Centroid Method of Cluster Control for a Group of Mobile Robots for Collision Avoidance

Abstract

Real-time coordination of large-scale mobile robot swarms remains a challenging problem, particularly in dynamic and obstacle-rich environments. The primary limitation arises from the computational complexity of peragent path planning, which becomes intractable as the number of robots increases. This paper proposes a centroid-based cluster control framework that integrates adaptive k-means clustering with graph-based global routing. The method reduces computational complexity by performing global path planning only for cluster centroids, rather than for individual robots. Local control laws combine attractive forces toward cluster centroids with repulsive inter-agent interactions to ensure collision avoidance and compliance with velocity constraints. The approach is validated through simulations with up to 200 robots at a control rate of 20 Hz. Experimental results demonstrate up to 92% reduction in computation time compared to per-agent A* planning, while path elongation remains below 9%. The proposed framework provides a scalable and efficient solution for decentralized multi-robot coordination in real-time applications.