Institute for Mathematics and Its Applications
Randall D. Beer, Case Western Reserve University
Animals are remarkably effective and robust in complex real-world environments. For this reason, they are increasingly serving as a source of inspiration for autonomous robots. Unfortunately, many of the very same properties of biological nervous systems that are responsible for these impressive behavioral capabilites can also make them very difficult to understand and design. This talk will survey a variety of ongoing projects attempting to grapple with some of these difficulties. First, I will briefly describe a series of legged robots whose control is based on Pearson's flexor-burst generator model of cockroach walking and Cruse's model of stick insect walking. Next, I will briefly describe the use of genetic algorithms to evolve dynamical neural controllers for legged locomotion. One of the interesting results of these experiments is that chain reflex, central or "mixed" pattern generation architectures consistently evolve depending on the availability of sensory feedback during evolution. Finally, I will present an analysis of the general principles of operation of a large population of evolved single-leg central pattern generators, as well as a preliminary analysis of phase-locking in Cruse's model of stick insect walking.