We’ve discovered that evolution strategies (ES), an optimization technique that’s been known for decades, rivals the performance of standard reinforcement learning (RL) techniques on modern RL benchmarks (e.g. Atari/MuJoCo), while overcoming many of RL’s inconveniences.

In particular, ES is simpler to implement (there is no need for backpropagation), it is easier to scale in a distributed setting, it does not suffer in settings with sparse rewards, and has fewer hyperparameters. This outcome is surprising because ES resembles simple hill-climbing in a high-dimensional space based only on finite differences along a few random directions at each step.