Reachability and capability analysis for manipulation tasks

Abstract

An offline analysis of the reachability of a robotic arm saves time for online queries like grasp selection or path planning. Reachability data is complemented with indices that quantify the goodness of one region in space to create a capability map, which can be computed based either on forward or inverse kinematics. This paper discusses the advantages and limitations of those methods, and proposes a hybrid method to improve the generation time while guaranteeing complete exploration of the space. The correctness of the results is studied with a prediction accuracy test. To illustrate the utility of a capability map, real-time visual information is incorporated to the map to help in the selection of grasp poses, or in path planning from an initial to a final pose.