Products·OpenAI·Apr 2016

2. OpenAI Gym

Standardized RL benchmarks and environments

Blog Post

Summary

Release of OpenAI Gym, a toolkit for developing and comparing reinforcement learning algorithms, providing standardized environments and benchmarks for the RL research community.

Key Concepts

Standardized RL environments gave reinforcement learning its ImageNet moment

OpenAI Gym provided a standardized interface for RL environments, from simple cart-pole balancing to Atari games to continuous control. The key insight was that RL needed what MNIST and ImageNet had given supervised learning: shared benchmarks that enabled apples-to-apples comparison.

Simple, language-agnostic API: reset, step, render

The toolkit was open-source and language-agnostic, with a simple API: env.reset(), env.step(action), env.render().

Connections

Influenced by

1. Introducing OpenAI

Dec 2015

Influences

5. Proximal Policy Optimization (PPO)

Aug 2017