Tag
OpenAI demonstrates a method for training a reinforcement learning agent to play Montezuma's Revenge from a single human demonstration, addressing the challenge of sparse rewards through curriculum learning and careful hyperparameter tuning. The approach achieves strong performance on the notoriously difficult Atari game while showing generalization limitations on other titles.