next up previous
Next: Message Targeting/Distinguishing Up: Communication in Domains with Previous: Team Member Architecture

Communication Paradigm


The challenge for an agent to distinguish messages that are meant for it from those that are not is the first of five challenges that arise in the type of environment considered here. Second, since there is a single communication channel, agents must be prepared for active interference by hostile agents. A hostile agent could mimic messages it has previously heard at random times. Third, since the communication channel has low bandwidth, the team must prevent itself from all ``talking at once.'' Many communication utterances call for responses from all team members. However, if all team members respond simultaneously, few of the responses will get through. Fourth, since communication is unreliable, agents must be robust to lost messages: their behaviors cannot depend upon receiving communications from a teammate. Fifth, teams must determine how to maximize the chances that they are using the same team strategy despite the facts that each is acting autonomously and that communication is unreliable.

Table 1: The characteristics and challenges of the type of communication environment considered in this paper.

In order to meet these challenges, we propose that a team should use messages of the following form:

(<team-identifier> <unique-team-member-ID> <encoded-time-stamp> <time-stamped-team-strategy> <selected-internal-state> <target> <message-type> <message-data>)
Such a formulation assumes that the bandwidth allows for messages of several bytes in length to be transmitted in a reasonable amount of time. Some aural communication scenarios may need fewer, or condensed fields.

The contents of these fields are the product of the locker-room agreement. When forming the team, the agents must agree upon their team name (<team-identifier>) and a unique ID number for each member. For simplicity, the member IDs can be sequential numbers. These first two fields ensure that any teammate hearing the message knows precisely who uttered it. Teammates also agree ahead of time upon the security code used to create the field <encoded-time-stamp>. To coordinate, they agree upon a method for encoding and changing team strategies, and possibly upon positions of their internal states that should be communicated to help keep teammate information up to date. In addition, they must choose a set of acceptable message-types. The messages can use any syntactic and semantic codes (KQML [3] and KIF [5] for example). The only requirement is that the agents also agree on a mapping from message type to response requirements. Finally, the <target> field can be used to indicate the intended recipient(s) of the message. It could be intended for a single team member, for some subset of them, or for all team members.

The remainder of this section details how these particular message fields can be used to meet the challenges summarized in Table 1.

next up previous
Next: Message Targeting/Distinguishing Up: Communication in Domains with Previous: Team Member Architecture

Peter Stone
Mon Nov 24 11:31:14 EST 1997