Communication Paradigm

 

The teamwork structure defined in Section 4 is designed to be appropriate for all PTS domains. In the subclass of PTS domains with single-channel, low-bandwidth, unreliable communication during the periods of limited communication, such as the soccer server, the communication paradigm defined in this section is also appropriate. The communication paradigm further illustrates the use of the locker-room agreement within the team member agent architecture. Recall that PTS domains may have no communication possible during the ``on-line'' periods. In those cases, of course, the communication paradigm presented here does not apply.

Domains with single-channel, low-bandwidth, unreliable communication raise several challenges for inter-agent communication. The locker-room agreement can be used to make inter-agent communication more efficient and reliable. The five challenges addressed by our communication approach are:

1. Team members need some method of identifying which messages on the single channel are intended for which agent.
2. 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.
3. Since the communication channel has low bandwidth, the team must prevent itself from ``talking all 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.
4. Since communication is unreliable, agents must be robust to lost messages: their behaviors cannot depend upon receiving communications from a teammate.
5. Teams must determine how to maximize the chances that they are using the same team strategy (formation) despite the facts that each is acting autonomously and that communication is unreliable.

The characteristics and challenges of this communication environment are summarized in Table 1.

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

In order to meet these challenges, a team uses messages with the following fields, all of whose syntax and semantics are defined within the locker-room agreement:

• The <team-identifier> identifies messages from within the team as opposed to another team in an adversarial environment.
• The <unique-team-member-ID> is a different sequential integer assigned to each team member.
• The <encoded-time-stamp> is a security code that can be used to verify a message's authenticity.
• The <time-stamped-team-strategy> indicates the current formation that the sender believes the team is using.
• The <selected-internal-state> can contain portions of the sender's internal state.
• The <message-type> and <message-data> contain the semantic content of the individual message. The messages can use any syntactic and semantic codes (KQML [10] and KIF [12] for example). The locker-room agreement also includes a mapping from message type to response requirements.
• The <target> indicates the intended recipient(s) of the message. It could be intended for a single team member identified either by <unique-team-member-ID> or by role within the team's current formation; for a unit of the current formation; or for all team members.

In addition to the protocol defined within the locker-room agreement, some internal state variables need to be devoted to communication. When an agent hears a message, it interprets it and updates the world state to reflect any information transmitted by the message. It also stores the content of the message as a special variable last-message. Furthermore, based on the locker-room agreement, an internal behavior then updates the internal state. If the message requires a response, three variables in the internal state are manipulated by an internal behavior: response, response-flag, and communicate-delay. response is the actual response that should be given by the agent as determined in part by the locker-room agreement. All three of these variables are then referenced by an external behavior to determine when a response should be given. For example one condition-action pair of the top-level external behavior might be: if (response-flag set and communicate-delay==0) then say(response).

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

• Message Targeting and Distinguishing
• Robustness to Active Interference
• Multiple Simultaneous Responses
• Robustness to Lost Messages
• Team Coordination