Reinforcement Learning through Asynchronous Advantage Actor-Critic on a GPU

We introduce a hybrid CPU/GPU version of the Asynchronous Advantage-Critic (A3C) algorithm, currently the state-of-the-art method in learning for various gaming tasks. We analyze its computational and concentrate on aspects critical to leveraging the GPU's power. We introduce a system of queues and a dynamic scheduling, potentially helpful for other asynchronous algorithms as well. Our CPU/GPU version of A3C, based on TensorFlow, achieves a significant up compared to a CPU implementation; we make it publicly available to researchers at https: //github. com/NVlabs/GA3C.