Walking pattern based on simulated annealing for biped robotics

Abstract

Biped robotics has attracted many research efforts for several decades. There are many researches on stability issues of biped robots. Among these, how to control the walking patterns is especially important in biped robot walking. This paper proposes a walking pattern control methodology inspired from biped animals. These animals tend to lower down the mass center to some specific position at first, and then rise up from the position to regular height of standing when they're walking with large steps. Inspired from this behavior, we first use an inverse pendulum model to interpret it, and then design a walking pattern used in biped robot walking to realize human-like natural walking. From our experiment, we find out that the lowest position of the body mass center do have a large influence on the time of performing a full walking pattern. More seriously, a walking pattern with a bad lowest position may result in the falling down of the biped robot. In order to maintain the stability and precisely control the execution time of the walking pattern, Simulated Annealing is used in the machine learning. A 2D biped robot walking simulation system is established to realize this method. We demonstrate this algorithm in this simulation system from a fall down situation to reaching the desired step time precisely. The system can be fully controlled through this proposed approach.