Scalable Potential-Field Multi-Agent Coordination In Resource Distribution Tasks

this paper, we look at a scalable, generic methodology for coor- dination in large, embodied multi-agent systems (MAS), operating in noisy, real-time environments. More specifically, we look at real- world task assignment problems and use the testbed of robots per- forming resource distribution in large storage facilities. The main problem when using MAS for task assignment problems in real- world environments is the limited scalability of traditional MAS approaches such as centralized planning. As a first step toward a generic methodology for embodied, dis- tributed MAS, we analyse the behavior and the scalability of an embodied MAS which is driven by simulated potential fields. Re- sources in a storage facility emit a certain simulated potential while robots emit an opposite potential. Applying principles inspired by physics to local behavior, idle robots move toward resources but away from other robots. Whenever a resource appears, a central agent system registers this appearance and creates a plan for a ro- bot to pick up the resource and deliver it to some destination area. We show that this system is highly scalable with respect to both the environment and the number of robots, while maintaining func- tionality, adaptivity and robustness. Furthermore, the integration of planning prevents local optima and increases fairness. In future work, the central agent system will be replaced by a dis- tributed system, such as a sensor network, to further increase scal- ability, adaptivity and robustness of the system.

[1]  Ann Nowé,et al.  Homo Egualis Reinforcement Learning Agents for Load Balancing , 2002, WRAC.

[2]  O. D. Kellogg Foundations of potential theory , 1934 .

[3]  Danny Weyns,et al.  Gradient field-based task assignment in an AGV transportation system , 2006, AAMAS '06.

[4]  Daniele Nardi,et al.  Multirobot systems: a classification focused on coordination , 2004, IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics).

[5]  G. Hardin The Tragedy of the Commons , 2009 .

[6]  Fabien Michel,et al.  Environments for Multi-Agent Systems III , 2008 .

[7]  David W. Payton,et al.  Pheromone Robotics , 2001, Auton. Robots.

[8]  Karl Tuyls,et al.  An Overview of Cooperative and Competitive Multiagent Learning , 2005, LAMAS.

[9]  H. Van Dyke Parunak,et al.  Swarming Distributed Pattern Detection and Classification , 2004, E4MAS.

[10]  Michael L. Littman,et al.  Markov Games as a Framework for Multi-Agent Reinforcement Learning , 1994, ICML.

[11]  P. Stone,et al.  Continuous area sweeping: a task definition and initial approach , 2005, ICAR '05. Proceedings., 12th International Conference on Advanced Robotics, 2005..

[12]  H. Van Dyke Parunak,et al.  Multiple Pheromones for Improved Guidance , 2000 .

[13]  Jacques Ferber,et al.  Environments for Multiagent Systems State-of-the-Art and Research Challenges , 2004, E4MAS.

[14]  Sean Luke,et al.  A pheromone-based utility model for collaborative foraging , 2004, Proceedings of the Third International Joint Conference on Autonomous Agents and Multiagent Systems, 2004. AAMAS 2004..

[15]  Andrew W. Moore,et al.  Reinforcement Learning: A Survey , 1996, J. Artif. Intell. Res..

[16]  Chris Melhuish,et al.  Stigmergy, Self-Organization, and Sorting in Collective Robotics , 1999, Artificial Life.

[17]  Sean Luke,et al.  Cooperative Multi-Agent Learning: The State of the Art , 2005, Autonomous Agents and Multi-Agent Systems.

[18]  H. Van Dyke Parunak,et al.  Digital Pheromones for Coordination of Unmanned Vehicles , 2004, E4MAS.

[19]  Edsger W. Dijkstra,et al.  A note on two problems in connexion with graphs , 1959, Numerische Mathematik.