Simulation-Based Planning for Computer Generated Forces

Planning in AI has been an active research topic for more than thirty years but only recently has it started to move in the direction of combining planning and execution to achieve what is sometimes called as`In-telligent Reactive Planning'. We propose simulation-based planning as a new way to perform intelligent re-active planning. Simulation-based planning |unlike most other planning systems|integrates simulation into the planning process. Once a set of plans is generated , simulations are used to test and evaluate the plans to choose the most applicable plan for that current situation. In most planning systems, plan evaluation depends on rules alone and because rules must be designed general enough to cover all possible cases, the evaluation is not speciic enough for some individual cases. However, when the plan evaluation is done through simulations, the evaluation can be more ne-tuned to individual cases and can allow better plans to be chosen for that individual case. From the military planning perspective, the simulation-based planner is also quite useful due to its ability to perform adver-sarial and multiagent planning. This is a natural consequence of using simulation in the planning process. By allowing other entities such as the enemy to simulate in parallel with the planner's forces, the planner is able to observe, prior to the actual execution, the eeects of adversarial and multiagent actions against its own plans.