Gal A. Kaminka: Publications

Sorted by DateClassified by Publication TypeClassified by TopicGrouped by Student (current)Grouped by Former Students

Swarms Can be Rational

Yinon Douchan, Ran Wolf, and Gal A. Kaminka. Swarms Can be Rational. In Proceedings of the International Joint Conference on Autonomous Agents and Multi-Agent Systems, 2019.

Download

[PDF]2.9MB  

Abstract

Multi-robot systems are comprised of multiple robots, each under its own control, typically carrying out tasks towards a global goal. In these, spatial coordination (avoiding collisions) is a fundamental challenge. Swarm methods, where by robots coordinate ad-hoc and locally, with little or no communications offer a promising approach. However, while empirically demonstrated to be viable in practice, no theoretical guarantees of performance are known, nor a formalization of the task in a way that admits analysis. This paper formalizes a class of multi-robot cooperative tasks as potential extensive-form games. We show that the system coordination overhead is a potential function, forming a connection between the theoretical maximum-payoff equilibrium of the system, and the rational choices of individual robots during task execution: the robot swarm can be rational in theory. We then show how to approximate the rational decision-making in practice using reinforcement learning, while operating strictly within the limited capabilities of simple swarm robots. We empirically evaluate the efficacy of these methods in two multi-robot domains.

Additional Information

BibTeX

@inproceedings{aamas19,
  author = {Yinon Douchan and Ran Wolf and Gal A. Kaminka},
  title = {Swarms Can be Rational},
  booktitle = AAMAS,
  year = {2019},
  wwwnote = {},
  abstract = {
   Multi-robot systems are comprised of multiple robots, each under its own control, typically carrying out tasks towards a global goal. In these, spatial coordination (avoiding collisions) is a fundamental challenge. Swarm methods, where by robots coordinate ad-hoc and locally, with little or no communications offer a promising approach. However, while empirically demonstrated to be viable in practice, no theoretical guarantees of performance are known, nor a formalization of the task in a way that admits analysis.
   This paper formalizes a class of multi-robot cooperative tasks as  potential extensive-form games. 
   We show that the system coordination overhead is a potential function, forming a connection between the theoretical maximum-payoff equilibrium of the system, and the rational choices of individual robots during task execution: the robot swarm can be rational in theory.  We then show how to approximate the rational decision-making in practice using reinforcement learning, while operating strictly within the limited capabilities of simple swarm robots. We empirically evaluate the efficacy of these methods in two multi-robot domains.
  }
}

Generated by bib2html.pl (written by Patrick Riley ) on Mon Nov 16, 2020 22:25:46