A centralized approach for representing and resolving interactions among multi-agent tasks while planning hierarchically

The advantages of hierarchical planning for construction projects and the difficulties associated with implementing hierarchical planning systems using artificial intelligence techniques are described. Needed extensions to a state-of-the-art AI planning system are identified and its subsequent hierarchical planning capabilities are demonstrated on an example project. The system for interactive planning and execution (SIPE) is a powerful, domain-independent planning system, augmented with frame representation and deductive capabilities. However, its heuristics for plan generation initially failed to meet the requirements of hierarchical planning in the construction domain. Two key limitations in the original SIPE planner are identified and corrected. First, SIPE was unable to consider all possible orderings among parallel branches and therefore was unable to solve problems with mutual interaction among parallel subplans. Second, SIPE provided only a limited representation language for problems that involve coordinating the tasks of several different agents.<
>