CWRU Dynamics of Adaptive Behavior Research Group

Dynamics of Adaptive Behavior Research Group

Electrical Engineering and Computer Science
Case Western Reserve University

Research Areas

Support provided by grants from the National Science Foundation, the Office of Naval Research, DARPA, and the Human Frontier Science Program.

Evolution of Dynamical "Nervous Systems" for Autonomous Agents
We use evolutionary algorithms to generate continuous-time recurrent neural networks (CTRNNs) that function as "nervous systems" for controlling the behavior of autonomous agents. To date, we have evolved CTRNNs for chemotaxis, walking, sequential decision-making, reinforcement learning and simple visually-guided behaviors orientation, discrimination and pointing). The evolved walking circuits have been successfully applied to the control of an actual hexapod robot. Current work is focused on extending this approach to more sophisticated forms of visually-guided behavior and learning, to the integration of multiple behaviors, and to the control of more physically realistic peripheries.

Dynamical Analysis of Coupled Brain/Body/Environment Systems
We use the mathematical tools of dynamical systems theory to analyze the mechanisms of adaptive behavior in a variety of systems. We have studied the general dynamical behavior of small CTRNNs, and have analyzed the operation of a variety of evolved CTRNNs, especially walking circuits. These analyses have emphasized the importance of viewing behavior not merely as a product of a nervous system, but rather as arising from the ongoing interaction between a nervous system, body and environment. Currently, we are studying the operation of evolved CTRNNs for visually-guided object discrimination and analysing the properties of distributed control systems for hexapod locomotion. In addition, we are exploring the broader implications of our dynamical perspective for understanding the neural mechanisms underlying adaptive behavior and cognition.

Biologically-Inspired Robotics
Animals are remarkably well-adapted to the environments in which they must survive, and they exhibit an efficiency and flexibility of operation and a robustness to contingency and damage that we would like to emulate in our autonomous robots. In collaboration with Prof. Roger Quinn's Bio-Robotics Group in the Dept. of Mechanical and Aerospace Engineering and with Profs. Hillel Chiel and Roy Ritzmann in the Dept. of Biology, we are exploring ways in which biological control principles abstracted from studies of the neural basis of behavior in simpler animals can be applied to autonomous robots. We also believe that appropriately constrained robots can serve as physical models for testing biological hypotheses. Most of our work to date has focused on the application of studies of insect walking to the control of locomotion in legged robots.

Miscellaneous
In addition, we have been known to dabble in a variety of other areas, including models of the neural and biomechanical basis of behavior in simpler animals (Computational Neuroethology), models of biological development, and applications of AI to the generation of music by computer.

Publication List

Current Members		Recent Alumni
Dr. Randall Beer, Director Barry Drennan Chad Seys Sean Psujek Jacob Garcowski Eldan Goldenberg		Dr. Leslie Picardo Dr. Frank Dellaert Dr. Brian Yamauchi Dr. John Gallagher Eva Pierce Andy Slocum Doug Downey Gwendid van der Voort van der Kleij Boonyanit Mathayomchan Andrew Fife Dr. Phattanard Phattanasri Jeff Ames Dr. Alan Calvitti (Thesis)

beer@eecs.cwru.edu

	Evolution of Dynamical "Nervous Systems" for Autonomous Agents We use evolutionary algorithms to generate continuous-time recurrent neural networks (CTRNNs) that function as "nervous systems" for controlling the behavior of autonomous agents. To date, we have evolved CTRNNs for chemotaxis, walking, sequential decision-making, reinforcement learning and simple visually-guided behaviors orientation, discrimination and pointing). The evolved walking circuits have been successfully applied to the control of an actual hexapod robot. Current work is focused on extending this approach to more sophisticated forms of visually-guided behavior and learning, to the integration of multiple behaviors, and to the control of more physically realistic peripheries.
	Dynamical Analysis of Coupled Brain/Body/Environment Systems We use the mathematical tools of dynamical systems theory to analyze the mechanisms of adaptive behavior in a variety of systems. We have studied the general dynamical behavior of small CTRNNs, and have analyzed the operation of a variety of evolved CTRNNs, especially walking circuits. These analyses have emphasized the importance of viewing behavior not merely as a product of a nervous system, but rather as arising from the ongoing interaction between a nervous system, body and environment. Currently, we are studying the operation of evolved CTRNNs for visually-guided object discrimination and analysing the properties of distributed control systems for hexapod locomotion. In addition, we are exploring the broader implications of our dynamical perspective for understanding the neural mechanisms underlying adaptive behavior and cognition.
	Biologically-Inspired Robotics Animals are remarkably well-adapted to the environments in which they must survive, and they exhibit an efficiency and flexibility of operation and a robustness to contingency and damage that we would like to emulate in our autonomous robots. In collaboration with Prof. Roger Quinn's Bio-Robotics Group in the Dept. of Mechanical and Aerospace Engineering and with Profs. Hillel Chiel and Roy Ritzmann in the Dept. of Biology, we are exploring ways in which biological control principles abstracted from studies of the neural basis of behavior in simpler animals can be applied to autonomous robots. We also believe that appropriately constrained robots can serve as physical models for testing biological hypotheses. Most of our work to date has focused on the application of studies of insect walking to the control of locomotion in legged robots.
	Miscellaneous In addition, we have been known to dabble in a variety of other areas, including models of the neural and biomechanical basis of behavior in simpler animals (Computational Neuroethology), models of biological development, and applications of AI to the generation of music by computer.