Planning safe and efficient movement of a manipulator in an environment with static obstacles
Devanshi Suri, Sumanta Guha, I. Suzuki · 1999
A powerful new method, randomized preprocessing for motion planning, has emerged recently with great success in robotics. Its success can be attributed to the fact that it works very well for robots whose configuration space is high-dimensional. In this thesis, we first enhance the randomized preprocessing approach by using a cell division based heuristic which generates configurations in the preprocessing phase by controlled and iterative splitting of the configuration space. Next, a new approach to randomized preprocessing, termed ray shooting is developed where we randomly generate rays, instead of isolated configurations in the configuration space. Our experiments indicate that for certain robots, both proposed approaches offer significant performance improvements over the naive randomized preprocessing where configurations are generated randomly in the free configuration space. Addressed next in this dissertation is a novel two-phase approach to planning safe motion of a robot—the motion of a robot is considered safe if the robot stays as far away as possible from the nearest obstacle. First, in the preprocessing phase, an approximation of the medial axis of the free configuration space is constructed by either (i) probing the full configuration space along a grid or by (ii) probing the full configuration space randomly. This is followed by a careful selection of a subset of configurations that are likely to lie near the medial axis. In the subsequent path planning phase, a given source-goal configuration pair is first connected to the medial axis by an inexpensive local planner and then a motion is planned as a path within the medial axis. Experiments with 2and 3-DOF robot manipulators show that the proposed approach generates longer, but significantly safer motion, when compared to the naive planning approach that always finds a shortest motion. In addition, once preprocessing is complete the safe approach spends much less time planning motion than the naive approach. The safe approach that has been proposed herein can be easily extended to robots with many DOF. Lastly, we present a novel topology-based approach for motion planning. In this approach, we explicitly use the geometry of the workspace to identify and generate a set of configurations that form a “good cover” of the entire configuration space. The configurations in the cover are from distinct homotopy classes of paths in the workspace. We run simulation experiments for a 7-DOF robot manipulator that indicates that a topology-based approach, informed by the geometry of the workspace, performs as well as the approach based on “blind” random probing of the workspace. It, however, has the added advantages of simplicity, reduction in size of the configuration graph and stability under small perturbations of the workspace.