Interpretable Apprenticeship Learning with Temporal Logic Specifications
This work addresses the problem of making apprenticeship learning more interpretable for AI and robotics researchers, but it is incremental as it builds on existing LTL specification methods.
The paper tackles the inverse problem of inferring linear temporal logic (LTL) specifications from demonstrated behavior trajectories in Markov Decision Processes, formulating it as a multiobjective optimization problem and demonstrating efficacy in two simple domains using genetic programming.
Recent work has addressed using formulas in linear temporal logic (LTL) as specifications for agents planning in Markov Decision Processes (MDPs). We consider the inverse problem: inferring an LTL specification from demonstrated behavior trajectories in MDPs. We formulate this as a multiobjective optimization problem, and describe state-based ("what actually happened") and action-based ("what the agent expected to happen") objective functions based on a notion of "violation cost". We demonstrate the efficacy of the approach by employing genetic programming to solve this problem in two simple domains.