A survey on Velocity Obstacle paradigm

Collision avoidance for multi-agent systems in dynamic and possibly crowded environment is a field of research whose popularity has increased during the last three decades. The approaches to solve the problem belong to two categories according to where the computation is done: centralized and decentralized (i.e. at the agent level). The principal decentralized collision avoidance methods currently available in literature are Artificial Potential Fields, Velocity Obstacle, Social Force, Dynamic Window and Model Predictive Controller. The aim of this paper is to offer a complete and comprehensive review of the most popular decentralized Velocity Obstacle based collision avoidance schemes. A taxonomy will describe the evolution of the field of research back to the original paper in the early 1990s.