OpenAI has released two new implementations of its Baselines library: ACKTR and A2C. A2C is a version of Asynchronous Advantage Actor Critic (A3C) that runs synchronously, achieving similar performance to the original. ACKTR is another reinforcement learning algorithm that's more sample-efficient than TRPO and A2C, but requires only slightly more computation per update. --- Why it matters: These new algorithms are important for researchers and engineers working on reinforcement learning tasks because they offer improved efficiency and performance compared to existing methods. Source: https://openai.com/index/openai-baselines-acktr-a2c