The Joy of Forgetting: Faster Anytime Search via Restarting

Anytime search algorithms solve optimisation problems by quickly finding a usually suboptimal solution and then find- ing improved solutions when given additional time. To de- liver a solution quickly, they are typically greedy with re- spect to the heuristic cost-to-go estimate h. In this paper, we first show that this low-h bias can cause poor perfor- mance if the heuristic is inaccurate. Building on this observa- tion, we then present a new anytime approach that restarts the search from the initial state every time a new solution is found. We demonstrate the utility of our method via ex- periments in PDDL planning as well as other domains. We show that it is particularly useful for hard optimisation prob- lems like planning where heuristics may be quite inaccurate and inadmissible, and where the greedy solution makes early mistakes.