Dynamic neural learning for obstacle avoidance of humanoid robot performing cooperative tasks

How to avoid the collision between a humanoid robot's two manipulators is an important problem when two arms cooperate to perform the end task. In this paper, a dynamic neural learning for obstacle avoidance (DNLOA) scheme of a humanoid robot performing cooperative tasks is proposed. In this scheme, the trajectory tracking and mutual collision avoidance of the dual robot arms are converted to a quadratic programming (QP) framework, and the end-effector's trajectory tracking, mutual collision avoidance and joint angle range are described as equality constraints or inequality constraints. Then the QP framework is solved by a simplified recurrent neural network (S-RNN). The framework is applied to the trajectory tracking tasks of drawing double crossed circles and writing Chinese characters with the humanoid robot's dual manipulators. With the proposed DNLOA scheme, the tasks are done precisely without dual-arm mutual collision. Compared the scheme we proposed with the minimize velocity norm scheme (MVN) existed, experiments show that the proposed DNLOA scheme for humanoid robot is effective, accurate and practical.