A biological perspective on autonomous agent design

Abstract

The inability of current “classical” AI systems to handle unconstrained interaction with the real world has recently lead to a search for new control architectures for autonomous agents. We argue that simpler natural animals already exhibit most of the properties required by an autonomous agent, and suggest that designers of autonomous agents should draw directly upon the neural basis of behavior in these animals. The relevant behavioral and neurobiological literature is briefly reviewed. An artificial nervous system for controlling the behavior of a simulated insect is then developed. The design of this artificial insect is based in part upon specific behaviors and neural circuits from several natural animals. The insect exhibits a number of characteristics which are remarkably reminiscent of natural animal behavior.