Lorenzo Steccanella

New framework learns state representations from trajectories without rewards

Researchers have developed a novel state representation framework for Markov decision processes (MDPs) that learns directly from state trajectories without needing reward signals or explicit action data. This method foc…