PeGLearn
RA-L and ICRA 2023
Our prior work in LfD-STL required the demonstrators to explicitly specify their preferences by ranking the STL specifications. The ranked specifications were represented by a directed acyclic graph (DAG) to capture the preferences and dependencies. In this paper, we propose an algorithm - Performance Graph Learning (PeGLearn) - to automatically infer the specification DAG from demonstrations. We also show how PeGLearn facilitates explainability for AI-based systems via a user study on CARLA, a simulated driving environment.
This paper was published in IEEE Robotics and Automation Letters (RA-L). It was one of the 10 papers in the Reinforcement Learning category accepted for oral presentation at the 2023 IEEE International Conference on Robotics and Automation (ICRA).