Reinforcement learning architecture with a policy network (actor) and a value estimator (critic) that learn together.