Shaping Collective Behaviors in Swarm Robotics Through Probabilistic Motion Decision-Making

Swarm robotics exhibits scalable and adaptive collective behaviors, providing an effective solution for complex tasks in real-world applications. However, reliance on velocity and global positioning information of neighbors limits the practical deployment of swarm robots. In this letter, we propose a sensorimotor-based swarm model that directly maps first-person visual perception to motion decisions through probabilistic decision-making. Based on numerical simulations, we find the emergence of flocking, milling, and swarming behaviors without explicit velocity alignment and positional interactions. In addition, we investigate the effectiveness of the proposed swarm model under non-omniscient perception. Moreover, we show that probabilistic motion decision-making enhances the resilience of group coordination under sensory noise. Finally, we achieve flocking, milling and swarming behaviors in a swarm of 50 real robots under motion noise disturbance and simulated visual constraints, highlighting the potential of the proposed swarm model in real-world tasks.