Aggressive Moving Obstacle Avoidance Using a Stochastic Reachable Set Based Potential Field

Identifying collision-free trajectories in environments with dynamic obstacles is a significant challenge. However, many pertinent problems occur in dynamic environments, e.g., flight coordination, satellite navigation, autonomous driving, and household robotics. Stochastic reachable (SR) sets assure collision-free trajectories with a certain likelihood in dynamic environments, but are infeasible for multiple moving obstacles as the computation scales exponentially in the number of Degrees of Freedom (DoF) of the relative robot-obstacle state space. Other methods, such as artificial potential fields (APF), roadmap-based methods, and tree-based techniques can scale well with the number of obstacles. However, these methods usually have low success rates in environments with a large number of obstacles. In this paper, we propose a method to integrate formal SR sets with ad-hoc APFs for multiple moving obstacles. The success rate of this method is 30% higher than two related methods for moving obstacle avoidance, a roadmap-based technique that uses a SR bias and an APF technique without a SR bias, reaching over 86% success in an enclosed space with 100 moving obstacles that ricochet off the walls.