YUN Chao1, LIU Gang1, WANG Gang1, YANG Xue bing2

 

(1. School of Mechanical Engineering and Automation, Beihang University, Beijing 100191, China;

 

2. Wise Welding Technology & Engineering Co. Ltd., Beijing 100076, China)

 

 

Abstract: Aiming at the flexibility of the redundant manipulator, a novel obstacle avoidance strategy based on the radial basis function (RBF) neural networks and the quadratic programming method was proposed. An obstacle avoidance model was established to describe the constraints of the manipulators such as obstacles in the workspace and the joint angle limits. Meantime, by updating the output weights, joint angles can be obtained to form an optimized configuration. The approach was proved to be feasible based on the Lyapunov stability theory. A preselected vital links method and a fuzzy look up table of the initial configuration weights obtained from the offline training algorithm can both improve the convergence rate to a great extent, which make it possible for manipulators to avoid obstacles in real time. The simulations were carried out on a new type 7 degrees of freedom robot manipulator and a comparison was made with the classical Jacobian′s pseudo inverse approach. The results indicate that the proposed approach is suitable for obtaining accurate configurations of the manipulator while avoiding the obstacles and joint angle limits with a high efficiency.

 

Key words: RBF neural networks; quadratic programming; redundant manipulators; obstacle avoidance