Simulation of a three-links planar redundant manipulator for obstacle avoidance using neural networks

This paper presents a method of solving the inverse kinematics of a three-links planar redundant manipulator to avoid obstacle and to reach a destination point with the implementation of neural networks. An algorithm of the pseudoinverse method with gradient projection technique is first used to determine the trajectory of the manipulator. The adjustable gains in the algorithm are assumed to increase to a constant value from zero in a step when the obstacle avoidance point reaches within a predefined distance with the obstacle. This causes a discontinuity in the trajectory. The results are edited and used as the input for the training of radial basis function and back-propagation neural networks in order to solve the mentioned problem. The outcomes of the two types of neural networks are compared and the results show that both improved the original trajectory while the back propagation neural network produces a `smoother' path and more uniform rate of change of joint velocities compared to that of radial basis function neural network. The complexity of the pseudo-inverse was highly reduced using the proposed method.