Robot motion planning by a hierarchical search on a modified discretized configuration space

Abstract

or map the set of feasible motions onto a network of In this paper, a hierarchical A* -based method for motion planning of robot manipulators is presented. The algorithm's advantage is its simplicity. The algorithm is also shown to be resolution complete.The algorithm searches the robot's configuration space with many different "step sizes" at the same time. The step size denotes the distance between the nodes A' expands during the search. When a path candidate goes far from the obstacles, a big step size is used. When it goes near the obstacle surfaces, a smaller step size is used.Two simulated robot workcells are provided for experimental study. The motion planning of 5 and 6-degrees-of-freedom industrial robots appears to be reasonably fast. P IntroductionThis paper studies a point-to-point path planning problem for robot manipulators. Currently the most effective path planners utilize hierarchy in searching. They can be roughly categorized cell decompositions and roadmap or skeleton methods [11] and [8].Cell decomposition approaches are based on decomposing (either exactly or approximateiy) the set of free configurations into simple non-overlapping regions called cells, see e.g. [2], [9], [ll], [3] and [l]. The adjacency of these cells is represented in a connectivity graph that is searched for a path. If a cell is not free then it is subdivided into some smaller cells. If they are not free then the subdivision process is further continued hierarchically.Cell decomposition approaches use sophisticated algorithms for constructing adjacent free cells. To check whether a cell is free needs to model swept volumes of the manipulator links [ I l l or to calculate distances between the robot and the obstacles [6], 181. Roadmap or skeleton approaches attempt to retract one-dimensional lines, called the roadmap, skeleton, or subgoal network, see e.g. [4], [ll], [8], [lo]. The nodes of the roadmap graph or subgoals are generated by heuristic rules guiding the intelligent sampling of the search space. The subgoal generation can be done by preprocessing the search space or on-line, during the path planning. This stage can be called global planning. Local planning refers to finding a collision free path from one subgoal to another.In skeleton approaches, the local planner usually do not know how difficult it is to find a path from one subgoal to another. Therefore it has to decide on when to stop searching and to choose another subgoal. In other words, the path planners have to allocate the computational resources between the global and the local planners well to be effective. This may need sophisticated algorithms.The points mentioned above do not claim that these path planners are not effective. However, they may have quite complicated algorithmic structures.In this paper, we propose a simple path planning algorithm that take into account hierarchy in search. However, it avoids solving the complicated cell decomposition problem. It uses onedimensional lines as the roadmap methods do. On the other hand, it does not gen...