EFFICIENT GAIT LEARNING IN SIMULATION: CROSSING THE REALITY GAP BY EVOLUTIONARY MODEL IDENTIFICATION

This paper presents concepts and selected results concerning a simulation-based approach to gait development for a hexapod walking robot. The gaits should enable the fastest movement that is possible with the given robot kinematics and dynamics on flat terrain. We focus on the role of an appropriate simulation model of the walking robot and the reality gap problem, which must be solved to safely use the gaits developed in simulation on the real machine. We show that biologically inspired optimization methods can solve both the problems of gait learning and simulated robot model parametrization by efficiently finding solutions in complex, multidimensional search spaces.