such as search and rescue using a robotic team, as well as potential connections to natural systems including search strategies in the human vision system.https://resolver.caltech.edu/CaltechETD:etd-05312007-024822Exploration into the Feasibility of Underwater Synthetic Jet Propulsion
https://resolver.caltech.edu/CaltechETD:etd-09252006-134742
Year: 2007
DOI: 10.7907/72SZ-T823
<p>This thesis explores the feasibility of using synthetic jet actuators for the propulsion of small underwater vehicles. This work was inspired by the widespread use of pusatile jet propulsion by sea creatures such as squid, salp, and jellyfish. The jets created by these animals utilize vortex rings for thrust production. A method for creating similar vortex ring-based jets is the use of synthetic, or zero net mass flux, jets. These jets, which form a jet structure through the alternating sucking and blowing of fluid through a single orifice, have previously been investigated for the utility in air flow control.</p>
<p>The design, construction, and testing of aquatic synthetic jet prototypes is presented. Force measurement and flow visualization experiments are performed on these jets to gain an understanding of the forces and flow structures produced. The flow visualizations confirm the outflow vortex ring observations reported previously in the literature and present the first images of vortex ring formation inside the synthetic jet chamber. A new phenomenon, that of self-induced coflow upstream of the jet orifice, is discussed. The force measurements present confirmation that a net thrust is produced by the jets and give insight to the relationship between jet forcing parameters (such as frequency) and the resulting thrust. An automated genetic algorithmic approach to optimizing the thrust for a given jet geometry is also presented and tested.</p>
<p>Using the results of these experiments I propose a model for synthetic jet thrust. This model asserts that there are three force producing components to the flow: orifice inflow, orifice outflow, and a self-induced coflow. The contribution of each of these components is derived and compared with experimental results.</p>
<p>Included at the end of this thesis is a preliminary study into possible vehicle architecture for the utilization of synthetic jet thrusters.</p>https://resolver.caltech.edu/CaltechETD:etd-09252006-134742Robotics Training Algorithms for Optimizing Motor Learning in Spinal Cord Injured Subjects
https://resolver.caltech.edu/CaltechETD:etd-08142006-165844
Year: 2007
DOI: 10.7907/EH12-WD80
<p>The circuitries within the spinal cord are remarkably robust and plastic. Even in the absence of supraspinal control, such circuitries are capable of generating functional movements and changing their level of excitability based on a specific combination of properceptive inputs going into the spinal cord. This has led to an increase in locomotor training, such as Body Weight Support Treadmill training (BWST) for spinal cord injured (SCI) patients. However, today, little is known about the underlying physiological mechanisms responsible for the locomotor recovery achieved with this type of rehabilitative training, and the optimal rehabilitative strategy is still unknown.</p>
<p>This thesis describes a mouse model to study the effect of rehabilitative training on SCI. Using this model, the effects of locomotor recovery on adult spinal mice following complete spinal cord transaction is examined. Results that indicate adult spinal mice can be robotically trained to step, and when combined with the administration of quipazine (a broad serotonin agonist), there is an interaction and retention effect. Results also demonstrate that the training paradigm can be optimized in using “Assisted-as-Needed” (AAN) training. To find the optimal AAN training parameters, a learning model is developed to test the effect of various parameters of the AAN training algorithm. Simulation results from our model show that learning is training-dependent. In addition, the model predicts that improved motor learning can improve post-SCI by making the AAN training more adaptable.</p>
<p>The primary contributions of this thesis are twofold, in biology and engineering. We develop a mouse model using novel robotic devices and controls that can be used to study SCI and other locomotor disorders in the future by taking advantage of the many different strains of transgenic mice that are commercially available. We also further confirm that sensory integration responsible for motor control is distributed throughout the hierarchy of the neuromuscular system and can be achieved within the isolated spinal cord. Lastly, by developing a learning model, we can start looking into how variability plays a role in motor learning, the understanding of which will have profound implications in neurophysiology, machine learning and adaptive optimal controls research.</p>https://resolver.caltech.edu/CaltechETD:etd-08142006-165844Automated Visual Tracking for Behavioral Analysis of Biological Model Organisms
https://resolver.caltech.edu/CaltechETD:etd-05272008-161801
Year: 2008
DOI: 10.7907/TSQ7-SN68
<p>Capturing the detailed motion and behavior of biological organisms plays an important role in a wide variety of research disciplines. Many studies in biomechanics, neuroethology, and developmental biology rely on analysis of video sequences to understand the underlying behavior. However, the efficient and rapid quantification of these complex behavioral traits imposes a major bottleneck on the elucidation of many interesting scientific questions. The goal of this thesis is to develop a suite of model-based visual tracking algorithms that will apply across a variety of model organisms used in biology. These automated tracking algorithms operate in a high-throughput, high-resolution manner needed for a productive synthesis with modern genetic approaches. To this end, I demonstrate automated estimation of the detailed body posture of nematodes, zebrafish, and fruit flies from calibrated video.</p>
<p>The current algorithm utilizes a generative geometric model to capture the organism's shape and appearance. To accurately predict the organism's motion between video frames, I incorporate a motion model that matches tracked motion patterns to patterns in a training set. This technique is invariant with respect to the organism's velocity and can easily incorporate training data from completely different motion patterns. The prediction of the motion model is refined using measurements from the image. In addition to high-contrast feature points, I introduce a region, segmentation model based on level sets that are formally integrated into the observation framework of an Iterated Kalman Filter (IKF). The prior knowledge provided by the geometric and motion models improves tracking accuracy in the presence of partial occlusions and misleading visual cues.</p>
<p>The method is used to track the position and shape of multiple nematodes during mating behavior, zebrafish of different ages during escape response, and fruit flies during take off maneuvers. These applications demonstrate the modular design of this model-based visual tracking system, where the user can specify which components are appropriate to a given experiment. In contrast to other approaches, which are customized to a particular organism or experimental setup, my approach provides a foundation that requires little re-engineering whenever the experimental parameters are changed.</p>https://resolver.caltech.edu/CaltechETD:etd-05272008-161801Adaptive Feature Selection in Pattern Recognition and Ultra-Wideband Radar Signal Analysis
https://resolver.caltech.edu/CaltechETD:etd-05302008-134607
Year: 2008
DOI: 10.7907/7NR6-AR24
<p>Feature selection from measured data aims to extract informative features to reveal the statistic or stochastic mechanism underlying the complicated or high dimensional original data. In this thesis, the feature selection problem is probed under two situations, one is pattern recognition and the other is ultra-wideband radar signal analysis.</p>
<p>Classical pattern recognition methods select features by their ability to separate the multiple classes with certain gauge measure. The deficiency in this general strategy is its lack of adaptation in specific situations. This deficiency may be overcome by viewing the selected features as a function of not only the training samples but also the unlabeled test data. From this perspective, this thesis proposes an adaptive sequential feature selection algorithm which utilizes an information-theoretic measure to reduce the classification task complexity sequentially, and finally outputs the probabilistic classification result and its variation level. To verify the potential advantage of this algorithm, this thesis applies it to one important problem of neural prosthesis, which concerns decoding a finite number of classes, intended reach directions, from recordings of neural activities in the Parietal Reach Region of one rhesus monkey. Experimental results show that the classification scheme of combining the adaptive sequential feature selection algorithm and the information fusion method outperforms some classical pattern recognition rules, such as the nearest neighbor rule and support vector machine, in decoding performance.</p>
<p>The second scenario in this thesis targets developing a human presence and motion pattern detector through ultra-wideband radar signal analysis. To augment the detection robustness, both static and dynamic features should be utilized. The static features reflect the information of target geometry and its variability, while the dynamic features extract the temporal structure among radar scans. The problem of static feature selection is explored in this thesis, which utilizes the Procrustes shape analysis to generate the representative template for the target images, and makes statistical inference in the tangent space through the Hotelling one sample test. After that, the waveform shape variation structure is decomposed in the tangent space through the principal component analysis. The selected principal components not only accentuate the prominent dynamics of the target motion, but also generate another informative classification feature.</p>
https://resolver.caltech.edu/CaltechETD:etd-05302008-134607Robotic Training for Motor Rehabilitation after Complete Spinal Cord Injury
https://resolver.caltech.edu/CaltechETD:etd-09202007-135027
Year: 2008
DOI: 10.7907/T01R-P904
<p>The spinal cord circuits have a great degree of automaticity and plasticity. They are able to generate complex locomotor patterns such as stepping and scratching even without input from supraspinal nervous systems. When provided with ensembles of afferent sensory information input associated with a specific motor task, e.g., stepping, the spinal cord can "learn" to perform that task even if it is isolated from the supraspinal nervous systems.</p>
<p>The plasticity of the spinal cord led researchers to study the use of physical locomotor training, e.g., treadmill step training with body weight support, to rehabilitate locomotor function after spinal cord injury (SCI). With intensive training, the spinal-cord-injured subject can recover some level of stepping ability. Explorations were made in this thesis to find an optimal training paradigm. Novel assist-as-needed paradigms were developed to allow variability during training since it is an intrinsic feature of normal stepping. Comparative experiments were conducted against fixed-trajectory training. Results demonstrated that variability is an important factor to induce more improvement in step training.</p>
<p>Standing is another important function in one's daily life, though it received less research attention than stepping. A prototype stand platform with 6 degrees of freedom was developed as an experimental tool for stand and postural study. Analogous to step training, we tested the effect of daily training on extensor responses in the hind limbs of complete spinal rats. The results showed no significant effect of the training. This led to the conclusion that without tonic input, the spinal cord has very limited ability to generate enough extensor muscle tone and to respond to postural disturbance. Further studies in standing rehabilitation should combine other methods to provide tonic inputs to the spinal cord.</p>https://resolver.caltech.edu/CaltechETD:etd-09202007-135027Target Tracking Using Clustered Measurements, with Applications to Autonomous Brain-Machine Interfaces
https://resolver.caltech.edu/CaltechETD:etd-05292008-105504
Year: 2008
DOI: 10.7907/6Y3F-8M87
<p>This thesis presents new methods for classifying and tracking the signals of targets that produce clusters of observations, measured in successive recording intervals or scans. This multitarget tracking problem arises, for instance, in extracellular neural recordings, in which an electrode is inserted into the brain to detect the spikes of individual neurons. Since multiple active neurons may lie near the electrode, each detected spike must be assigned to the neuron that produced it, a task known as spike sorting. In the scenario considered in this thesis, the electrode signal is sampled over many brief recording intervals. In each recording interval, all spikes must first be clustered according to their generating neurons, and then each cluster must be associated to clusters from previous recording intervals, thus tracking the signals of putative neuron "targets."</p>
<p>This thesis introduces a novel multitarget tracking solution for the above problem, called multiple hypothesis tracking for clusters (MHTC). The MHTC algorithm has two main parts: a Bayesian clustering algorithm for associating observations to clusters in each interval and a probabilistic supervisory system that manages association hypotheses across intervals. The clustering procedure provides significantly more consistent results than previously available methods, enabling more accurate tracking of targets over time. Such consistency is promoted by a maximum a posteriori (MAP) approach to optimizing a Gaussian mixture model via expectation-maximization (EM), in which information from the preceding intervals serves as a prior for the current interval while still allowing the number and locations of targets to change. MHTC's hypothesis management system, like that of traditional multiple hypothesis tracking (MHT) algorithms, propagates various possibilities for how to assign measurements to existing targets and uses a delayed decision-making logic to resolve data association ambiguities. It also, however, maintains several options, termed model hypotheses, for how to cluster the observations of each interval. This combination of clustering and tracking in a single solution enables MHTC to robustly maintain the identities of cluster-producing targets in challenging recording scenarios.</p>
<p>In addition to these classification and tracking techniques, this thesis presents advances in a miniature robotic electrode microdrive capable of extracellular recordings lasting for days at a time. As a whole, these contributions can play an important role in enabling an autonomous neural interface, which, by frequent automatic repositioning of its recording electrodes, can optimize the recording quality of extracellular signals associated with individual neurons and maintain high quality recordings for long periods of time. Such autonomous movable electrodes may eventually overcome key barriers to engineering a practical neuroprosthetic device and, in the near term, can significantly improve state-of-the-art neuroscience experimental procedures.</p>
https://resolver.caltech.edu/CaltechETD:etd-05292008-105504Inference in Hybrid Systems with Applications in Neural Prosthetics
https://resolver.caltech.edu/CaltechETD:etd-12312008-184713
Year: 2009
DOI: 10.7907/REB5-BB43
<p>This thesis develops new hybrid system models and associated inference algorithms to create a ``supervisory decoder' for cortical neural prosthetic devices that aim to help the severely handicapped. These devices are a brain-machine interface, consisting of surgically implanted electrode arrays and associated computer decoding algorithms, that enable a human to control external electromechanical devices, such as artificial limbs, by thought alone.</p>
<p>Hybrid systems are characterized by discrete switching between sets of continuous dynamical activity. New hybrid models, which are flexible enough to model neurological activity, are created that incorporate both duration and dynamical state based switching paradigms. Combining generalized linear models with nonstationary and semi-Markov chains gives rise to three new hybrid systems: generalized linear hidden Markov models (GLHMM), hidden semi-Markov models (HSMM) with generalized linear model dynamics, and hidden regressor dependent Markov models (HRDMM). Bayesian inference methods, including variational Bayes and Gibbs sampling, are derived for the identification of existing and developed hybrid models. The developed inference algorithms provide advances over the current hybrid system identification literature by providing a principled way to incorporate prior knowledge and select between alternative model classes and orders, including the number of discrete system states.</p>
<p>Future neuroprostheses that seek to provide a facile interface for the paralyzed patient will require a supervisory decoder that classifies, in real time, the discrete cognitive, behavioral, or planning state of the brain. The developed hybrid models and inference algorithms provide a framework for supervisory decoding, where first, a hybrid-state neurological activity model is identified from data, and then used to estimate the discrete state in real time. The electrical activity of multiple neurons from a cortical area in the brain associated with motor planning (the parietal reach region), and multiple signal types, including both spike arrival times and local field potentials, are fused to give more accurate results. The model structure, including the number of discrete cognitive states, can also be estimated from the data, resulting in significantly improved decoding performance compared to existing methods.</p>
<p>Additional demonstrated applications include the automated segmentation of honey bee motion into discrete primitives, and generating mechanical system models for a pick-and-place machine.</p>
https://resolver.caltech.edu/CaltechETD:etd-12312008-184713Neuro-Evolution Using Recombinational Algorithms and Embryogenesis for Robotic Control
https://resolver.caltech.edu/CaltechTHESIS:06092010-140839602
Year: 2010
DOI: 10.7907/YNED-VN66
Control tasks involving dramatic nonlinearities, such as decision making, can be challenging for classical design methods. However, autonomous, stochastic design methods such as evolutionary computation have proved effective. In particular, genetic algorithms that create designs via the application of recombinational rules are robust and highly scalable. Neuro-Evolution Using Recombinational Algorithms and Embryogenesis (NEURAE) is a genetic algorithm that creates C++ programs that in turn create neural networks which can function as logic gates. The neural networks created are scalable and robust enough to feature redundancies that allow the network to function despite internal failures. An analysis of NEURAE evinces how biologically inspired phenomena apply to simulated evolution. This allows for an optimization of NEURAE that enables it to create controllers for a simulated swarm of Khepera-inspired robots.https://resolver.caltech.edu/CaltechTHESIS:06092010-140839602Real-Time Applications of 3D Object Detection and Tracking
https://resolver.caltech.edu/CaltechTHESIS:01152010-143831008
Year: 2010
DOI: 10.7907/4N1K-GK74
Robot perception is a fundamental aspect of any autonomous system. It gives the robot the capacity to make sense of vast amounts of data and gain an understanding of the world around it. An active problem in the area of robot perception is real-time detection and pose estimation of 3D objects. This thesis presents an approach to 3D object detection and tracking utilizing a stereo-camera sensor. Geometric object models are learned in short order time via a training phase and real-time detection and tracking is made possible by performing sparse stereo calculations on the chosen features within an adaptive region of interest of the camera image. The experimental results obtained by using this method will show the effectiveness of the approach as compared against ground truth measures in real-time. Using that framework as a basis, extensions to two other problems in robot sensing are then considered: (1) sensor-planning for model identification, and (2) sensor-planning for object-search. In the former, a novel algorithm for determining the next-best-view for a mobile sensor to identify an unknown 3D object from among a database of known models is presented and tested across two experiments involving real robotic systems. An information theoretic approach is taken to quantify the utility of each potential sensing action and the validity of the algorithm is discussed. In the latter area, a novel approach is presented that allows an autonomous mobile robot to search for a 3D object using an onboard stereo camera sensor mounted on a pan-tilt head. Search efficiency is realized by the combination of a coarse-scale global search coupled with a fine-scale local search, guided by a grid-based probability map. Obstacle avoidance during the search is naturally integrated into the method with additional experimental results on a mobile robot presented to illustrate and validate the proposed search strategy. All presented experiments were carried out in real-time processing with modest computation done by a single laptop computer.https://resolver.caltech.edu/CaltechTHESIS:01152010-143831008Robot Motion Planning in Dynamic, Cluttered, and Uncertain Environments: the Partially Closed-Loop Receding Horizon Control Approach
https://resolver.caltech.edu/CaltechTHESIS:02042010-152638957
Year: 2010
DOI: 10.7907/SD3N-JR18
This thesis is concerned with robot motion planning in dynamic, cluttered, and uncertain environments. Successful and efficient robot operation in such environments requires reasoning about the future system evolution and the uncertainty associated with obstacles and moving agents in the environment. Current motion planning strategies ignore future information and are limited by the resulting growth of uncertainty as the system is evolved. This thesis presents an approach that accounts for future information gathering (and the quality of that information) in the planning process. The Partially Closed-Loop Receding Horizon Control approach, introduced in this thesis, is based on Dynamic Programming with imperfect state information. Probabilistic collision constraints, due to the need for obstacle avoidance between the robot and obstacles with uncertain locations and geometries, are developed and imposed. By accounting for the anticipated future information, the uncertainty associated with the system evolution is managed, allowing for greater numbers of moving agents and more complex agent behaviors to be handled. Simulation results demonstrate the benefit of the proposed approach over existing approaches in static and dynamic environments. Complex agent behaviors, including multimodal and interactive agent-robot models, are considered.https://resolver.caltech.edu/CaltechTHESIS:02042010-152638957Axel Rover Tethered Dynamics and Motion Planning on Extreme Planetary Terrain
https://resolver.caltech.edu/CaltechTHESIS:08312011-003358925
Year: 2012
DOI: 10.7907/MPHD-PC75
<p>Some of the most appealing science targets for future exploration missions in our solar system lie in terrains that are inaccessible to state-of-the-art robotic rovers such as NASA's Opportunity, thereby precluding in situ analysis of these rich opportunities. Examples of potential high-yield science areas on Mars include young gullies on sloped terrains, exposed layers of bedrock in the Victoria Crater, sources of methane gas near Martian volcanic ranges, and stepped delta formations in heavily cratered regions. In addition, a recently discovered cryovolcano on Titan and frozen water near the south pole of our own Moon could provide a wealth of knowledge to any robotic explorer capable of accessing these regions.</p>
<p>To address the challenge of extreme terrain exploration, this dissertation presents the Axel rover, a two-wheeled tethered robot capable of rappelling down steep slopes and traversing rocky terrain. Axel is part of a family of reconfigurable rovers, which, when docked, form a four-wheeled vehicle nicknamed DuAxel. DuAxel provides untethered mobility to regions of extreme terrain and serves as an anchor support for a single Axel when it undocks and rappels into low-ground.</p>
<p>Axel's performance on extreme terrain is primarily governed by three key system components: wheel design, tether control, and intelligent planning around obstacles. Investigations in wheel design and optimizing for extreme terrain resulted in the development of grouser wheels. Experiments demonstrated that these grouser wheels were very effective at surmounting obstacles, climbing rocks up to 90% of the wheel diameter. Terramechanics models supported by experiments showed that these wheels would not sink excessively or become trapped in deformable terrain.</p>
<p>Predicting tether forces in different configurations is also essential to the rover's mobility. Providing power, communication, and mobility forces, the tether is Axel's lifeline while it rappels steep slopes, and a cut, abraded, or ruptured tether would result in an untimely end to the rover's mission. Understanding tether forces are therefore paramount, and this thesis both models and measures tension forces to predict and avoid high-stress scenarios.</p>
<p>Finally, incorporating autonomy into Axel is a unique challenge due to the complications that arise during tether management. Without intelligent planning, rappelling systems can easily become entangled around obstacles and suffer catastrophic failures. This motivates the development of a novel tethered planning algorithm, presented in this thesis, which is unique for rappelling systems.</p>
<p>Recent field experiments in natural extreme terrains on Earth demonstrate the Axel rover's potential as a candidate for future space operations. Both DuAxel and its rappelling counterpart are rigorously tested on a 20 meter escarpment and in the Arizona desert. Through analysis and experiments, this thesis provides the framework for a new generation of robotic explorers capable of accessing extreme planetary regions and potentially providing clues for life beyond Earth.</p>https://resolver.caltech.edu/CaltechTHESIS:08312011-003358925Estimation and Inference for Grasping and Manipulation Tasks Using Vision and Kinesthetic Sensors
https://resolver.caltech.edu/CaltechTHESIS:04052013-105520483
Year: 2013
DOI: 10.7907/PZB6-QJ39
<p>This thesis presents a novel framework for state estimation in the context of robotic grasping and manipulation. The overall estimation approach is based on fusing various visual cues for manipulator tracking, namely appearance and feature-based, shape-based, and silhouette-based visual cues. Similarly, a framework is developed to fuse the above visual cues, but also kinesthetic cues such as force-torque and tactile measurements, for in-hand object pose estimation. The cues are extracted from multiple sensor modalities and are fused in a variety of Kalman filters.</p>
<p>A hybrid estimator is developed to estimate both a continuous state (robot and object states) and discrete states, called contact modes, which specify how each finger contacts a particular object surface. A static multiple model estimator is used to compute and maintain this mode probability. The thesis also develops an estimation framework for estimating model parameters associated with object grasping. Dual and joint state-parameter estimation is explored for parameter estimation of a grasped object's mass and center of mass. Experimental results demonstrate simultaneous object localization and center of mass estimation.</p>
<p>Dual-arm estimation is developed for two arm robotic manipulation tasks. Two types of filters are explored; the first is an augmented filter that contains both arms in the state vector while the second runs two filters in parallel, one for each arm. These two frameworks and their performance is compared in a dual-arm task of removing a wheel from a hub.</p>
<p>This thesis also presents a new method for action selection involving touch. This next best touch method selects an available action for interacting with an object that will gain the most information. The algorithm employs information theory to compute an information gain metric that is based on a probabilistic belief suitable for the task. An estimation framework is used to maintain this belief over time. Kinesthetic measurements such as contact and tactile measurements are used to update the state belief after every interactive action. Simulation and experimental results are demonstrated using next best touch for object localization, specifically a door handle on a door.
The next best touch theory is extended for model parameter determination. Since many objects within a particular object category share the same rough shape, principle component analysis may be used to parametrize the object mesh models. These parameters can be estimated using the action selection technique that selects the touching action which best both localizes and estimates these parameters. Simulation results are then presented involving localizing and determining a parameter of a screwdriver.</p>
<p>Lastly, the next best touch theory is further extended to model classes. Instead of estimating parameters, object class determination is incorporated into the information gain metric calculation. The best touching action is selected in order to best discern between the possible model classes. Simulation results are presented to validate the theory.</p>https://resolver.caltech.edu/CaltechTHESIS:04052013-105520483Robot Navigation in Dense Crowds: Statistical Models and Experimental Studies of Human Robot Cooperation
https://resolver.caltech.edu/CaltechTHESIS:05182013-191132413
Year: 2013
DOI: 10.7907/BHGM-0C65
<p>This thesis explores the problem of mobile robot navigation in dense human crowds. We begin by considering a fundamental impediment to classical motion planning algorithms called the freezing robot problem: once the environment surpasses a certain level of complexity, the planner decides that all forward paths are unsafe, and the robot freezes in place (or performs unnecessary maneuvers) to avoid collisions. Since a feasible path typically exists, this behavior is suboptimal. Existing approaches have focused on reducing predictive uncertainty by employing higher fidelity individual dynamics models or heuristically limiting the individual predictive covariance to prevent overcautious navigation. We demonstrate that both the individual prediction and the individual predictive uncertainty have little to do with this undesirable navigation behavior. Additionally, we provide evidence that dynamic agents are able to navigate in dense crowds by engaging in joint collision avoidance, cooperatively making room to create feasible trajectories. We accordingly develop interacting Gaussian processes, a prediction density that captures cooperative collision avoidance, and a "multiple goal" extension that models the goal driven nature of human decision making. Navigation naturally emerges as a statistic of this distribution.</p>
<p>Most importantly, we empirically validate our models in the Chandler dining hall at Caltech during peak hours, and in the process, carry out the first extensive quantitative study of robot navigation in dense human crowds (collecting data on 488 runs). The multiple goal interacting Gaussian processes algorithm performs comparably with human teleoperators in crowd densities nearing 1 person/m<sup>2</sup>, while a state of the art noncooperative planner exhibits unsafe behavior more than 3 times as often as the multiple goal extension, and twice as often as the basic interacting Gaussian process approach. Furthermore, a reactive planner based on the widely used dynamic window approach proves insufficient for crowd densities above 0.55 people/m<sup>2</sup>. We also show that our noncooperative planner or our reactive planner capture the salient characteristics of nearly any dynamic navigation algorithm. For inclusive validation purposes, we show that either our non-interacting planner or our reactive planner captures the salient characteristics of nearly any existing dynamic navigation algorithm. Based on these experimental results and theoretical observations, we conclude that a cooperation model is critical for safe and efficient robot navigation in dense human crowds.</p>
<p>Finally, we produce a large database of ground truth pedestrian crowd data. We make this ground truth database publicly available for further scientific study of crowd prediction models, learning from demonstration algorithms, and human robot interaction models in general.</p>
https://resolver.caltech.edu/CaltechTHESIS:05182013-191132413Spinal Cord Injury Therapy through Active Learning
https://resolver.caltech.edu/CaltechTHESIS:07252013-120308708
Year: 2014
DOI: 10.7907/X5M7-EC09
Therapy employing epidural electrostimulation holds great potential for improving therapy for patients with spinal cord injury (SCI) (Harkema et al., 2011). Further promising results from combined therapies using electrostimulation have also been recently obtained (e.g., van den Brand et al., 2012). The devices being developed to deliver the stimulation are highly flexible, capable of delivering any individual stimulus among a combinatorially large set of stimuli (Gad et al., 2013). While this extreme flexibility is very useful for ensuring that the device can deliver an appropriate stimulus, the challenge of choosing good stimuli is quite substantial, even for expert human experimenters. To develop a fully implantable, autonomous device which can provide useful therapy, it is necessary to design an algorithmic method for choosing the stimulus parameters. Such a method can be used in a clinical setting, by caregivers who are not experts in the neurostimulator's use, and to allow the system to adapt autonomously between visits to the clinic. To create such an algorithm, this dissertation pursues the general class of active learning algorithms that includes Gaussian Process Upper Confidence Bound (GP-UCB, Srinivas et al., 2010), developing the Gaussian Process Batch Upper Confidence Bound (GP-BUCB, Desautels et al., 2012) and Gaussian Process Adaptive Upper Confidence Bound (GP-AUCB) algorithms. This dissertation develops new theoretical bounds for the performance of these and similar algorithms, empirically assesses these algorithms against a number of competitors in simulation, and applies a variant of the GP-BUCB algorithm in closed-loop to control SCI therapy via epidural electrostimulation in four live rats. The algorithm was tasked with maximizing the amplitude of evoked potentials in the rats' left tibialis anterior muscle. These experiments show that the algorithm is capable of directing these experiments sensibly, finding effective stimuli in all four animals. Further, in direct competition with an expert human experimenter, the algorithm produced superior performance in terms of average reward and comparable or superior performance in terms of maximum reward. These results indicate that variants of GP-BUCB may be suitable for autonomously directing SCI therapy.https://resolver.caltech.edu/CaltechTHESIS:07252013-120308708Formal Methods for Control Synthesis in Partially Observed Environments: Application to Autonomous Robotic Manipulation
https://resolver.caltech.edu/CaltechTHESIS:05292014-063852576
Year: 2014
DOI: 10.7907/RQKC-N871
<p>Modern robots are increasingly expected to function in uncertain and dynamically challenging environments, often in proximity with humans. In addition, wide scale adoption of robots requires on-the-fly adaptability of software for diverse application. These requirements strongly suggest the need to adopt formal representations of high level goals and safety specifications, especially as temporal logic formulas. This approach allows for the use of formal verification techniques for controller synthesis that can give guarantees for safety and performance. Robots operating in unstructured environments also face limited sensing capability. Correctly inferring a robot's progress toward high level goal can be challenging.</p>
<p>This thesis develops new algorithms for synthesizing discrete controllers in partially known environments under specifications represented as linear temporal logic (LTL) formulas. It is inspired by recent developments in finite abstraction techniques for hybrid systems and motion planning problems. The robot and its environment is assumed to have a finite abstraction as a Partially Observable Markov Decision Process (POMDP), which is a powerful model class capable of representing a wide variety of problems. However, synthesizing controllers that satisfy LTL goals over POMDPs is a challenging problem which has received only limited attention.</p>
<p>This thesis proposes tractable, approximate algorithms for the control synthesis problem using Finite State Controllers (FSCs). The use of FSCs to control finite POMDPs allows for the closed system to be analyzed as finite global Markov chain. The thesis explicitly shows how transient and steady state behavior of the global Markov chains can be related to two different criteria with respect to satisfaction of LTL formulas. First, the maximization of the probability of LTL satisfaction is related to an optimization problem over a parametrization of the FSC. Analytic computation of gradients are derived which allows the use of first order optimization techniques. </p>
<p>The second criterion encourages rapid and frequent visits to a restricted set of states over infinite executions. It is formulated as a constrained optimization problem with a discounted long term reward objective by the novel utilization of a fundamental equation for Markov chains - the Poisson equation. A new constrained policy iteration technique is proposed to solve the resulting dynamic program, which also provides a way to escape local maxima.</p>
<p>The algorithms proposed in the thesis are applied to the task planning and execution challenges faced during the DARPA Autonomous Robotic Manipulation - Software challenge.</p>https://resolver.caltech.edu/CaltechTHESIS:05292014-063852576Efficient Methods for Stochastic Optimal Control
https://resolver.caltech.edu/CaltechTHESIS:05312014-011052261
Year: 2014
DOI: 10.7907/D40A-9E03
<p>The Hamilton Jacobi Bellman (HJB) equation is central to stochastic optimal control (SOC) theory, yielding the optimal solution to general problems specified by known dynamics and a specified cost functional. Given the assumption of quadratic cost on the control input, it is well known that the HJB reduces to a particular partial differential equation (PDE). While powerful, this reduction is not commonly used as the PDE is of second order, is nonlinear, and examples exist where the problem may not have a solution in a classical sense. Furthermore, each state of the system appears as another dimension of the PDE, giving rise to the curse of dimensionality. Since the number of degrees of freedom required to solve the optimal control problem grows exponentially with dimension, the problem becomes intractable for systems with all but modest dimension.</p>
<p>In the last decade researchers have found that under certain, fairly non-restrictive structural assumptions, the HJB may be transformed into a linear PDE, with an interesting analogue in the discretized domain of Markov Decision Processes (MDP). The work presented in this thesis uses the linearity of this particular form of the HJB PDE to push the computational boundaries of stochastic optimal control.</p>
<p>This is done by crafting together previously disjoint lines of research in computation. The first of these is the use of Sum of Squares (SOS) techniques for synthesis of control policies. A candidate polynomial with variable coefficients is proposed as the solution to the stochastic optimal control problem. An SOS relaxation is then taken to the partial differential constraints, leading to a hierarchy of semidefinite relaxations with improving sub-optimality gap. The resulting approximate solutions are shown to be guaranteed over- and under-approximations for the optimal value function. It is shown that these results extend to arbitrary parabolic and elliptic PDEs, yielding a novel method for Uncertainty Quantification (UQ) of systems governed by partial differential constraints. Domain decomposition techniques are also made available, allowing for such problems to be solved via parallelization and low-order polynomials.</p>
<p>The optimization-based SOS technique is then contrasted with the Separated Representation (SR) approach from the applied mathematics community. The technique allows for systems of equations to be solved through a low-rank decomposition that results in algorithms that scale linearly with dimensionality. Its application in stochastic optimal control allows for previously uncomputable problems to be solved quickly, scaling to such complex systems as the Quadcopter and VTOL aircraft. This technique may be combined with the SOS approach, yielding not only a numerical technique, but also an analytical one that allows for entirely new classes of systems to be studied and for stability properties to be guaranteed.</p>
<p>The analysis of the linear HJB is completed by the study of its implications in application. It is shown that the HJB and a popular technique in robotics, the use of navigation functions, sit on opposite ends of a spectrum of optimization problems, upon which tradeoffs may be made in problem complexity. Analytical solutions to the HJB in these settings are available in simplified domains, yielding guidance towards optimality for approximation schemes. Finally, the use of HJB equations in temporal multi-task planning problems is investigated. It is demonstrated that such problems are reducible to a sequence of SOC problems linked via boundary conditions. The linearity of the PDE allows us to pre-compute control policy primitives and then compose them, at essentially zero cost, to satisfy a complex temporal logic specification.</p> https://resolver.caltech.edu/CaltechTHESIS:05312014-011052261Convex Model Predictive Control for Vehicular Systems
https://resolver.caltech.edu/CaltechTHESIS:06052014-200112345
Year: 2014
DOI: 10.7907/PNN7-SC35
In this work, the author presents a method called Convex Model Predictive Control (CMPC) to control systems whose states are elements of the rotation matrices SO(n) for n = 2, 3. This is done without charts or any local linearization, and instead is performed by operating over the orbitope of rotation matrices. This results in a novel model predictive control (MPC) scheme without the drawbacks associated with conventional linearization techniques such as slow computation time and local minima. Of particular emphasis is the application to aeronautical and vehicular systems, wherein the method removes many of the trigonometric terms associated with these systems’ state space equations. Furthermore, the method is shown to be compatible with many existing variants of MPC, including obstacle avoidance via Mixed Integer Linear Programming (MILP).https://resolver.caltech.edu/CaltechTHESIS:06052014-200112345Two and Three Finger Caging of Polygons and Polyhedra
https://resolver.caltech.edu/CaltechTHESIS:12062015-164238181
Year: 2016
DOI: 10.7907/Z93X84KR
<p>Multi-finger caging offers a rigorous and robust approach to robot grasping. This thesis provides several novel algorithms for caging polygons and polyhedra in two and three dimensions. Caging refers to a robotic grasp that does not necessarily immobilize an object, but prevents it from escaping to infinity. The first algorithm considers caging a polygon in two dimensions using two point fingers. The second algorithm extends the first to three dimensions. The third algorithm considers caging a convex polygon in two dimensions using three point fingers, and considers robustness of this cage to variations in the relative positions of the fingers.</p>
<p>This thesis describes an algorithm for finding all two-finger cage formations of planar polygonal objects based on a contact-space formulation. It shows that two-finger cages have several useful properties in contact space. First, the critical points of the cage representation in the hand’s configuration space appear as critical points of the inter-finger distance function in contact space. Second, these critical points can be graphically characterized directly on the object’s boundary. Third, contact space admits a natural rectangular decomposition such that all critical points lie on the rectangle boundaries, and the sublevel sets of contact space and free space are topologically equivalent. These properties lead to a caging graph that can be readily constructed in contact space. Starting from a desired immobilizing grasp of a polygonal object, the caging graph is searched for the minimal, intermediate, and maximal caging regions surrounding the immobilizing grasp. An example constructed from real-world data illustrates and validates the method.</p>
<p>A second algorithm is developed for finding caging formations of a 3D polyhedron for two point fingers using a lower dimensional contact-space formulation. Results from the two-dimensional algorithm are extended to three dimension. Critical points of the inter-finger distance function are shown to be identical to the critical points of the cage. A decomposition of contact space into 4D regions having useful properties is demonstrated. A geometric analysis of the critical points of the inter-finger distance function results in a catalog of grasps in which the cages change topology, leading to a simple test to classify critical points. With these properties established, the search algorithm from the two-dimensional case may be applied to the three-dimensional problem. An implemented example demonstrates the method.</p>
<p>This thesis also presents a study of cages of convex polygonal objects using three point fingers. It considers a three-parameter model of the relative position of the fingers, which gives complete generality for three point fingers in the plane. It analyzes robustness of caging grasps to variations in the relative position of the fingers without breaking the cage. Using a simple decomposition of free space around the polygon, we present an algorithm which gives all caging placements of the fingers and a characterization of the robustness of these cages.</p>https://resolver.caltech.edu/CaltechTHESIS:12062015-164238181Kinematics and Local Motion Planning for Quasi-static Whole-body Mobile Manipulation
https://resolver.caltech.edu/CaltechTHESIS:05222016-095145651
Year: 2016
DOI: 10.7907/Z9KK98RX
<p>This thesis studies mobile robotic manipulators, where one or more robot manipulator arms are
integrated with a mobile robotic base. The base could be a wheeled or tracked vehicle, or it might be a
multi-limbed locomotor. As robots are increasingly deployed in complex and unstructured environments,
the need for mobile manipulation increases. Mobile robotic assistants have the potential to revolutionize human
lives in a large variety of settings including home, industrial and outdoor environments.</p>
<p>Mobile Manipulation is the use or study of such mobile robots as they interact with physical
objects in their environment. As compared to fixed base manipulators, mobile manipulators can take
advantage of the base mechanism’s added degrees of freedom in the task planning and execution process.
But their use also poses new problems in the analysis and control of base system stability, and the
planning of coordinated base and arm motions. For mobile manipulators to be successfully and
efficiently used, a thorough understanding of their kinematics, stability, and capabilities is required.
Moreover, because mobile manipulators typically possess a large number of actuators, new and efficient
methods to coordinate their large numbers of degrees of freedom are needed to make them practically
deployable. This thesis develops new kinematic and stability analyses of mobile manipulation, and new
algorithms to efficiently plan their motions.</p>
<p>I first develop detailed and novel descriptions of the kinematics governing the operation of multi-
limbed legged robots working in the presence of gravity, and whose limbs may also be simultaneously
used for manipulation. The fundamental stance constraint that arises from simple assumptions about
friction and the ground contact and feasible motions is derived. Thereafter, a local relationship between
joint motions and motions of the robot abdomen and reaching limbs is developed. Baseeon these
relationships, one can define and analyze local kinematic qualities including limberness, wrench
resistance and local dexterity. While previous researchers have noted the similarity between multi-
fingered grasping and quasi-static manipulation, this thesis makes explicit connections between these two
problems.</p>
<p>The kinematic expressions form the basis for a local motion planning problem that that
determines the joint motions to achieve several simultaneous objectives while maintaining stance stability
in the presence of gravity. This problem is translated into a convex quadratic program entitled the
balanced priority solution, whose existence and uniqueness properties are developed. This problem is
related in spirit to the classical redundancy resoxlution and task-priority approaches. With some simple
modifications, this local planning and optimization problem can be extended to handle a large variety of
goals and constraints that arise in mobile-manipulation. This local planning problem applies readily to
other mobile bases including wheeled and articulated bases. This thesis describes the use of the local
planning techniques to generate global plans, as well as for use within a feedback loop. The work in this
thesis is motivated in part by many practical tasks involving the Surrogate and RoboSimian robots at
NASA/JPL, and a large number of examples involving the two robots, both real and simulated, are
provided.</p>
<p>Finally, this thesis provides an analysis of simultaneous force and motion control for multi-
limbed legged robots. Starting with a classical linear stiffness relationship, an analysis of this problem for
multiple point contacts is described. The local velocity planning problem is extended to include
generation of forces, as well as to maintain stability using force-feedback. This thesis also provides a
concise, novel definition of static stability, and proves some conditions under which it is satisfied.</p>https://resolver.caltech.edu/CaltechTHESIS:05222016-095145651Electromyographic Signal Processing With Application To Spinal Cord Injury
https://resolver.caltech.edu/CaltechTHESIS:05312016-211459301
Year: 2016
DOI: 10.7907/Z9QJ7F99
<p>An Electromyogram or Electromyographic (EMG) signal is the recording of the electrical activity produced by muscles. It measures the electric currents generated in muscles during their contraction. The EMG signal provides insight into the neural activation and dynamics of the muscles, and is therefore important for many different applications, such as in clinical investigations that attempt to diagnose neuromuscular deficiencies. In particular, the work in this thesis is motivated by rehabilitation for patients with spinal cord injury. The EMG signal is very important for researchers and practitioners to monitor and evaluate the effect of the rehabilitation training and the condition of muscles, as the EMG signal provides information that helps infer the neural activity in the spinal cord. Before the work in this thesis, EMG analysis required significant amounts of manual labeling of interesting signal features. The motivation of this thesis is to fully automate the EMG analysis tasks and yield accurate, consistent results.</p>
<p>The EMG signal contains multiple muscle responses. The difficulty in processing the EMG signal arises from the fact that the transient muscle response is a transient signal with unknown arrival time, unknown duration, and unknown shape. In addition, the EMG signal recorded from patients with spinal cord injury during rehabilitation is very different from the EMG signal of normal healthy people undergoing the same motions. For example, some of the muscle responses are very weak and thus hard to detect. Because of this, general EMG processing tools and methods are either not applicable or insufficient.</p>
<p>The primary contribution of this thesis is the development of a wavelet-based, double-threshold algorithm for the detection of transient peaks in the EMG signal. The application of wavelet transform in the detection of transient signals has been studied extensively and employed successfully. However, most of the theories assume certain knowledge about the shapes of the transient signals, which makes it hard to be generalized to the transient signals with arbitrary shapes. The proposed detection scheme focuses on the more fundamental feature of most transient signals (in particular the EMG signal): peaks, instead of the shapes. The continuous wavelet transform with Mexican Hat wavelet is employed. This thesis theoretically derived a framework for selecting a set of scales based on the frequency domain information. Ridges are identified in the time-scale space to combine the wavelet coefficients from different scales. By imposing two thresholds, one on the wavelet coefficient and one on the ridge length, the proposed detection scheme can achieve both high recall and high precision. A systematic approach for selecting the optimal parameters via simulation is proposed and demonstrated. Comparing with other state-of-the-art detection methods, the proposed method in this thesis yields a better detection performance, especially in the low Signal-to-Noise-Ratio (SNR) environment.</p>
<p>Based on the transient peak detection result, the EMG signal is further segmented and classified into various groups of monosynaptic Motor Evoked Potentials (MEPs) and polysynaptic MEPs using techniques stemming from Principal Component Analysis (PCA), hierarchical clustering, and Gaussian mixture model (GMM). A theoretical framework is proposed to segment the EMG signal based on the detected peaks. The scale information of the detected peak is used to derive a measure for its effective support. Several different techniques have been adapted together to solve the clustering problem. An initial hierarchical clustering is first performed to obtain most of the monosynaptic MEPs. PCA is used to reduce the number of features and the effect of the noise. The reduced feature set is then fed to a GMM to further divide the MEPs into different groups of similar shapes. The method of breaking down a segment of multiple consecutive MEPs into individual MEPs is derived.</p>
<p>A software with graphic user interface has been implemented in Matlab. The software implements the proposed peak detection algorithm, and enables the physiologists to visualize the detection results and modify them if necessary. The solutions proposed in this thesis are not only helpful to the rehabilitation after spinal cord injury, but applicable to other general processing tasks on transient signals, especially on biological signals.</p>https://resolver.caltech.edu/CaltechTHESIS:05312016-211459301Online Learning for the Control of Human Standing via Spinal Cord Stimulation
https://resolver.caltech.edu/CaltechTHESIS:04172017-163725367
Year: 2017
DOI: 10.7907/Z9BK19DN
<p>Many applications in recommender systems or experimental design need to make decisions online. Each decision leads to a stochastic reward with initially unknown distribution, while new decisions are made based on the observations of previous rewards. To maximize the total reward, one needs to balance between exploring different strategies and exploiting currently optimal strategies within a given set of strategies. This is the underlying trade-off of a number of clinical neural engineering problems, including brain-computer interface, deep brain stimulation, and spinal cord injury therapy. In these systems, complex electronic and computational systems interact with the human central nervous system. A critical issue is how to control the agents to produce results which are optimal under some measure, for example, efficiently decoding the user's intention in a brain-computer interface or performs temporal and spatial specific stimulation in deep brain stimulation. This dissertation is motivated by electrical sipnal cord stimulation with high dimensional inputs(multi-electrode arrays). The stimulation is applied to promote the function and rehabilitation of the remaining neural circuitry below the spinal cord injury, and enable complex motor behaviors such as stepping and standing. To enable the careful tuning of these stimuli for each patient, the electrode arrays which deliver these stimuli have become increasingly more sophisticated, with a corresponding increase in the number of free parameters over which the stimuli need to be optimized. Since the number of stimuli is growing exponentially with the number of electrodes, algorithmic methods of selecting stimuli is necessary, particularly when the feedback is expensive to get.</p>
<p>In many online learning settings, particularly those that involve human feedback, reliable feedback is often limited to pairwise preferences instead of real valued feedback. Examples include implicit or subjective feedback for information retrieval and recommender systems, such as clicks on search results, and subjective feedback on the quality of recommended care. Sometimes with real valued feedback, we require that the sampled function values exceed some prespecified ``safety'' threshold, a requirement that existing algorithms fail to meet. Examples include medical applications where the patients' comfort must be guaranteed; recommender systems aiming to avoid user dissatisfaction; and robotic control, where one seeks to avoid controls that cause physical harm to the platform.</p>
<p>This dissertation provides online learning algorithms for several specific online decision-making problems. \selfsparring optimizes the cumulative reward with relative feedback. RankComparison deals with ranking feedback. \safeopt considers the optimization with real valued feedback and safety constraints. \cduel is designed for specific spinal cord injury therapy.</p>
<p>A variant of \cduel was implemented in closed-loop human experiments, controlling which epidural stimulating electrodes are used in the spinal cord of SCI patients. The results obtained are compared with concurrent stimulus tuning carried out by human experimenter. These experiments show that this algorithm is at least as effective as the human experimenter, suggesting that this algorithm can be applied to the more challenging problems of enabling and optimizing complex, sensory-dependent behaviors, such as stepping and standing in SCI patients.</p>
<p>In order to get reliable quantitative measurements besides comparisons, the standing behaviors of paralyzed patients under spinal cord stimulation are evaluated. The potential of quantifying the quality of bipedal standing in an automatic approach is also shown in this work.</p>https://resolver.caltech.edu/CaltechTHESIS:04172017-163725367Heading Estimation via Sun Sensing for Autonomous Navigation
https://resolver.caltech.edu/CaltechTHESIS:06142017-153929873
Year: 2017
DOI: 10.7907/Z9BG2M1S
<p>In preparation for the mission to Mars in 2020, NASA JPL and Caltech have been exploring the potential of sending a scout robot to accompany the new rover. One of the leading candidates for this scout robot is a lightweight helicopter that can fly every day for ~1 to 3 minutes. Its findings would be critical in the path planning for the rover because of its ability to see over and round local terrain elements. The inconsistent Mars magnetic field and GPS-denied environment would require the navigation system of such a vehicle to be completely overhauled. In this thesis, we present a novel technique for heading estimation for autonomous vehicles using sun sensing via fisheye camera. The approach results in accurate heading estimates within 2.4° when relying on the camera alone. If the information from the camera is fused with our sensors, the heading estimates are even more accurate. While this does not yet meet the desired error bound, it is a start with the critical flaws in the algorithm already identified in order to improve performance significantly. This lightweight solution however shows promise and does meet the weight constraints for the 1 kg Mars 2020 Helicopter Scout.</p>https://resolver.caltech.edu/CaltechTHESIS:06142017-153929873Dynamic Modeling and Control of Spherical Robots
https://resolver.caltech.edu/CaltechTHESIS:05302018-110559204
Year: 2018
DOI: 10.7907/E5CW-8H41
<p>In this work, a rigorous framework is developed for the modeling and control of spherical robotic vehicles. Motivation for this work stems from the development of Moball, which is a self-propelled sensor platform that harvests kinetic energy from local wind fields. To study Moball's dynamics, the processes of Lagrangian reduction and reconstruction are extended to robotic systems with symmetry-breaking potential energies, in order to simplify the resulting dynamic equations and expose mathematical structures that play an important role in subsequent control-theoretic tasks. These results apply to robotic systems beyond spherical robots. A formulaic procedure is introduced to derive the reduced equations of motion of most spherical robots from inspection of the Lagrangian. This adaptable procedure is applied to a diverse set of robotic systems, including multirotor aerial vehicles.</p>
<p>Small time local controllability (STLC) results are derived for barycentric spherical robots (BSR), which are spherical vehicles whose locomotion depends on actuating the vehicle's center of mass (COM) location. STLC theorems are introduced for an arbitrary BSR on flat, sloped, or smooth terrain. I show that STLC depends on the surjectivity of a simple <i>steering matrix</i>. An STLC theorem is also derived for a class of commonly encountered multirotor vehicles.</p>
<p>Feedback linearizing and PID controllers are proposed to stabilize an arbitrary spherical robot to a desired trajectory over smooth terrain, and direct collocation is used to develop a feedforward controller for Moball specifically. Moball's COM is manipulated by a novel system of magnets and solenoids, which are actuated by a "ballistic-impulse" controller that is also presented. Lastly, a motion planner is developed for energy-harvesting vehicles. This planner charts a path over smooth terrain while balancing the desire to achieve scientific objectives, avoid hazards, and the imperative of exposing the vehicle to environmental sources of energy such as local wind fields and topology. Moball's design details and experimental results establishing Moball's energy-harvesting performance (7<i>W</i> while rolling at a speed of 2 <i>m/s</i>), are contained in an Appendix.</p>https://resolver.caltech.edu/CaltechTHESIS:05302018-110559204Optimal Controller Synthesis for Nonlinear Systems
https://resolver.caltech.edu/CaltechTHESIS:12162017-121220572
Year: 2018
DOI: 10.7907/Z9TX3CK8
<p>Optimal controller synthesis is a challenging problem to solve. However, in many applications such as robotics, nonlinearity is unavoidable. Apart from optimality, correctness of the system behaviors with respect to system specifications such as stability and obstacle avoidance is vital for engineering applications. Many existing techniques consider either the optimality or the correctness of system behavior. Rarely, a tool exists that considers both. Furthermore, most existing optimal controller synthesis techniques are not scalable because they either require ad-hoc design or they suffer from the curse of dimensionality.</p>
<p>This thesis aims to close these gaps by proposing optimal controller synthesis techniques for two classes of nonlinear systems: linearly solvable nonlinear systems and hybrid nonlinear systems. Linearly solvable systems have associated Hamilton- Jacobi-Bellman (HJB) equations that can be transformed from the original nonlinear partial differential equation (PDE) into a linear PDE through a logarithmic transformation. The first part of this thesis presets two methods to synthesize optimal controller for linearly solvable nonlinear systems. The first technique uses a hierarchy of sums-of-square programs to compute a sequence of suboptimal controllers that have non-increasing suboptimality for first exit and finite horizon problems. This technique is the first systematic approach to provide stability and suboptimal performance guarantees for stochastic nonlinear systems in one framework. The second technique uses the low rank tensor decomposition framework to solve the linear HJB equation for first exit, finite horizon, and infinite horizon problems. This technique scale linearly with dimensions, alleviating the curse of dimensionality and enabling us to solve the linear HJB equation for a quadcopter model that is a twelve-dimensional system on a personal laptop. A new algorithm is proposed for a key step in the controller synthesis algorithm to solve the ill-conditioning issue that arises in the original algorithm. A MATLAB toolbox that implements the algorithms is developed, and the performance of these algorithms is illustrated by a few engineering examples.</p>
<p>Apart from stability, in many applications, more complex specifications such as obstacle avoidance, reachability, and surveillance are required. The second part of the thesis describes methods to synthesize optimal controllers for hybrid nonlinear systems with quantitative objectives (i.e., minimizing cost) and qualitative objectives (i.e., satisfying specifications). This thesis focuses on two types of qualitative objectives, regular objectives, and ω-regular objectives. Regular objectives capture bounded time behavior such as reachability, and ω-regular objectives capture long term behavior such as surveillance. For both types of objectives, an abstraction-refinement procedure that preserves the cost is developed. A two-player game is solved on the product of the abstract system and the given objectives to synthesize the suboptimal controller for the hybrid nonlinear system. By refining the abstract system, the algorithms are guaranteed to converge to the optimal cost and return the optimal controller if the original systems are robust with respect to the initial states and the optimal controller inputs. The proposed technique is the first abstraction-refinement based technique to combine both quantitative and qualitative objectives into one framework. A Python implementation of the algorithms are developed, and a few engineering examples are presented to illustrate the performance of these algorithms.</p>https://resolver.caltech.edu/CaltechTHESIS:12162017-121220572An Electrophysiological Study Of Voluntary Movement and Spinal Cord Injury
https://resolver.caltech.edu/CaltechTHESIS:06012018-140912331
Year: 2018
DOI: 10.7907/K6P2-ZH75
<p>Voluntary movement is generated from the interaction between neurons in our brain and the neurons in our spinal cord that engage our muscles. A spinal cord injury destroys the connection between these two regions, but parts of their underlying neural circuits survive. A new class of treatment (the brain-machine interface) takes advantage of this fact by either a) recording neural activity from the brain and predicting the intended movement (neural prosthetics) or b) stimulating neural activity in the spinal cord to facilitate muscle activity (spinal stimulation). This thesis covers new research studying the brain-machine interface and its application for spinal injury.</p>
<p>First, the electrical properties of the microelectrode (the main tool of the brain-machine interface) are studied during deep brain recording and stimulation. This work shows that the insulation coating the electrode forms a capacitor with the surrounding neural tissue. This capacitance causes large spikes of voltage in the surrounding tissue during deep brain stimulation, which will cause electrical artifacts in neural recordings and may damage the surrounding neurons. This work also shows that a coaxially shielded electrode will block this effect.</p>
<p>Second, the activity of neurons in the parietal cortex is studied during hand movements, which has applications for neural prosthetics. Prior work suggests that the parietal cortex encodes a state-estimator [1], which combines sensory feedback with the internal efference copy to predict the state of the hand. To test this idea, we used a visual lag to misalign sensory feedback from the efference copy. The expectation was that a state-estimator would unknowingly combine the delayed visual feedback with the current efference information, resulting in incorrect predictions of the hand. Our results show a drop in correlation between neural activity in the parietal cortex and hand movement during a visual lag, supporting the idea that the parietal cortex encodes a state-estimator. This correlation gradually recovers over time, showing that parietal cortex is adaptive to sensory delays.</p>
<p>Third, while the intention of spinal stimulation was to interact locally with neural circuits in the spinal cord, results from the clinic show that electrical stimulation of the lumbosacral enlargement enables paraplegic patients to regain voluntary movement of their legs [2]. This means that spinal stimulation facilitates communication across an injury site. To further study this effect, we developed a new behavioral task in the rodent. Rats were trained to kick their right hindlimb in response to an auditory cue. The animals then received a spinal injury that caused paraplegia. After injury, the animals recovered the behavior (they could kick in response to the cue), but only during spinal stimulation. Their recovered behavior was slower and more stereotyped than their pre-injury response. Administering quipazine to these rodents disrupted their ability to respond to the cue, suggesting that serotonin plays an important role in the recovered pathway. This work proves that the new behavioral task is a successful tool for studying the recovery of voluntary movement.</p>
<p>Future work will combine cortical recordings with this behavioral task in the rodent to study plasticity in the nervous system and improve treatment of spinal cord injuries.</p>
<p>[1] Mulliken, Grant H., Sam Musallam, and Richard A. Andersen. "Forward estimation of movement state in posterior parietal cortex." Proceedings of the National Academy of Sciences105.24 (2008): 8170-8177.</p>
<p>[2] Harkema, Susan, et al. "Effect of epidural stimulation of the lumbosacral spinal cord on voluntary movement, standing, and assisted stepping after motor complete paraplegia: a case study." The Lancet 377.9781 (2011): 1938-1947.</p>https://resolver.caltech.edu/CaltechTHESIS:06012018-140912331Towards High Performance Robotic Actuation
https://resolver.caltech.edu/CaltechTHESIS:05222019-132217207
Year: 2019
DOI: 10.7907/W64Q-1R69
<p>The main objective of this thesis is to enable development of high performance actuation for legged, limbed and mobile robots. Due to the fact that such robots need to support their own weight, their actuators need to be light weight, compact and efficient. Furthermore, a dynamics analysis, shows that the actuators' design may have significant impact on a robot's dynamics sensitivity. These consideration motivate improvements in all actuator design aspects.</p>
<p>First, the application-specific design of outer rotor motors with concentrated windings is considered. It is shown that an intrinsic design trade-off exists between a motor's copper loss, core loss and mass, which allows development of motors with superior performance for a particular application. The three main application categories of interest are: electric vehicles, drones and robotic joints. Due to their outstanding torque density, high pole count outer rotor motors are analysed in terms of their design and optimization for robotic applications. Motor design scaling modes are also described in order to outline the main challenges in the implementation of high torque motors.</p>
<p>Next, the design of gearboxes for robotic actuation is discussed. A novel type of high reduction Bearingless Planetary Gearbox is introduced which allows large range of reduction ratios to be achieved in a compound planetary stage. In this concept, all gear components float in an unconstrained manner as the planet carrier is substituted with a secondary sun gear. The advantages of the Bearingless Planetary Gearbox over current approaches in terms of improved robustness, load distribution, manufacturability, and assembly are outlined.</p>
<p>Finally, analysis, design, and prototyping of rotary planar springs for rotary series elastic actuators is described. A mathematical model, based on curved beam theory, that allows rapid design, analysis, and comparison of rotary springs is developed. Mass reduction techniques based on composite arm structures are introduced and internal arm contact modeling is presented. Motivated by strain energy density analysis, an optimization based spring design approach is developed that allows significant increase in the torque and torque density.</p>
https://resolver.caltech.edu/CaltechTHESIS:05222019-132217207Numerical Investigation of Spinal Neuron Facilitation with Multi-electrode Epidural Stimulation
https://resolver.caltech.edu/CaltechTHESIS:11302018-185025297
Year: 2019
DOI: 10.7907/2DVK-G212
<p>Approximately 1,275,000 people in the US have a spinal cord injury severe enough to cause some paralysis of the arms and/or legs. Epidural stimulation using implanted multi-electrode stimulating arrays over the lumbosacral spinal cord has recently shown promise in assisting individuals with severe spinal cord injuries to stand, walk, and even facilitate voluntary movement. Both animal model and human studies have shown that sub-threshold facilitation of motor recovery gives the best results. The underlying neural mechanisms by which sub-threshold epidural stimulation leads to motor recovery are incompletely known.</p>
<p>This thesis uses computational methods to study the <i>facilitation effect</i>. A neuron is facilitated if a sub-threshold synaptic input can cause a neuronal output under the influence of a stimulating electric field. The analysis in this thesis is based on a computational model of the epidural spinal stimulation process in the rat spinal cord. This model includes a time-domain finite element simulation (using COMSOL®) of the various tissues in the spinal cord with the appropriate anisotropic and frequency-dependent complex relative permittivities. The voltages obtained from the finite element simulations were used as the extracellular voltage in NEURON simulations.</p>
<p>A population of neurons were simulated under a wide variety of conditions. These simulations highlight the effect of neuron orientation, location, and synaptic timing as key parameters which influence facilitation.</p>
<p>This study indicates that regions of the spinal cord that have previously been ignored may be actively involved in motor recovery. These results may also enable the design of specialized epidural electrode arrays and the design of new stimulation protocols.</p>https://resolver.caltech.edu/CaltechTHESIS:11302018-185025297Functional Autonomy Techniques for Manipulation in Uncertain Environments
https://resolver.caltech.edu/CaltechTHESIS:06082020-104419929
Year: 2020
DOI: 10.7907/0kgt-yg76
<p>As robotic platforms are put to work in an ever more diverse array of environments, their ability to deploy visuomotor capabilities without supervision is complicated by the potential for unforeseen operating conditions. This is a particular challenge within the domain of manipulation, where significant geometric, semantic, and kinetic understanding across the space of possible manipulands is necessary to allow effective interaction. To facilitate adoption of robotic platforms in such environments, this work investigates the application of functional, or behavior level, autonomy to the task of manipulation in uncertain environments. Three functional autonomy techniques are presented to address subproblems within the domain.</p>
<p>The task of reactive selection between a set of actions that incur a probabilistic cost to advance the same goal metric in the presence of an operator action preference is formulated as the Obedient Multi-Armed Bandit (OMAB) problem, under the purview of Reinforcement Learning. A policy for the problem is presented and evaluated against a novel performance metric, disappointment (analogous to prototypical MAB's regret), in comparison to adaptations of existing MAB policies. This is posed for both stationary and non-stationary cost distributions, within the context of two example planetary exploration applications of multi-modal mobility, and surface excavation.</p>
<p>Second, a computational model that derives semantic meaning from the outcome of manipulation tasks is developed, which leverages physics simulation and clustering to learn symbolic failure modes. A deep network extracts visual signatures for each mode that may then guide failure recovery. The model is demonstrated through application to the archetypal manipulation task of placing objects into a container, as well as stacking of cuboids, and evaluated against both synthetic verification sets and real depth images.</p>
<p>Third, an approach is presented for visual estimation of the minimum magnitude grasping wrench necessary to extract massive objects from an unstructured pile, subject to a given end effector's grasping limits, that is formulated for each object as a "wrench space stiction manifold". Properties are estimated from segmented RGBD point clouds, and a geometric adjacency graph used to infer incident wrenches upon each object, allowing candidate extraction object/force-vector pairs to be selected from the pile that are likely to be within the system's capability.</p>https://resolver.caltech.edu/CaltechTHESIS:06082020-104419929Tethered Motion Planning for a Rappelling Robot
https://resolver.caltech.edu/CaltechTHESIS:06012020-230913819
Year: 2020
DOI: 10.7907/h7d4-ww72
<p>The Jet Propulsion Laboratory and Caltech developed the Axel rover to investigate and demonstrate the potential for tethered extreme terrain mobility, such as allowing access to science targets on the steep crater walls of other planets. Tether management is a key issue for Axel and other rappelling rovers. Avoiding tether entanglement constrains the robot's valid motions to the set of outgoing and returning path pairs that are homotopic to each other. In the case of a robot on a steep slope, a motion planner must additionally ensure that this ascent-descent path pair is feasible, based on the climbing forces provided by the tether. This feasibility check relies on the taut tether configuration, which is the shortest path in the homotopy class of the ascent-descent path pair. </p>
<p>This dissertation presents a novel algorithm for tethered motion planning in extreme terrains, produced by combining shortest-homotopic-path algorithms from the topology and computational geometry communities with traditional graph search methods. The resulting tethered motion planning algorithm searches for this shortest path, checks for feasibility, and then generates waypoints for an ascent-descent path pair in the same homotopy class. I demonstrate the implementation of this algorithm on a Martian crater data set such as might be seen for a typical mission. By searching only for the shortest path, and ordering that search according to a heuristic, this algorithm proceeds more efficiently than previous tethered path-planning algorithms for extreme terrain. </p>
<p>Frictional tether-terrain interaction may cause dangerously intermittent and unstable tether obstacles, which can be categorized based on their stability. Force-balance equations from the rope physics literature provide a set of tether and terrain conditions for static equilibrium, which can be used to determine if a given tether configuration will stick to a given surface based on tether tension. By estimating the tension of Axel's tether when driving, I divide potential tether tension obstacles into the following categories: acting as obstacles, acting as non-obstacles, and hazardous intermittent obstacles where it is uncertain whether the tether would slip or stick under normal driving tension variance. This dissertation describes how to modify the obstacle map as the categorization of obstacles fluctuates, and how to alter a motion plan around the dangerous tether friction obstacles. Together, these algorithms and methods form a framework for tethered motion planning on extreme terrain.</p>https://resolver.caltech.edu/CaltechTHESIS:06012020-230913819Towards Learning Robotic Dynamics: Application to Multirotor Takeoff and Landing
https://resolver.caltech.edu/CaltechTHESIS:03152021-082447788
Year: 2021
DOI: 10.7907/199j-dk87
<p>Multirotors have become widespread but their usage is still limited. Ensuring safety during take-off and landing is still an open problem. Towards this goal this thesis proposes two different solutions to address this problem. The two approaches complement each other and they are tested on hardware.</p>
<p>The first approach is to design a vehicle that is stable during take-off, despite hardware failures or unsteady take-off platforms. A solution is to use a ballistic launch to impose a deterministic path, preventing collisions with its environment. Following this approach led to the development of several SQUID (<i>Streamlined Quick Unfolding Investigation Drone</i>) vehicles. The main challenges are the ballistic initial flight, large accelerations during launch, and limited volume. A first prototype was developed, which is able to transition mid-flight from stable ballistic flight to a fully controllable multirotor. The system has been fabricated and field tested from a moving vehicle up to 50mph to successfully demonstrate the feasibility of the concept and experimentally validate the design's aerodynamic stability and deployment reliability. A second prototype expanded the first one's capabilities incorporating fully-autonomous vision-based navigation, while keeping the ballistic passive stability and stable transition abilities. The new design includes a more reliable plate-based structure and more effective folding fins.</p>
<p>The second approach focuses on designing controllers that are safe regardless of the platform. For that purpose, a Model Predictive Control (MPC) is used to ensure state and input constraints. Given the highly non-linear dynamics platforms and fast dynamics that require a quick controller evaluation, the work in this thesis is built using Koopman Operator theory, which allows tools from linear analysis to be applied to systems with inherently non-linear dynamics. One of the main contributions is a novel method to find Koopman Eigenfunctions directly from data. Another key contribution is an episodic approach to model non-linear actuation dynamics. The proposed method is first tested on simulation and it outperforms comparable approaches. The method is also demonstrated on-board a multirotor for a fast landing application, where the nonlinear ground effect is learned and used to improve landing speed and quality. An additional extension considers model uncertainty in the MPC architecture, where an Ensemble Kalman Sampler is used to learn the uncertainty distribution.</p>https://resolver.caltech.edu/CaltechTHESIS:03152021-082447788Assuring Safety under Uncertainty in Learning-Based Control Systems
https://resolver.caltech.edu/CaltechTHESIS:01052021-195655093
Year: 2021
DOI: 10.7907/9kye-rn93
<p>Learning-based controllers have recently shown impressive results for different robotic tasks in well-defined environments, successfully solving a Rubiks cube and sorting objects in a bin. These advancements promise to enable a host of new capabilities for complex robotic systems. However, these learning-based controllers cannot yet be deployed in highly uncertain environments due to significant issues relating to learning reliability, robustness, and safety.</p>
<p>To overcome these issues, this thesis proposes new methods for integrating model information (e.g. model-based control priors) into the reinforcement learning framework, which is crucial to ensuring reliability and safety. I show, both empirically and theoretically, that this model information greatly reduces variance in learning and can effectively constrain the policy search space, thus enabling significant improvements in sample complexity for the underlying RL algorithms. Furthermore, by leveraging control barrier functions and Gaussian process uncertainty models, I show how system safety can be maintained under uncertainty without interfering with the learning process (e.g. distorting the policy gradients).</p>
<p>The last part of the thesis will discuss fundamental limitations that arise when utilizing machine learning to derive safety guarantees. In particular, I show that widely used uncertainty models can be highly inaccurate when predicting rare events, and examine the implications of this for safe learning. To overcome some of these limitations, a novel framework is developed based on assume-guarantee contracts in order to ensure safety in multi-agent human environments. The proposed approach utilizes contracts to impose loose responsibilities on agents in the environment, which are learned from data. Imposing these responsibilities on agents, rather than treating their uncertainty as a purely random process, allows us to achieve both safety and efficiency in interactions.</p>https://resolver.caltech.edu/CaltechTHESIS:01052021-195655093Online Learning from Human Feedback with Applications to Exoskeleton Gait Optimization
https://resolver.caltech.edu/CaltechTHESIS:12092020-162149429
Year: 2021
DOI: 10.7907/gvtx-1586
<p>Systems that intelligently interact with humans could improve people's lives in numerous ways and in numerous settings, such as households, hospitals, and workplaces. Yet, developing algorithms that reliably and efficiently personalize their interactions with people in real-world environments remains challenging. In particular, one major difficulty lies in adapting to human-in-the-loop feedback, in which an algorithm makes sequential decisions while receiving online feedback from humans; throughout this interaction, the algorithm seeks to optimize its decision-making quality, as measured by the utility of its performance to the human users. Such algorithms must balance between exploration and exploitation: on one hand, the algorithm must select uncertain strategies to fully explore the environment and the interacting human's preferences, while on the other hand, it must exploit the empirically-best-performing strategies to maximize its cumulative performance.</p>
<p>Learning from human feedback can be difficult, as people are often unreliable in specifying numerical scores. In contrast, humans can often more accurately provide various types of qualitative feedback, for instance pairwise preferences. Yet, sample efficiency is a significant concern in human-in-the-loop settings, as qualitative feedback is less informative than absolute metrics, and algorithms can typically pose only limited queries to human users. Thus, there is a need to create theoretically-grounded online learning algorithms that efficiently, reliably, and robustly optimize their interactions with humans while learning from online qualitative feedback.</p>
<p>This dissertation makes several contributions to algorithm design for human-in-the-loop learning. Firstly, this work develops the Dueling Posterior Sampling (DPS) algorithmic framework, a model-based, Bayesian approach for online learning in the settings of preference-based reinforcement learning and generalized linear dueling bandits. DPS is developed together with a theoretical regret analysis framework, and yields competitive empirical performance in a range of simulations. Additionally, this thesis presents the CoSpar and LineCoSpar algorithms for sample-efficient, mixed-initiative learning from pairwise preferences and coactive feedback. CoSpar and LineCoSpar are both deployed in human subject experiments with a lower-body exoskeleton to identify optimal, user-preferred exoskeleton walking gaits. This work presents the first demonstration of preference-based learning for optimizing dynamic crutchless exoskeleton walking for user comfort, and makes progress toward customizing exoskeletons and other assistive devices for individual users.</p>https://resolver.caltech.edu/CaltechTHESIS:12092020-162149429Autonomous Mission-Driven Robots in Extreme Environments
https://resolver.caltech.edu/CaltechTHESIS:05172022-043237609
Year: 2022
DOI: 10.7907/a78d-kv42
<p>Robotic autonomy systems that can negotiate harsh environments under time and communication constraints are critical to accomplishing many real-world missions. Such systems require an integrated software-hardware solution capable of robustly reasoning about a time-limited mission across a complex environment and negotiating extreme physical conditions during mission execution. To this end, I will discus the development of two field-tested systems designed for operation in GPS-denied areas: (i) a coverage planning framework that enables efficient exploration of large, unknown environments, and (ii) a ballistically-launched aircraft that converts to an autonomous, free-flying multirotor in order to provide rapid aerial surveillance.</p>
<p>The first system addresses the time-limited exploration problem by providing a planning strategy that seeks to maximize the area covered by a robot’s sensor footprint along a planned trajectory. In order to find solutions over large spatial extents (>1 km) and long temporal horizons (>1 hour), this coverage problem is decomposed into tractable subproblems by introducing spatial and temporal abstractions. Spatially, the robot-world belief is approximated by a task-dependent structure, enriched with environment map estimates. Temporally, the belief is approximated by the aggregation of multiple structures, each spanning a different spatial range. Cascaded uncertainty-aware solvers return a coverage plan over the stratified belief in real time.
Coverage policies are constructed in a receding horizon fashion to ensure motion smoothness and resiliency to real-world stochasticity in perception and control. This coverage planning framework was extensively tested on physical robots in various real-world environments (caves, mines, subway systems, etc.) and served as the exploration strategy for a competing entry in the DARPA Subterranean Challenge.</p>
<p>The second system addresses rapid multirotor deployment for aerial data collection during emergencies. While multirotors are advantageous over fixed-winged systems due to their high maneuverability, their rotating blades are hazardous and require stable, uncluttered takeoff sites. To overcome this issue, a ballistically-launched, autonomously-stabilizing multirotor (SQUID -- Streamlined Quick Unfolding Investigation Drone) was designed, fabricated, and tested. SQUID follows a deterministic trajectory, transitioning from a folded launch configuration to an autonomous, fully-controllable hexacopter. The entire process from launch to position stabilization requires no user- or GPS-input and demonstrates the viability of using ballistically-launched multirotors to achieve safe and rapid deployment from moving vehicles.</p>https://resolver.caltech.edu/CaltechTHESIS:05172022-043237609Koopman-based Learning and Control of Agile Robotic Systems
https://resolver.caltech.edu/CaltechTHESIS:10122021-213903517
Year: 2022
DOI: 10.7907/2t6d-j206
<p>Learning methods to enable high performance control systems have recently shown promising results in selected environments and applications. These advances promote the next generation of autonomous robots capable of significantly improving efficiency, cost, and safety in their respective domains. Importantly, these systems are <i>safety-critical</i> and operate in proximity to humans in diverse and uncertain environments. As a result, operational failures may cause significant material and societal losses. Additionally, robot learning and control are further complicated by requiring fast controller update rates and operational constraint satisfaction.</p>
<p>To address these challenges, this thesis presents multiple methods based on Koopman operator theory. The first approach develops algorithms to learn lifted-dimensional models of nonlinear systems and leverages the models in model predictive control (MPC) design. Koopman-based methods typically employ hand-crafted observable functions to "lift" the state variables to the higher dimensional space. For most systems, this leads to poor prediction performance and inefficient use of data and computational resources. Instead, I present methods that generate observable functions from data, both based on underlying theory and by incorporating the observable functions and model structure in a neural network model. This allows lower dimensional models, important for real-time control, and enables the nonlinearities of control-affine dynamics to be captured, crucial to describing many robotic systems. I use quadrotor drones to experimentally demonstrate that the learned models combined with MPC can achieve close to optimal behavior while respecting important operational constraints.</p>
<p>The last part of the thesis is concerned with endowing systems with an arbitrary nominal control policy with safety guarantees. Control barrier functions (CBFs) are a powerful tool to achieve this, yet they rely on the computation of control invariant sets, which is notoriously difficult. To avoid this, a backup strategy can be used to implicitly define a control invariant set. However, this requires forward integration of the system dynamics under a backup controller, which is prohibitively expensive for realistic systems. I present a method that replaces the expensive integration using learned Koopman operators of the closed-loop dynamics. As a result, the online computation time required to evaluate the controller is drastically reduced, enabling real-time use. I also derive an error bound on the unmodeled dynamics in order to robustify the CBF controller and demonstrate the method on multi-agent collision avoidance for wheeled robots and quadrotors.</p>https://resolver.caltech.edu/CaltechTHESIS:10122021-213903517Autonomous Temporal Understanding and State Estimation during Robot-Assisted Surgery
https://resolver.caltech.edu/CaltechTHESIS:05272022-171138586
Year: 2022
DOI: 10.7907/n58k-tr61
<p>Robot-Assisted Surgery (RAS) has become increasingly important in modern surgical practice for its many benefits and advantages for both the patient and the healthcare professionals, as compared to traditional open surgeries and minimally invasive surgeries such as laparoscopy. Artificial intelligence applications during RAS and post-operative analysis can provide various surgeon-assisting functionalities and could potentially achieve a better surgery outcome. These applications, ranging from providing surgeons with advisory information during RAS and post-operative analysis to virtual fixture and supervised autonomous surgical tasks, share a necessary prerequisite of a comprehensive understanding of the current surgical scene. This understanding should include the knowledge of the current surgical task being performed, the surgeon's actions and gestures, the state of the patient, etc. Currently, there is yet to be a unified effort to achieve the autonomous temporal understanding and perception of an RAS at the high accuracy and efficiency required in the highly safety-critical field of medicine.</p>
<p>This thesis develops novel modeling methodologies and deep learning-based models for the autonomous perception and temporal segmentation of the current surgical scene during an RAS. An RAS procedure is modeled as a hierarchical system consisting of discrete surgical states at multiple levels of temporal granularity. These surgical states take the form of surgical tasks, operational steps, fine-grained surgical actions, etc. A broad range of computational experiments were performed to develop methods that achieve an accurate, robust, and efficient estimation of these surgical states. Multiple novel deep learning-based models for feature extraction, noise elimination, and efficient training were proposed and tested. This thesis also shows the significant benefits of incorporating multiple types of data streams recorded by the surgical robotic system to a more accurate surgical state estimation effort.</p>
<p>Two new RAS datasets that contains real-world RAS procedures and diverse experimental settings were collected and annotated--filling a gap in the data sets available for the development and testing of of robust surgical state estimation models. The performance and robustness of models in this thesis work were showcased with these highly complex and dynamic real-world RAS datasets and compared against state-of-the-art methods. A significant model performance improvement was observed in both surgical state estimation accuracy and efficiency. The modeling methodologies and deep learning-based models developed in this work have diverse potential applications to the development of a next-generation surgical robotic systems.</p>https://resolver.caltech.edu/CaltechTHESIS:05272022-171138586Risk-Aware Planning and Control in Extreme Environments
https://resolver.caltech.edu/CaltechTHESIS:02082023-223824752
Year: 2023
DOI: 10.7907/xv2b-tj24
<p>Safety-critical control and planning for autonomous systems operating in unstructured environments is a challenging problem must be addressed as autonomous vehicles, surgical robots, and autonomous industrial robots become more pervasive. This thesis addresses some of the issues in safety critical autonomy by introducing new techniques for computationally tractable and efficient safety-critical control. The approach developed in this thesis arises from taking a deeper look at two questions: 1) How can we obtain better uncertainty quantification of the disturbances that affect autonomous systems either as a result of unmodeled changes in the environment or due to sensor imperfections? 2) Given richer uncertainty quantification techniques, how do incorporate the diverse uncertainty descriptions into the control and planning framework without sacrificing the tractability and efficiency of existing approaches?</p>
<p>I address the above two questions by developing risk-aware control and planning techniques for traversal of a mobile robot over static but extreme terrain and in the presence of dynamic obstacles. We first look at algorithms for risk-aware terrain assessment, and extensively test them on wheeled and legged robots that were deployed in subterranean tunnel, urban, and cave environments for search and rescue operations in the DARPA Subterranean Challenge. I then present a theory for risk-aware model predictive control in static environments and in the presence of dynamic obstacles. Coherent risk measures are applied to this planning and control framework in order to account for diverse uncertainty descriptions. Computationally tractable reformulations of the optimal control problem are realized through constraint tightening techniques.</p>
<p>I then investigate algorithms for uncertainty assessment and prediction of apriori unknown, dynamic obstacles using data-driven techniques. We use a technique from signal processing literature called Singular Spectrum Analysis for making linear predictions of dynamic obstacles. The obstacle motion predictions are equipped with error predictions to account for the uncertainty in the sensing heuristically using bootstrapping techniques. We use a statistical tool, Adaptive Conformal Inference, to further calibrate the heuristic error prediction online to obtain true uncertainty prediction while using nonstationary data to analyze the performance of the data-driven predictor. These techniques provide reactive, real-time, risk-aware obstacle avoidance in dynamic environments.</p>https://resolver.caltech.edu/CaltechTHESIS:02082023-223824752Data-Driven Safety-Critical Autonomy in Unknown, Unstructured, and Dynamic Environments
https://resolver.caltech.edu/CaltechTHESIS:03042024-201031352
Year: 2024
DOI: 10.7907/qpbp-0x81
<p>This thesis addresses the critical challenge of ensuring safety in autonomous exploration within unknown, unstructured, dynamic environments, a domain filled with various types of uncertainties. These include model uncertainties in system dynamics, localization uncertainties stemming from measurement noises, and the risks of collision in environments with dynamic obstacles. Traditional models for vehicle planning and control are often simplified for computational feasibility, but this simplification without careful analysis can compromise safety and system stability. My research introduces a novel, comprehensive framework to provide probabilistically safe planning and control for robot autonomy, structured around three components:</p>
<p>(1) Probabilistic Uncertainty Quantification for Model Mismatches: </p>
<p>This segment focuses on identifying model discrepancies given closed-loop tracking data in an unstructured environment where a reduced-order robot model is used for planning and control. The disturbance is modeled as a scalar-valued stochastic process of a norm on the difference between the reduce-order robot model and actual system evolution. In an online and risk-aware framework, Gaussian Process Regression is employed to extract the probabilistic upper bound to such stochastic process, referred to as the Surface-at-Risk. Theoretical guarantees on the accuracy of the fitted discrepancy surface are analyzed and verified to the data sets collected during system operation. </p>
<p>In an offline setting, conformal prediction, a statistical inference tool, is employed to obtain probabilistic upper bounds of matched and unmatched model disturbance in the system from data, without any assumption of the latent probability distribution governing these discrepancies. Building on these bounds, the robot's nominal ancillary controller is augmented for extending robustness and stability guarantees of the closed-loop system in the face of such discrepancies. Additionally, a maximum tracking error tube is constructed along the planned trajectory using the reduced-order model. Such error tubes describe the maximum permissible deviation in actual trajectory tracking under the augmented ancillary controller and the worst-case matched and unmatched model uncertainties, thereby delineating safe operational boundaries for the system. </p>
<p>(2) Data-Driven Unsafe Set Prediction for Dynamic Obstacles: </p>
<p>This thesis topic develops an online, data-driven predictive model for dynamic obstacles, accounting for measurement noise and low-frequency data rates.
First inspired by singular spectrum analysis (SSA), a time-series forecast technique, obstacle models characterized by linear recurrence relationships are extracted from real-time position observables. Using the statistical bootstrap technique, a set of predicted obstacle trajectories are constructed, which in turn are reformulated into deterministic distributionally robust obstacle avoidance constraints, reflecting a user-defined risk tolerance. </p>
<p>Further refining the obstacle predictor for intention-unknown obstacles, a linear, time-varying model is learned from data using time-delay embedding of obstacle position observables. Additive process and measurement noises are anticipated in the learned model, where their intensities are estimated from data. For inferring prediction uncertainties, a companion data-driven Kalman Filter (DDKF) is constructed to forecast obstacle positions and uncertainties. This "heuristic unsafe set" from DDKF is then dynamically calibrated using adaptive conformal prediction, ensuring safety without relying on any distribution assumptions regarding the uncertainties or model accuracy. The calibrated sets, called conformal prediction sets, are then reformulated into convex state constraints.</p>
<p>(3) Safety-Critical Planning:</p>
<p>The thesis proposes two methods for ensuring safety in planning and navigation: Probabilistic-Safe Model Predictive Control (MPC) and Probabilistic-Safe Model Predictive Path Integral (MPPI) given uncertainties arising from operating in unknown, unstructured, and dynamic environments. The MPC approach integrates the quantified obstacle avoidance constraints into a convex program to balance computational tractability while providing probabilistic safety guarantees. In contrast, the MPPI method, a sampling-based strategy, incorporating unsafe sets into a cost map derived from sensory data, optimizes reference tracking trajectory while guaranteeing collision avoidance up to a user-defined risk tolerance.</p>
<p>In unknown and cluttered environments automatically, the proposed framework learns an upper bound on model residuals from data and systematically calculates the safety buffers needed to provide the desired probabilistic safe navigation of robotics systems. Additionally, in the presence of dynamic obstacles, the proposed data-driven predictor systematically extracts an obstacle model and makes obstacle-occupied unsafe set forecasts. These features largely eliminate the "hand tuning" of the underlying planner and controller that is normally required in heuristic-based algorithms. The efficacy of these proposed frameworks is empirically validated through Monte Carlo Simulations, alongside hardware validations on both ground and aerial vehicles, demonstrating their robustness, versatility, and applicability in real-world scenarios.</p>https://resolver.caltech.edu/CaltechTHESIS:03042024-201031352