Control of a robot leg with an adaptive aVLSI CPG chip
Introduction
A central pattern generator (CPG) is most generally defined as a biological circuit capable of autonomously generating sustained oscillations [2], [7]. In particular, locomotor CPGs produce a rhythmic pattern of neuronal discharge that can drive muscles in a fashion similar to that seen during normal locomotion. Locomotor CPGs are autonomous in the sense that they can operate without input from higher centers or from sensors. Under normal conditions, however, they make extensive use of sensory feedback from the muscles and skin, as well as descending input [3].
In the early 1980s Cohen and colleagues [4] analyzed a CPG model using a system of phase-coupled oscillators. Lewis et al. [9], [12], [10] developed a model of the CPG called adaptive ring rules (ARR). Fundamentally, ARR is based on phase-coupled oscillators, but is highly elaborated. The elaboration includes the use of plastic output functions and phase-coupling functions, and allows modeling of plasticity and learning in CPG elements. ARR has sufficient complexity to drive robotic devices [9], [10]. In hardware, there have been several VLSI implementations of CPGs used to control robot movement [16], [14], [15]. In particular, Still and Schölkopf [16] developed a supervised-learning method for tuning a CPG based on knowledge of desired phase and duty cycle.
Here we present an adaptive VLSI chip, based on established principles of locomotor-control circuits in the nervous system, that mimics many features of a biological CPG. The circuit can successfully control a robot leg running on a treadmill. It uses real-time sensory feedback from position sensors to stabilize running rhythmicity, and a competitive mechanism to establish the value for an adaptive parameter. Unlike Still and Schölkopf [16] we do not use supervised learning, and thus view the present work as complementary to that effort.
In the presence of sensory feedback, the adaptive properties of the circuit can produce stable running even when the robot leg is “underactuated,” meaning that not all of its moving joints are actively driven. The leg can be driven stably with a hip actuator alone, allowing the knee to be passive. We show that the sensory feedback serves to entrain the CPG, such that the gait can recover from large asymmetries in the backward and forward swings of the leg. Selective “lesions” of this sensory feedback resulted in a deterioration of the gait, but the gait recovered when sensory feedback was restored. We also show that sensory feedback is critical for compensating for momentary external perturbations.
It should be noted that we have, in essence, the minimal system that can produce running behavior. This system has three essential elements: (1) A driven hip and passive knee, (2) a CPG capable of entrainment and (3) a CPG capable of output amplitude modulation. If any essential element is removed, the system will not work. The mechanisms of adjustment are simple and straight-forward. Our approach represents an extreme in the spectrum of modeling. We tend towards an absolutely minimal system, removing obfuscating detail. From this starting point we can make our model increasingly complex to incorporate additional biological details.
Section snippets
A novel CPG chip and circuits
The circuits described below were implemented on a VLSI chip which is not shown here for reasons of space. The circuits consume less than 1 μW of power and occupy less than 0.4 mm2 in area. Those interested in more details should refer to [11].
The robot leg and sensors
The robotic leg is a small (10-cm height) two-joint mechanism. For all experiments described, only the hip joint was driven, and the “knee” remained completely passive. The knee rotated on a low-friction ball bearing joint, and was prevented from hyperextension with a hard mechanical stop. The leg ran on a drum that was free to rotate under the contact forces of the leg.
The leg has three sensors on it. Two inductive (LVDT) sensors monitor the position of the knee and hip joints, and a pressure
Results
The foremost result is that the circuit adapts such that the passive knee joint has the correct dynamics to enable running. Fig. 4 shows a phase plot of the hip and knee position, and foot contact force. The bulk of the trajectory describes a tight “spinning top” shape, while the few outlying trajectories are caused by external disturbances. After a disturbance the trajectory quickly returns to its nominal orbit, implying that the system is stable.
We next established that sensory feedback was
Conclusions
In this paper we present the first experimental results of an adaptive aVLSI CPG chip controlling a robotic leg. Our work differs from previous studies in several respects. First, we allow adaptation based on sensory input. Second, our chip has short-term onboard memory devices that allow the continuous, real-time adaptation of both center-of-stride and stride amplitude. In addition, we make use of integrate-and-fire neurons for the output motor neurons. Finally, our abstraction is at a higher
M. Anthony Lewis is President and a founder of Iguana Robotics, Inc. He received a B.S. in Cybernetics from the University of California, Los Angeles, and a M.S. and a Ph.D. in Electrical Engineering from the University of Southern California in 1994 and 1996, respectively. His research areas include learning in legged locomotion, visuomotor coordination, and medical applications of biomimetic technology. He began his career in robotics at Hughes Aircraft building snake-like robot manipulators.
References (16)
- et al.
Peripheral control of the cat's step cycle II. Entrainment of the central pattern generators for locomotion by sinusoidal hip movements during fictive locomotion
Acta Physiol. Scand.
(1983) - T. G. Brown, On the nature of the fundamental activity of the nervous centres; together with an analysis of the...
- et al.
Neural Control of Rhythmic Movements in Vertebrates
(1988) - et al.
The nature of the coupling between segmental oscillators of the lamprey spinal generator for locomotiona mathematical model
J. Math. Biol.
(1982) - R.J. Full, Integration of individual leg dynamics with whole body movement in arthropod locomotion, in: R.D. Beer,...
- et al.
The simplest walking modelstability, complexity, and scaling
ASME J. Biomech. Eng.
(1998) Neurobiological bases of rhythmic motor acts in vertebrates
Science.
(1985)- F.C. Hoppensteadt, An introduction to the mathematics of neurons, Cambridge Studies in Mathematical Biology, vol. 6,...
Cited by (32)
Neuromorphic electronics for robotic perception, navigation and control: A survey
2023, Engineering Applications of Artificial IntelligenceFPGA implementation of a configurable neuromorphic CPG-based locomotion controller
2013, Neural NetworksCitation Excerpt :The circuit power consumption is 4.8 mW and its size including I/O pads is 2.2 mm2. Lewis, Hartmann, Etienne-Cummings, and Cohen (2001) present an adaptive CPG in an analog VLSI chip used to control a robot leg. All neurons in their circuit are modified integrate-and-fire neurons composed of capacitors, MOSFET (metal–oxide–semiconductor field-effect transistor) transistors and a hysteretic comparator.
Using hardware models to quantify sensory data acquisition across the rat vibrissal array
2007, Bioinspiration and BiomimeticsComputer simulation study on central pattern generator: From biology to engineering
2006, International Journal of Neural Systems
M. Anthony Lewis is President and a founder of Iguana Robotics, Inc. He received a B.S. in Cybernetics from the University of California, Los Angeles, and a M.S. and a Ph.D. in Electrical Engineering from the University of Southern California in 1994 and 1996, respectively. His research areas include learning in legged locomotion, visuomotor coordination, and medical applications of biomimetic technology. He began his career in robotics at Hughes Aircraft building snake-like robot manipulators. During graduate school, he worked for two years at the California Institute of Technology's Jet Propulsion Laboratory in the area of anthropomorphic telemanipulators, and he served two years as the director of UCLA Commotion Laboratory working in the area of collective robotics. After graduation, he joined the University of Illinois at Urbana-Champaign and worked in the area of visuomotor control and multi-robot systems.
Ralph Etienne-Cummings received his B.Sc. in Physics, 1988, from Lincoln University, Pennsylvania. He completed his M.S.E.E. and Ph.D. in electrical engineering at the University of Pennsylvania in 1991 and 1994, respectively. Currently, Dr. Etienne-Cummings is an assistant professor of electrical and computer engineering at the Johns Hopkins University, Baltimore, MD. His research interest includes mixed signal VLSI systems, computational sensors, computer vision, neuromorphic engineering, smart structures, mobile robotics and robotics-assisted surgery. He is a recipient of the NSF's CAREER and ONR's YIP Awards.
Mitra J. Hartmann received a B.S. in Applied and Engineering Physics from Cornell University, and a Ph.D. in Integrative Neuroscience from the California Institute of Technology. She is currently a post-doctoral scholar in the Center for Integrated Space Microsystems at Caltech's Jet Propulsion Laboratory. Her research interests include active sensing, neuromorphic sensory processing, sensorimotor control systems, and biomimetic robotics. She is a recipient of Caltech's Everhart Distinguished Graduate Student award.
Avis H. Cohen obtained her Ph.D. from Cornell University, and held post-doctoral positions at the Karolinska Institute in Stockholm with Sten Grillner, and at Washington University, St. Louis with Carl Rovained, developing the lamprey spinal cord preparation for the study of the central pattern generator for locomotion. She then spent 10 years at Cornell in an independent research position continuing her lamprey work, and extending it to mathematical modeling of the intersegmental coordinating system using dynamical systems models. In 1990 she moved to University of Maryland as an Associate Professor in the Biology Department where she has been ever since. She was the director of the fledgling Neuroscience and Cognitive Science Program at UMD, and is a member of the Institute for Systems Research. She is also one of the co-directors of the Telluride Summer Workshop on Neuromorphic Engineering.