Gotta Learn Fast: A new benchmark for generalization in RL

Read on openai.com