Collision Free Path Planning for Multi-DoF Manipulators

Abstract

Path planning is a very important issue in robotics. It has been widely studied for the last decades. This subject has gathered three interesting fields that were quite different in the past. These fields are robotics, artificial intelligence and control. The general problem of path planning consists of searching a collision free trajectory that drives a robot from an initial location (position and orientation of the end effector) to a goal location. This problem is very wide and it has many variants such as planning for mobile robots, planning for multiple robots, planning for closed kinematic chains and planning under differential constraints. It includes also time varying problems and molecular modeling, see (LaValle, 2006) for a complete review. In this study we focus on the case of multi-Degrees of Freedom (DoF) serial manipulators. The first works on serial manipulators path planning began in the seventies with Udupa (Udupa, 1977), then with Lozano-Perez and Wesley (LozanoPerez & Wesley, 1979) who proposed solving the problem using the robot's configuration space (CSpace). Since then, most of path planning important works have been carried out in the CSpace. There are two kinds of path planning methods: Global methods and Local methods. Global methods (Paden et al., 1989; Lengyel et al., 1990; Kondo, 1991) generally act in two stages. The first stage, which is usually done off-line, consists of making a representation of the free configuration space (CSFree). There are many ways proposed for that: the octree, the Voronoi diagram, the grid discretization and probabilistic roadmaps. For each chosen representation, an adapted method is used in order to construct the CSFree, see (Tournassoud, 1992; LaValle, 2006). The representation built in the first stage is used in the second one to find the path. This is not very complicated since the CSFree is known in advance. Global methods give a good result when the number of degrees of freedom (DoF) is low, but difficulties appear when the number of DoF increases. Moreover, these methods are not suitable for dynamic environments, since the CSFree must be recomputed as the environment changes. Local methods are suitable for robots with a high number of DoF and thus they are used in real-time applications. The