An Evolutionary Based Approach for Control Programming of Humanoids

We describe the first instance of a novel approach for control programming of humanoid robots, based on evolution. To overcome some of the difficulties with evolution on real hardware, we use a physically realistic simulation of the robot. The essential idea is to evolve control programs from first principles on a simulated robot, transfer the programs to the real robot, and continue to evolve on the robot. As the key motivation for using simulators, we describe an on-line learning experiment with a humanoid robot. The Genetic Programming system is implemented as a Virtual Register Machine, with a linear genome, and steady state tournament selection. Evolution created controller programs that made the simulated robot produce forward locomotion behavior. A further application of this system, with two phases of evolution, would be to have a flexible adaptation mechanism that can react to hardware failures in the robot.