IPC Planning Problems

该数据集包含了来自不同规划领域的标准IPC规划问题，这些领域包括方块世界、送货、机械手、物流、电梯调度、奖励、扳手以及遍历等。此外，该数据集不仅涵盖多种规划和实例，还着重于实现近乎100%的泛化能力，或理解导致失败的原因。对于大型实例，数据集可达高达40,000个采样的可达状态。其任务是对在学习到的价值函数中获得的贪婪策略进行测试，以评估其泛化能力、覆盖范围以及计划质量。

This dataset comprises standard International Planning Competition (IPC) planning problems across various planning domains, including Blocks World, Delivery, Manipulator, Logistics, Elevator Scheduling, Reward, Wrench, and Traversal, among others. Beyond encompassing diverse planning domains and problem instances, this dataset focuses on either achieving near-100% generalization capability or understanding the causes of failures. For large-scale instances, the dataset can include up to 40,000 sampled reachable states. Its task is to test the greedy policy derived from the learned value function to evaluate its generalization capability, coverage, and planning quality.