Evolution and analysis of dynamical neural networks for agents integrating vision, locomotion, and short-term memory
John Christopher Gallagher, Randall D. Beer · 1999
The use of evolutionary approaches to create dynamical "nervous systems" for autonomous agents is becoming increasingly widespread. In previous work, we have successfully applied this approach to chemotaxis, walking, learning, and such minimally cognitive behavior as visually-guided orientation, object discrimination and pointing. In this paper, we extend this approach to the integration of visually-guided orientation and walking and to an object orientation task that requires short-term memory. In addition, we examine the neural dynamics underlying the operation of some of these evolved agents. 1 INTRODUCTION The potential of evolutionary approaches for autonomous agents is widely recognized (Beer & Gallagher, 1992; Brooks, 1992; Cliff et al., 1993; Nolfi et al., 1994). While much of this work has emphasized relatively low-level motor behavior, there is a growing interest in applying evolutionary approaches to more sophisticated kinds of behavior, such as predator/prey interactions, ...