Aggressive Collision-Inclusive Motion Planning

Abstract

Traditional drone systems typically prioritize collision avoidance as a fundamental tenet of their motion planning paradigms. Nevertheless, in some narrow scenarios, collisions might be inevitable. Notably, collisions can be strategically employed to minimize task duration and energy consumption. This article delves into the intriguing realm of actively and reasonably utilizing collisions for aggressive motion planning. To generate aggressive collision-utilized primitives, we design a sample-based front-end called improved collision-inclusive rapidly-exploring random trees, which solves linear quadratic minimum time problems. Unlike traditional derivative rapidly-exploring random trees, the generated primitives undergo deformation when collisions are detected rather than being discarded. To achieve better collision actions, a collision-utilized trajectory optimization is constructed, with the core being the design of a maneuverability constraint and a collision relaxation constraint. The corresponding constraint elimination and transcription methods are proposed to accelerate the solving process. Finally, the constrained trajectory optimization is transformed into an unconstrained optimization problem that can be solved within $\text{6}\,\text{ms}$. To determine whether to implement collisions, we propose a receding horizon decision strategy where the decision between collision-utilized and collision-free trajectories is made by minimizing the time cost within each horizon. The integrated collision-inclusive planning system enables robots to judiciously weigh the risks and benefits of collisions, allowing them to autonomously decide whether and how to engage in such interactions. Extensive experiments verify the effectiveness, capability, advancement, and robustness of our proposed method, showing that the strategic utilization of collisions can enhance aggressiveness in narrow environments, opening up a new perspective for motion planning.