Information Technology Reference
In-Depth Information
The old PGP model has many constraints especially in the heterogeneous
agents' different problems solution. How to deal with dynamic agents change
solution policy again? How to finish the concrete tasks in the limited time? How
to negotiation between agents? There are some methods to extend the PGP
model. For example, using the general partial global planning, GPGP provides
new components to meet the need(Decker,1995).
Osawa proposed an approach to construct cooperative plan in open
environment based on robots' collaboration (Osawa,1993). In the approach,
rational agent used the “multi world model” construct incomplete single agent
plan and then according to the effect-based model to balance the cost. This
process can be described as following:
(1) Requestor sends RFP to call-board
(2) Leisure agent apply a RFP from call-board
(3) Call-board sends RFP to leisure agents
(4) Leisure agent generate the plan
(5) Leisure agent sends it to requestor
(6) Requestor investigates the possibility of cooperation
(7) Requestor sends cooperation rewards
(8) Applicants form cooperative plan
The utility is computed using the following formula
utility(
))
The average utility is the principal of cooperation. Although Osawa solved
the cooperative problems in open environment, it is not a good way to add the
utility to be divided for average value since the utility is only the order relation of
agent to goal. We can not compare the utilities between different kind of agents.
a,g
)=worth(
a,g
)-cost(plan(
a,g
14.5.4
The Planning based on constraint propagation
In multi-agent System, The coordination work of agents can be realized by
situated action and planning. The Method based situated action emphasized the
interaction between agents and environment, and the effects of perception——
action looping upon agent actions. The planning decides the sequence of agent
actions which will be executed, and the sequence will be decided by some
searching way. Compared with situated action, planning can make the agent
actions more correct and more rational, therefore planning is applied generally in
deliberative agent.
