Update (June 2008):I have successfully defended my thesis, and am now a postdoctoral researcher at the University of Pennsylvania.

Introduction

I am a Ph.D. student in the robotics program at Carnegie Mellon University's Robotics Institute, where I study the application of gait regulation and feedback control for underactuated, dynamic legged robots. My research interests broadly focus on the development of control methods for unique and novel robotic mechanisms, using both analytic studies as well as pragmatic experimental results to drive the state-of-the-art in robotics research. In particular, I am interested in interpreting robotic systems as topological spaces, and studying constraints that exist in those spaces to determine useful control approaches.

I am currently seeking a postdoctoral and/or research position in robotics, commencing mid-2008.

Curriculum Vitae

Development of Climbing Behaviors

RiSE climbing a multistory building using a gait-based feedback climbing behavior.

My thesis work develops feedback behaviors for underactuated legged robots, via a dimensionality reduction in the parameters used to control complex behaviors. Using this control approach, I have developed real-time feedback behaviors that allow a legged robot to climb surfaces such as vertical exterior building walls as well as other challenging surfaces, such as tree trunks.

Rather than working with the full configuration space of a legged robot's joints, I model typical behaviors using open-loop gaits—patterns of leg motions that, in the absence of any feedback, can produce effective locomotion—and introduce parametrizations of gaits on which layers of feedback are added. Within this basic paradigm of mixing feedforward and feedback, I have added another control decomposition, whereby one control component performs sensor-based feedback, modifying a gait based upon sensor measurements. With a time-varying gait, another controller then performs gait regulation, closely monitoring the gait a robot uses, to keep the system near safe, known gaits. My thesis work seeks to provide useful solutions for each of these types of control.

Control diagram of an example gait-based behavior. At the core of the feedback behavior is a set of components of an open-loop gait.

Gait Control via Temporal Parameters

Focusing specifically on the temporal parameters of gaits, I develop control laws that modify the gait timing as a robot locomotes. The result is temporally adaptive behaviors that offer greater robustness and flexibility, when compared to traditional legged behaviors and gaits that remain fixed in their gait timing.

To add sensor-based feedback, I develop controllers that try to balance foot forces while climbing. I incorporate gait regulation with specifically designed potential functions whose minima are associated with gait limit cycles.

RiSE climbing a multistory building while using the behavioral control strategies described here.

Reactive Force Controllers

Using force sensors integrated into the distal ends of a robot's legs, I have designed task controllers that, using measured ground reaction forces, compute gait modifications to correct for errors in force. The result is that the robot may move away from the current gait timing, but does so to adjust forces to locomote better.

One controller looks specifically at traction forces, those aligned in the fore-aft direction, and in line with the motion a foot makes while on the ground. Speeding or slowing down legs, as they pass through stance, results in loading or unloading forces at the compliances of the robot's feet. In this way, it is possible to balance traction forces by modifying gait timing parameters.

A second controller activates a pawing reflex, whereby, when the robot notices that a foot has slipped off the wall, the foot will reattempt to attach. Compared to force balancing, which modifies gait timing continuously, this reflex adds discontinuous jumps, as a foot in stance suddenly must recirculate to reattach.

RiSE activating a pawing reflex when detecting failed attachment to a climbing surface.

Relevant publications: [INPREP2008] [JFR2008] [RSS2006] [SPIE2006]

Gait Regulation via Toroidal Potential Functions

Given the possibility that these reactive task controllers, such as force balancing and pawing, will move the robot away from its preferred gait, I have built CPG-based dynamical systems that converge to preferred gait timings. The design of these control systems, in the form of potential functions on complex spaces, is a critical portion of my thesis research.

The timing of legs in a gait is topologically equivalent to a high-dimensional torus, on which I have noted the existence of obstacles sets, corresponding to incorrect groups of legs recirculating together, thus losing static stability. Using these obstacles as a basis, I have built control laws that reshape the torus in order to introduce stable limit cycles at desired gaits, yet converge to these gaits while attempting to avoid the obstacles specified. Furthermore, by guaranteeing certain convergence properties, I am able to use these systems to perform gait switching, allowing the robot to actively change amongst gaits while locomoting. The algorithms and control laws that I have developed for gait regulation run in real-time, owing to use of a cell decomposition of the torus, paired with a simple hybrid control approach, to guarantee certain convergence properties.

Using gait regulation, RiSE is able to actively switch amongst gaits, while continuing to balance foot forces as it locomotes up a challenging surface.

Relevant publications: [INPREP2008] [RSS2006] [SPIE2006]

The RiSE Research Platform

The RiSE robot is a bio-inspired hexapod designed to climb on a variety of vertical surfaces as well as to demonstrate horizontal mobility. With a total of twelve actuated degrees of freedom, two joints per leg, the robot is severely underactuated and must rely upon tuned compliances, built into its leg and foot structures, to be able to properly hold onto climbing surfaces. RiSE is computationally and power autonomous, running real-time control software on a 266 MHz processor, with extended run-times allowed by its on-board batteries.

The RiSE team consists of members from the University of Pennsylvania, Stanford University, U.C. Berkeley (lab 1, lab 2), Lewis & Clark University, and Boston Dynamics, Inc., in addition to Carnegie Mellon.

Serving as the RiSE team's chief behavioral engineer for the last several years, I have built and refined control software that runs in real-time on RiSE's on-board CPU, processing sensor information and activating modular pieces of software as required by the given locomotion task.

Relevant publications: [JFR2008] [SPIE2006]

Gait Transitions

With tuned gaits operating open-loop without the presence of any sensors, some of my early work focused on developing advanced mobility using only open-loop gaits. To create pathways between gaits, I have developed several algorithms to path plan gait transitions, means of speeding up or slowing down recirculating legs in order to match another gait's timing. These methods allow us to pre-plan gait transitions, creating a mobility behavioral suite, available to the robot operator. Different gaits allow the operator to change robot locomotion style and speed, while specifically designed turning gaits, as well as forward and reverse gaits, add to the overall mobility. Gait transitions provide a means of switching between all of these possible gaits.

RiSE, without using any sensors to climb, executes a series of gait transitions, switching locomotion strategies, as requested by a human operator.

This work has also been applied to the RHex robot, to mount and climb a set of stairs and transition between walking and stair climbing gaits.

RHex, upon simply detecting the first stair, executes a gait transition that allows the robot to walk directly onto the stairs and slowly switch to the stair climbing gait.

Relevant publications: [ICRA2006]

Publications

Journals

[INPREP2008]

(1, 2) G. C. Haynes, A. A. Rizzi, and D. E. Koditschek. Topology-Based Navigation of Phase Space: An Application to Legged Robot Gaits. (in preparation)

[JFR2008]

(1, 2) M. Spenko, G. C. Haynes, J. A. Saunders, M. R. Cutkosky, A. A. Rizzi, R. J Full, and D. E. Koditschek. Biologically Inspired Climbing with a Hexapedal Robot. Journal of Field Robotics (to appear) PDF Preprint (password protected, e-mail for password)

Conference Proceedings

[RSS2006]

(1, 2) G. C. Haynes and A. A. Rizzi. Gait Regulation and Feedback on a Robotic Climbing Hexapod. Proceedings of Robotics: Science and Systems (August 2006), pp. 97-104. PDF

[SPIE2006]

(1, 2, 3) A. A. Rizzi, G. C. Haynes, D. E. Koditschek, and R. J. Full. Gait Generation and Control in a Climbing Hexapod Robot. SPIE Unmanned Systems Technology VIII (June 2006)

[ICRA2006]

G. C. Haynes and A. A. Rizzi. Gaits and Gait Transitions for Legged Robots. Proceedings of the IEEE International Conference on Robotics and Automation (May 2006). PDF

Galen Clark Haynes