Adaptive Anytime Motion Planning for Robust Robot Navigation in Natural Environments

Abstract

The problem of robot navigation is treated under constraints of limited perception horizon in complex, cluttered, natural environments. We propose a solution based on our previous work in fast constrained motion planning, where arbitrary mobility constraints could be satisfied while the planning problem is reduced to unconstrained heuristic search in state lattices. By trading off optimality, we improve planner run-times and increase robustness through achieving anytime planning quality, such that it becomes possible to integrate the planner within the high speed navigation framework. We show that using a planner in navigation works well and fast enough for real vehicle implementation, while it presents a number of important benefits over state-of-the-art in navigation.