Planning by autonomous agents with many concurrent goals in an elaborate simulated world

Complex simulated worlds populated by simulated agents having many concurrent goals present unique problems to planning systems. Neither a purely reactive model nor a traditional AI planning model is entirely adequate. The first because the agents need to know what they will be doing in the future to make good decisions about what to do in the present. The second because of the size and scope of the problems presented. The author describes a multi-level hybrid planning system consisting of a pre-formed plan language, a conflict resolution at the action level mechanism, and a coarse-grained, plan activities for the future facility that enables the Main Street simulated world to react to highly diverse sets of circumstances as well as to engage in long-range planning for the future