Learning and planning with logical automata

摘要

Abstract We introduce a method to learn policies from expert demonstrations that are interpretable and manipulable . We achieve interpretability by modeling the interactions between high-level actions as an automaton with connections to formal logic. We achieve manipulability by integrating this automaton into planning via Logical Value Iteration , so that changes to the automaton have predictable effects on the learned behavior. These qualities allow a human user to first understand what the model has learned, and then either correct the learned behavior or zero-shot generalize to new, similar tasks. Our inference method requires only low-level trajectories and a description of the environment in order to learn high-level rules. We achieve this by using a deep Bayesian nonparametric hierarchical model. We test our model on several domains of interest and also show results for a real-world implementation on a mobile robotic arm platform for lunchbox-packing and cabinet-opening tasks.