Decentralized collision avoidance for large teams of robots

Collision avoidance for large teams of robots in crowded environments is a challenging problem. If the environment is dynamic and the number of robots can vary during operation, centralized approaches are not suitable, since they cannot be recomputed online. This paper proposes a decentralized algorithm which allows the robots to navigate reactively to their goals. The algorithm does not require a high computational load and scales with the number of robots. The paper focuses on robots with differential drive, and in that case, the space used for each robot when avoiding collisions is optimized, so crowded environments can be tackled. In addition, realistic controllers are proposed to regulate the speed of robots. Simulations with different configurations are shown, including a complex experiment with 100 robots trying to go through the same point.