Real-Time Heuristic Search: Research Issues

Real-time (heuristic) search methods allow for fine-grained control over how much planning to do between plan exe-cutions. Many real-time search methods can use heuristic knowledge to guide planning, be interrupted at any state and resume execution at a different state, and improve their plan-execution time as they solve similar planning tasks, until their plan-execution time is optimal. Unfortunately, the behavior of real-time search methods is not yet well understood. In this paper, we show that the behavior of real-time search methods often differs from the behavior of more traditional and well-studied search methods and argue that it is impor-tant to investigate how properties of the heuristic functions, the domains, and the real-time search methods themselves influence their performance