An Update Method of Computer Simulation for Evolutionary Robotics.
Yoshiaki Katada, Kazuhiro Ohkura · IAS · 2006
One of advantages of evolutionary robotics over other approaches is its parallel population search. However, it generally takes an unrealistically long time to evaluate all candidate solutions by using a real robot. Thus, a technique of computer simulation is considered to be one of the most important topics in evolutionary robotics. Although it is quite difficult to provide a precise computer simulation model of a physical experiment, simulated robot/environment interaction dynamics must be synchronously consistent with the physical one during its evolution for reducing the gap between the simulated and physical one. In this paper, in order to overcome this problem, we propose an update method of simulated robot/environment interaction dynamics based on a statistical test during evolution, and then investigate the consistency between the performances of a robot in the simulated and physical environment. A series of experiments with a mobile robot illustrate the effectiveness of the proposed method in practice.