Gait Optimization of Biped Robot Based on Mix-encoding Genetic Algorithm

Abstract

A seven-link biped robot model with 12 rotational DOF was chosen for gait optimization. The vector describing robot's position and pose was established, then the vector's expected locus during a regular step was modeled by the 5th order polynomials. The mathematic descriptions of geometry restriction, stabilization, energy dissipation, and impact to swaying leg from floor were analyzed respectively, and then the optimal gait was worked out with genetic algorithm mixing binary number encoding and floating point number encoding. Experimental results show that the optimal gait maximizes dynamic stabilization while it minimizes both energy dissipation and impact to swaying leg from floor.