Inequality-based Manipulator-Obstacle Avoidance Using the LVI-based Primal-dual Neural Network

An important issue in the motion planning and control of redundant manipulators is the online obstacle-avoidance. This paper presents the algorithmic and computational aspects of inequality-based criteria/formulations for obstacle avoidance of PA10 robot arm. The formulations are unified as a quadratic- programming (QP) problem. In addition to handling environmental obstacles, this unified QP problem formulation could avoid joint physical limits as well as optimize various performance indices. Motivated by the online solution to such robotic optimization problems, four QP online algorithms/solvers are reviewed, especially the LVI-based primal-dual neural network. The inequality-based QP formulation and its solution for obstacle avoidance are substantiated by simulation results. This simulation also shows that joint-acceleration information could be generated online by using dynamic QP solvers for torque control even in the velocity-level redundancy resolution.