Real-time collision avoidance of manipulators with multiple redundancy

Abstract

A method is presented for the control of robot manipulators with multiple redundancies, using a technique based on Laplacian potential fields. Obstacles and a goal point are modelled as points of fixed potential within a continuous, closed region. All obstacles are modelled as having the same high potential and the goal point is defined as having some lower potential. From the field calculated, a method based on gradient descent is used to control the motion of the robot joints in an optimal manner, which will automatically avoid collisions between any of the links and any of the obstacles. The proposed field representation is compared with other potential field methods and in particular, the method is shown to be appropriate for fixed base, non-point robots. Four simulated case studies are presented which illustrate the technique in two-dimensional space. The studies consider the problems of: insertion of a robot into an annulus; negotiation of a high aspect-ratio corridor; a randomly cluttered field; and a symmetrical trap.