MCCA: A Decentralized Method for Collision and Deadlock Avoidance With Nonholonomic Robots

Navigation in dense and narrow environments with multiple robots is a standing challenge since deadlock is prone to occur. In this letter we present masked cooperative collision avoidance (MCCA), a fully decentralized method to avoid both collision and deadlock effectively. The concept of masked velocity is introduced, which is an implicit state of each robot and acts as an intention of avoiding deadlock. Robots are prioritized by a decentralized mechanism and masked velocities of robots with different priorities propagate among robots, promoting fluent and efficient deadlock avoiding behaviors in a local and collective manner. The solving process is reduced to a quadratic programming problem. Nonholonomic constraints are taken into account. We conduct extensive experiments in both simulation and real-world application, and the results verify the effectiveness of our method.