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.

Tim Salimans, Jonathan Ho, Xi Chen, Szymon Sidor, Ilya Sutskever · 2017

Related benchmarks

Task	Dataset	Result
Reinforcement Learning	LunarLanderContinuous v2	Mean Reward114.97	65
Reinforcement Learning	Atari 2600 MONTEZUMA'S REVENGE	Score0	45
Reinforcement Learning	HalfCheetah v3	Mean Reward2423.16	34
Reinforcement Learning	InvertedPendulum v2	Mean Reward651.86	27
Global Optimization	F2 benchmark function	Final Error0.012	25
Continuous Control	Humanoid 17-Dof	Final Return12544	21
Reinforcement Learning	Atari 2600 Qbert	Score147.5	20
Global Optimization	F6 benchmark function	F6 Final Error9.7	19
Continuous Control	Hopper 3-Dof	Final Return2557	18
Reinforcement Learning	Swimmer v3	Mean Reward318.39	15