Multi objective optimization of humanoid robot arm motion for obstacle avoidance

Abstract

Picking and placing objects on the table for an assistive humanoid robot requires good coordination and motion strategies. Obstacle avoidance is one of the main factor needs to be considered. In this paper, the arm motion generation for obstacle avoidance is formulated as an optimization problem. Multi-Objective Genetic Algorithm (MOGA) is utilized to generate the neural controller, optimizing three objective functions namely minimum execution time, minimum gripper distance and minimum arm acceleration. The main advantage of the proposed method is that in a single run of MOGA, multiple optimized neural controllers are generated. A wide range of initial and goal position can be achieved utilizing the same generated neural controller. The performance of the generated humanoid robot arm motion yields good results in simulation and experimental environments.