Evolution Strategies as a Scalable Alternative to Reinforcement Learning
About
We explore the use of Evolution Strategies (ES), a class of black box optimization algorithms, as an alternative to popular MDP-based RL techniques such as Q-learning and Policy Gradients. Experiments on MuJoCo and Atari show that ES is a viable solution strategy that scales extremely well with the number of CPUs available: By using a novel communication strategy based on common random numbers, our ES implementation only needs to communicate scalars, making it possible to scale to over a thousand parallel workers. This allows us to solve 3D humanoid walking in 10 minutes and obtain competitive results on most Atari games after one hour of training. In addition, we highlight several advantages of ES as a black box optimization technique: it is invariant to action frequency and delayed rewards, tolerant of extremely long horizons, and does not need temporal discounting or value function approximation.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Reinforcement Learning | LunarLanderContinuous v2 | Mean Reward115 | 59 | |
| Reinforcement Learning | Atari 2600 MONTEZUMA'S REVENGE | Score0.00e+0 | 45 | |
| Reinforcement Learning | HalfCheetah v3 | Mean Reward2.42e+3 | 34 | |
| Reinforcement Learning | InvertedPendulum v2 | Mean Reward651.9 | 27 | |
| Continuous Control | Humanoid 17-Dof | Final Return1.25e+4 | 21 | |
| Reinforcement Learning | Atari 2600 Qbert | Score147.5 | 20 | |
| Continuous Control | Hopper 3-Dof | Final Return2.56e+3 | 18 | |
| Reinforcement Learning | Swimmer v3 | Mean Reward318.4 | 15 | |
| Global Optimization | F5 benchmark function | Final Error0.0012 | 14 | |
| Global Optimization | F9 benchmark function | Final Error0.018 | 14 |