Learning STRIPS+ Action Models from Traces with Partial Information

This research addresses the problem of learning STRIPS+ action models from action traces when both actions and states are only partially observable. Previous work demonstrated that lifted STRIPS models could be learned from traces alone, but STRIPS actions include unnecessary arguments. A subsequent approach using STRIPS+ models, where some arguments are implicit, assumed fully observable states. This paper relaxes that assumption, formulating algorithms and completeness results for learning from traces with partial information about both actions and states. The work is theoretical and does not involve any art-related content.