An Artificial Life Approach for the Synthesis of Autonomous Agents

This paper describes an evolutionary process producing dy-namical neural networks used as \brains" for autonomous agents. The main concepts used : genetic algorithms, morphogenesis process, artiicial neural networks and artiicial metabolism, illustrate our conviction that some fundamental principles of nature may help to design processes from which emerge artiicial autonomous agents. The evolutionary process presented here is applied to a simulated autonomous robot. The resulting neural networks are then embedded on a real mobile robot. We emphasize the role of the artiicial metabolism and the role of the environment which appear to be the motors of evolution. The rst results observed are encouraging and motivate a deeper investigation of this research area.