Simulation-based planning for multi-agent environments

One of the key issues in reasoning with multiple interacting intelligent agents is how to model and code the decision making process of the agents. In Artificial Intelligence (AI), the major focus has been on modeling individual intelligence and a common approach has been to use operator or rule-based models to represent the decision making intelligence of an agent. If the purpose of the simulation is to precisely emulate a particular agent's intelligence, then such rule-based models may often be most appropriate. However, when the goal is to win the engagement in the battlefield, where the overall outcome may depend on individual execution of each task, the level of detail must be extended to the level of simulating individual task execution. In these cases, we have created a methodology, Simulation-Based Planning (SEP), that embeds one simulation inside another. The embedded simulation simulates the actions of agents before committing to a plan so that it may evaluate the desiredness of the actions. Plan alternatives are generated based on discrete paths through spatial regions of a domain, while specific optimal plans are generated through the use of experimental design and simulation. We have found that, through simulation-based planning, near-optimal plans can be selected by using simulation, in addition to using simulation once a plan has been adopted.