Neural Network Architecture to Solve Linear Complmentarity Problems: Application to the Grasping Problem

摘要

The numerical solution of Linear Complementarity Problem(LCP) by means of a heuristic technique is investigated. Complementarity problems provide a powerful framework to many applications in mechanics and engineering. Typical applications of these types of problems are unilateral contact conditions. The interaction in the gripper-object system exhibits unilateral contact conditions. The proposed neural network architecture finds almost exact solutions if it is compared with that found by Lemke algorithm. Training of the neural network is achieved by a new adaptive technique. This technique helps in the reduction of the iterations needed to reach the goal. Numerical examples that illustrate the proposed approach are presented.