Learning Synthetic Environments for Reinforcement Learning with Evolution Strategies