Sampling-based Motion Planning with High-Level Discrete Specifications

Motion planning has generally focused on computing a collision-free trajectory to a goal region. Enhancing the ability of robots in manipulation, automation, medicine, and other areas, however, often requires richer task specifications. Toward this goal, we study the problem of computing a collision-free trajectory that satisfies task specifications given by Finite Automata, STRIPS, Linear Temporal Logic, and other logic models. We propose to combine sampling-based motion planning with discrete planning. The search for a solution trajectory is conducted simultaneously over the continuous space of motions and the discrete space of the task specification. In this search, discrete plans guide motion planning as it extends a tree consisting of collision-free trajectories, while information gathered from motion planning is used to further improve the discrete plans. As a result of this interplay, the approach is able to selectively sample and explore those continuous regions and discrete plans that allow it to significantly advance the search for a trajectory that satisfies the task specification.