Kinematic control of a seven DOF robot manipulator with joint limits and obstacle avoidance using neural networks

In this paper, a numerical method based on neural network is presented to solve inverse kinematics problem of redundant manipulators subject to joint angle limits and obstacles in the workspace of the robot. The proposed method is performed in real time, where radial-basis function neural network is used to obtain joint angles of the robot. In order to satisfy constrains, a method called Nonlinear Quadratic Programming (NQP) is applied to update NN´s weights. Moreover, it will be shown that if the Kuhn-Tucker conditions are satisfied, then convergence of NN´s weights is guaranteed. Since the process is performed on-line, the computational time of obtaining the inverse kinematics solution must be suitable for real-time applications such as control of the robot manipulators. Moreover, since the convergence rate of the problem depends on the initial weights of the neural network, several initial weights are used relative to suitable factors such as feasibility of the solution and vicinity of the desired point. Simulations are carried out on the PA-10 redundant manipulator to show effectiveness of the proposed algorithm.