Kinematic Control and Obstacle Avoidance for Redundant Manipulators Using a Recurrent Neural Network

In this paper, a recurrent neural network called the Lagrangian network is applied for obstacle avoidance in kinematically redundant manipulators. Conventional numerical methods implemented in digital computers for obstacle avoidance redundancy resolution calculation could only compute the solution in milliseconds while neural network realized by hardware could complete the computation in microseconds, which is more desirable in real-time control of manipulators. By giving the desired end-effector velocities and the obstacle location, the neural network could generate the joint velocity vector which drives the manipulator to avoid obstacles and tracks the desired end-effector trajectory simultaneously. Simulation results show that the neural network is capable of computing the redundancy resolution for obstacle avoidance.