On the Effects of Collision Avoidance on Emergent Swarm Behavior

Abstract

Swarms of autonomous agents, through their decentralized and robust nature, show great promise as a future solution to the myriad missions of business, military, and humanitarian relief. Swarms can be useful purely as a theoretical abstraction or in simulation, but in many applications the swarm needs to be deployed on actual hardware platforms. The diverse nature of mission sets creates the need for a variety of hardware platforms, each with their own capabilities and limitations, for instance with sensing, actuation, communications, environmental disturbances, and structural robustness. In particular, the structural robustness of the platform, or lack thereof, seems to have a great effect on the viability of swarming behaviors where collisions might be an issue. Certain swarm behaviors have been demonstrated on platforms where collisions between agents are harmless, but on many platforms collisions are prohibited since they would damage the agents involved. The available literature typically assumes that collisions can be avoided by adding a collision avoidance algorithm on top of an existing swarm behavior. Through an illustrative example in our experience replicating a particular behavior, we show that this can be difficult to achieve since the swarm behavior can be disrupted by the collision avoidance. Furthermore, if collisions cause irreparable damage to the agents involved, we show that weakening the collision avoidance can also disrupt the intended swarm behavior, since destroyed agents are no longer able to interact with the rest of the swarm and their sudden disappearance dramatically alters the behavior of nearby agents. We introduce metrics quantifying the level of disruption in our swarm behavior and propose a technique that is able to assist in tuning the collision avoidance algorithm such that the goal behavior is achieved as best as possible while collisions are avoided. We validate our results through simulation.