Local motion planning for manipulators based on shrinking and growing geometry models

Abstract

A new approach to motion planning for manipulators is presented. A collision-free path is initially obtained by shrinking the robot's geometry model, and this path is modified to remain collision-free while the robot is re-inflated. For this purpose, a measure based on the factor needed to shrink the robot's geometry model in a certain configuration to get it free of collisions is defined. This measure serves as a potential and is used to modify whole trajectories. The planning information is evaluated locally, no preprocessing or global potential diffusion is required. To overcome local minima, the planner is embedded in a random global exploration algorithm. Simulation results demonstrate the power of this approach.