SAILOR is an inverse RL algorithm that learns world and reward models to search at test-time and recover from mistakes. - View it on GitHub