Solving Hard AI Planning Instances Using Curriculum-Driven Deep Reinforcement Learning
This addresses a critical bottleneck in AI planning for domains like Sokoban, offering a novel solution to previously unsolved problems, though it is incremental in applying deep RL with a curriculum to a specific hard domain.
The paper tackled the problem of solving hard Sokoban planning instances, which are PSPACE-complete and out of reach for current AI planners, by using a curriculum-driven deep reinforcement learning approach that automatically uncovers domain structure and solves these instances within one day of training, while other solvers fail within reasonable time limits.
Despite significant progress in general AI planning, certain domains remain out of reach of current AI planning systems. Sokoban is a PSPACE-complete planning task and represents one of the hardest domains for current AI planners. Even domain-specific specialized search methods fail quickly due to the exponential search complexity on hard instances. Our approach based on deep reinforcement learning augmented with a curriculum-driven method is the first one to solve hard instances within one day of training while other modern solvers cannot solve these instances within any reasonable time limit. In contrast to prior efforts, which use carefully handcrafted pruning techniques, our approach automatically uncovers domain structure. Our results reveal that deep RL provides a promising framework for solving previously unsolved AI planning problems, provided a proper training curriculum can be devised.