Altruism in Multi Robot Communities

As robots continue advancing and permeating our world, the frequency of interactions between robots that do not share a common goal will increase. Each robot in the world will likely have a different fitness function to optimize due to each robot having independent goals and owners. Such collections of robots, termed a Multi-Robot Community (MRC), generally, have no shared global fitness function to maximize. This is in contrast to most Multi-Robot Systems (MRSs), which are designed to cooperate to accomplish common goals and hence have a shared global fitness function. A MRC may contain individual robots, MRSs, or a combination. To establish cooperative task allocation in MRCs, a new task allocation framework is required.

Using a reward based system to facilitate coordination is impractical in a MRC, where a reward for accomplishing one robots task has no value for another robot. Rather, coordination in a MRC is focussed on reducing robot costs. To accomplish this, robots build relationships with other robots. These relationships, mimicking those in humans, grow over time between willing parties through robots performing altruistic actions for other robots. Robots performing an altruistic action incurs a cost without any reward or promise of payback. The relationships evolve while still protecting against selfishness. Three altruistic controllers have been developed that model the reciprocating altruistic relationships. The first, termed one-to-one, allows relationships to form only between pairs of robots. The second, termed one-to-many, allows relationships to form between individual robots and sets of robots. The third, termed one-to-mixed, is a hybrid of the previous two controllers.