Given a swarm of limited-capability robots, we seek to automatically discover the set of possible emergent behaviors. Prior approaches to behavior Robot swarms utilize many independent limited-sensing agents to produce complex emergent behaviors without requiring centralized control. However, relatively little research has explored how individual agents can form local beliefs of swarm-level behaviors based solely on their local perception. Such capability would greatly improve an agent’s ability to detect faults or behavior changes in decentralized swarms. In this paper, we introduce Synchronized Swarm Behavior Classification (SyncSBC), which combines improvements in machine learning and distributed consensus to classify collective swarm behavior and synchronize swarm decision-making in an entirely decentralized manner. Through a series of simulated experiments, we show that SyncSBC achieves high classification accuracy and low synchronization delay, making it a suitable paradigm for deployable swarm robotics. Finally, we use SyncSBC to demonstrate two promising swarm applications on real robots where we show that swarms utilizing SyncSBC can accurately identify anomalies in robot behavior and autonomously coordinate collective changes in swarm behavior.

Robot Hardware: The differential drive HeRo+ robot is an extension of the original open source HeRo robot that has an upgraded distance-sensing ToF (Time-of-Flight) sensor module that is used to detect other robots in its line-of-sight. We use 8 of these robots in our experiments. Each robot also contains 8 IR sensors placed around its circumference for close-range detection. However, we would not utilize this capability in our work, leaving it as a potential direction for future research in incorporating this sensor data for behavior predictions.

Robot Communication: HeRo+ robots operate through local sensing and interactions, while exchanging information with their nearby peers following a decentralized paradigm. For practical testing and coordinated management during deployment, all robots interface with a central ROS server that facilitates orchestration while preserving decentralized swarm operation. We use the centralized server to enable full swarm emergency stop, synchronized start, and wireless controller updates. Inter-robot communication remains strictly local, where each robot shares data only with neighbors within a radius r_c. Neighbor Detection is achieved using AprilTags present on each robot, which are tracked via a top-down camera to determine relative proximity and communication eligibility.