Deep reinforcement learning methods are capable of learning complex heuristics starting with no prior knowledge, but struggle in environments where the learning signal is sparse. In contrast, planning methods can discover ...