OpenAI Baselines is a set of high-quality implementations of reinforcement learning algorithms that're intended to benefit the artificial intelligence research community. They make it easier for researchers to replicate, refine, and identify new ideas that can be explored and developed further.
The OpenAI Baselines were open-sourced by OpenAI in May, 2017. The initial release included DQN -- a reinforcement learning algorithm that combines Q-Learning and deep neural networks to let reinforcement learning work for complex, high-dimensional environments such as video games and robotics -- as well as three of DQN's variants.
Purpose
OpenAI's stated goal in open-sourcing the OpenAI Baselines is to ensure that apparent reinforcement learning advances are never due to comparisons with buggy or untuned versions of existing algorithms. This is important because reinforcement learning algorithms are highly complex, making results tricky to reproduce. Providing the entire research community with known-good implementations and best practices for creating them is intended to save time implementing pre-existing algorithms so that more time can be spent designing new ones.
