Learning Domain-Independent Policies for Open List Selection

Since its proposal over a decade ago, LAMA has been considered one of the best-performing satisficing classical planners. Its key component is heuristic search with multiple open lists, each using a different heuristic function to order states. Even with a very simple, ad-hoc policy for open list selection, LAMA achieves state-of-the-art results. In this paper, we propose to use dynamic algorithm configuration to learn such policies in a principled and data-driven manner. On the learning side, we show how to train a reinforcement learning agent over several heterogeneous environments, aiming at zero-shot generalization to new related domains. On the planning side, our experimental results show that the trained policies often reach the performance of LAMA, and sometimes even perform better. Furthermore, our analysis of different policies shows that prioritizing states reached via preferred operators is crucial, explaining the strong performance of LAMA.