Real Time Control of Reasoning.

Abstract : This award led to six major technical advances during the contract period. Several of these (approximate planning, dynamic backtracking and limited discrepancy search) promise to substantially change the way various AI subcommunities solve problems. Approximate planning formalizes an approach to planning that, instead of being correct (every plan returned achieves the goal) and complete (all such plans are returned), is approximately correct and complete, in that most plans returned achieve the goal and that most such plans are returned. The cached plans used by case-based planners are best thought of as approximate as opposed to exact, and the approximate approach can be used to attack planning subgoals separately and then combine the plans generated to produce a plan for the original goal. The computational benefits of working with subgoals separately have long been recognized, but attempts to do so using correct and complete planners have failed. Dynamic backtracking and limited discrepancy search are new approaches to solving constraint-satisfaction problems of the sort that arise in scheduling and other applications. Both allow the flexibility of 'lateral' movements in the search space, enabling far more efficient searches and leading to significant performance improvements in systems solving realistic problems.