Physical Interactions Segregate Robot Swarms

Abstract

It is challenging to design rules of local interactions for global patterns in self-organized robot swarm systems, especially in robots with limited capacities of sensing, communication, and computing. Many natural swarms, such as animal swarms and cell collectives, generate complex global patterns through physical interactions. Mimicking the differential adhesion hypothesis in developmental biology, we developed a self-organizing method to segregate swarms of robots with different adhesion levels through physical interactions. Rather than using the electric sensing capability of robots, this method segregates different types of robots using their collisions as a sensor. In this method, the robots make decisions via competition between adhesion among robots and driving forces of robots rather than the electric computing capability of robots. Our experimental and simulation results illustrated that our method is effective. We further experimentally demonstrated the robustness of this method to the failures of robots. This physical-interaction-enabled segregation method exemplifies the concept of physical intelligence in swarm robotics.