Biped gait optimization using estimation of distribution algorithm

This paper proposes a new biped gait optimization method based on estimation of distribution algorithm (EDA). It is able to explicitly extract global statistical information from the selected solutions and build a posterior probability distribution model of promising solutions based on the extracted information. Biped gait for a nine-link robot is firstly formulated as a multiobjective optimization problem with consideration of multiconstraints including balance and torque. Optimization parameters are angles at transition poses. Instead of searching the joint space directly, EDA is applied to estimate the probability distribution of each joint degree. By this means, inherent mapping relationship between joint coordinates and cost function can be described in term of probability density. Compared to common intelligent learning method, the proposed optimization method can formulate a proper and feasible combination of impulses by tuning less parameters and visiting less states. The effectiveness of the proposed EDA based biped gait optimization method has been tested on a soccer-playing humanoid robot named Robo-Erectus. Experiment results demonstrate that the learned trajectory makes a good balance between stability and energy cost in short learning epochs