TITLE: Learning to Reduce Communication Cost on Task Negotiation among Multiple Autonomous Mobile Robots AUTHOR: Takuya Ohko Department of Computer Science, Keio University 3-14-1, Hiyoshi, Kohoku, Yokohama 223, Japan E-mail: ohko@aa.cs.keio.ac.jp Kazuo Hiraki Electrotechnical Laboratory, MITI 1-1-4, Umezono, Tsukuba 305, Japan E-mail: khiraki@etl.go.jp Yuichiro Anzai Department of Computer Science, Keio University 3-14-1, Hiyoshi, Kohoku, Yokohama 223, Japan E-mail: anzai@aa.cs.keio.ac.jp ABSTRACT: This paper describes LEMMING, a learning system for task negotiation in multi-robot environments. LEMMING focuses on the problem of communication costs on Contract Net Protocol. Contract Net Protocol has been recognized as an attractive way for task negotiation. However, it is difficult for multi-robot systems to use wide-band communication lines enough to utilize standard Contract Net Protocol. It has been observed that the main communication cost on Contract Net Protocol is caused by broadcasting task announcements. In order to reduce this cost LEMMING uses Case-Based Reasoning(CBR). By using CBR, LEMMING can derive useful knowledge from messages in Contract Net Protocol and can find a suitable robot that should be received task announcements. We evaluate the idea of LEMMING in a simulated multi-robot environment. The result shows the advantage of LEMMING over standard contract net systems.