Multiobjective optimization of a quadruped robot gait

This paper presents the methodology used for finding the optimal set of foot trajectories for a quadruped robot using multiobjective genetic algorithm optimization. The optimization evaluates the energy per distance and average speed criteria on a robot simulation model. Robot locomotion is achieved by open-loop execution of foot trajectories generated in the local leg coordinate system. Foot trajectory is formulated as a sum of harmonics which enabled great flexibility in determining the final trajectory shape. A multiobjective optimization is introduced to tune the foot trajectory parameters in order to achieve energy optimal and fast robot locomotion. The obtained Pareto frontier showed that the bound gait is optimal for lower speeds while the trot gait enabled the robot to reach its maximum speed. The paper identifies the correlation between the stride frequency and robot speed for each identified gait laying on the Pareto frontier. Finally we discuss the trajectory shape of solutions obtained using multiobjective optimization.