BRT: Biased Rapidly-exploring Tree

This paper describes the planner BRT (Biased Rapidlyexploring Tree). This planner is basically a Rapidlyexploring Random Tree (RRT) adapted to automated planning that employs Fast-Downward as the base planner. The novelty in this case is that it does not sample the search space in a random way; rather, it estimates which propositions are more likely to be achieved along some solution plan and uses that estimation (called bias) in order to sample more relevant intermediates states. The bias is computed using a messagepassing algorithm on the planning graph with landmarks as