Guiding sampling-based tree search for motion planning with dynamics via probabilistic roadmap abstractions

This paper focuses on motion-planning problems for high-dimensional mobile robots with nonlinear dynamics operating in complex environments. It is motivated by a recent framework that combines sampling-based motion planning in the state space with discrete search over a workspace decomposition. Building on this line of work, the premise of this paper is that the computational efficiency can be significantly improved by tightly coupling sampling-based motion planning with probabilistic roadmap abstractions instead of workspace decompositions. Probabilistic roadmap abstractions are constructed over a low-dimensional configuration space obtained by considering relaxed and simplified representations of the robot model and its feasible motions. By capturing the connectivity of the free configuration space, roadmap abstractions provide the framework with promising suggestions of how to effectively expand the sampling-based search in the full state space. Experiments with high-dimensional robot models, nonlinear dynamics, and nonholonomic constraints show significant computational speedups over related work.