Evolving morphologies and gaits of physically realistic simulated robots

Abstract

This paper describes our research and experiments with autonomous robots, in which were used genetic algorithms to evolve stable gaits of simulated legged robots in a physically based simulation environment. In our approach, gaits are defined using two different methods: a finite state machine based on the joint angles of the robot legs; and an Elman's recurrent neural network. The parameters for both methods are optimized using genetic algorithms, and the proposed model also allows the evolution of the robot body morphology. Several experiments are described, and the obtained results show that it is possible to generate stable gaits and efficient morphologies using machine learning techniques.