Prey Retrieval by a Swarm of Robots
Labella Thomas Halva

Abstract:

Prey retrieval, also known as foraging, is a widely used test application for Multi Robot Systems (MRS). The task consists in searching for objects spread in the environment and in bringing them to a speci c place called nest or home. Scienti c issues usually concern e cient exploration, mapping, communication among agents, task coordination and allocation, and con ict resolution. In particular, con icts prevent the performance of foraging (i.e. the number of items retrieved) from growing linearly with the number of robots. Collisions among robots or two robots that try to move an object but in di erent directions are examples of con icts. From an energetic point of view, con icts can be seen as energy that is lost during the retrieval. Several works in the literature investigate how the control system of each agent or some form of middle/long range communication can improve performance. In this work, we study a di erent approach to con ict resolution based only on information locally available to each robot. We show that it is possible to reduce con icts by adapting for each robot the probability to leave the nest according to previous successes or failures. We derive an estimator of the e ciency of the system and show that this form of adaptation improves it. E ciency increases through the allocation of tasks among members of the team and the exploitation of their individual abilities.