Deep Reinforcement Learning in Autonomous Driving Works best with discrete actions: 4 workers, 1e-4 learning rate Couldn't get it to work for the continuous actions space; something goes wrong with the actions it generates The A3C idea Overview Architecture of the Artificial Neural Network
