I. Introduction
Multi-robot collision avoidance has attracted considerable attention from researchers in recent years, as it finds extensive applications in scenarios such as autonomous driving, intelligent warehouse robotics, and crowd simulation [1], [2], [3], [4], [5], [6]. At its core, the robot avoids collisions in environments with multiple robots, including both static and dynamic obstacles [4], [7], [8], [9]. The existing multi-robot navigation methods are generally classified into centralized and decentralized approaches based on whether they rely entirely on a central server [8]. A central server is used in a centralized approach to plan the movement of each robot based on the information sensed by all robots.This process of planning the robot’s path first does not consider collisions, and then uses a scheduling scheme to avoid collisions at potential conflict locations [10]. However, the increasing number of robots results in a substantial rise in computational demands, leading to higher resource requirements and computation time costs, along with time delays in the control signals exchanged between each robot and the central server.