Hierarchical Genetic Path Planner for Highly Redundant Manipulators

This paper is concerned with path planning of highly redundant manipulators, which are particularly well suited for flexible manufacturing. We envision that highly redundant manipulators will be composed of a set of standard modules. In order to find the optimal configuration of modules, one has to perform a discrete search on possible assemblies of modules combined with a continuous search of each assembly to verify the feasibility for completing the given task. This requires efficient planning techniques in the workspace since configuration space conversion is too demanding when considering many possible assemblies. For this reason, we propose a hierarchical path planner consisting of a global path planner (GPP) and a local motion planner (LMP). The GPP generates multiple collision-free paths in the workspace by using the free space structuring algorithm MAKLINK. The GPP passes the paths to the LMP, which determines their feasibility. If one path is found to be unfeasible, another one is chosen automatically. Along the path the LMP selects a number of points as intermediate goals and for each goal it solves the inverse kinematics problem. The LMP is based on a genetic algorithm that reuses the population from one intermediate goal to the next. This nonrandom initial population technique significantly reduces the joint angle variations, thereby increasing the probability that a collision-free path exists in the configuration space.