Designing Path for Robot Arm Extensions Series with the Aim of Avoiding Obstruction with Recurring Neural Network

In this paper, recurrent neural network is used for path planning in the joint space of the robot with obstacle in the workspace of the robot. To design the neural network, first a performance index has been defined as sum of square of error tracking of final executor. Then, obstacle avoidance scheme is presented based on its space coordinate and its minimum distance between the obstacle and each of robot links and proper inequality equations have been derived which describe the qualification of obstacle avoidance. Moreover, nonlinear optimization problem with nonlinear constraint functions has been derived by considering the joint physical limits. In order to design the neural network, equivalent problem of projection theorem has been converted to quadratic programming by using Kuhn-Tucker 1optimality conditions. Based on the projection theorem, model of recurrent neural network has been determined which is as a first order differential equation. Simulation results show the good performance of the proposed method by applying the proposed algorithm on the seven-degree freedom PA-10 robot.