Reducing Communication Load on Contract Net by Case-Based Reasoning-- Extension with Directed Contract and Forgetting

This paper describes the communication load reduction on the task negotiation with Cont.rart Net Protor~d for multiph: autonomou~ lnobilc robots. Wc have dcvek)pe(l LEMMING. a task negotiation system with low ~xmnmtnication load for nmltiple atttonomolts mobile robots. For controlling multiple robots, Contract Net Protocol(CNP) is useful, but the broadcast of the Task Announcement messages on CNP tends to consnnic lnltdl comnnmicatitm load. In ordcr to overcome this probhun LEMMIN(; learns proper addressees for the Task Announct,nlcnt message.s with Case-Ba.~ed Rcasoning(CBl’t) so as to suppress the broadcast. The learning method is called Addrrssee Learning. In this paper, we extend LEMMING with "dirertcd contract" to r~lucc the communication load more cffectivcly. Moreover. wc extend LEMMING with "forgetting" to restrict the munbex of cases, sillcc it is impossible to have enough memory to keep MI the cases. The efficiency of LEMMI.NG is cvaluated in a silmtlated multi-robot environment to show that these extensions arc effec.tive for LEMMI.~e;.