coinrun-patchs I'm doing some interpretabilty experiment based on Goal Misgeneralization in Deep Reinforcement Learning