Commonsense Reasoning for RL Tasks | State-of-the-Art