Asynchronous Methods for Deep Reinforcement Learning: MuJoCo

The video shows agents trained using the Asynchronous Advantage Actor-Critic (A3C) algorithm performing a variety of motor control tasks. The tasks successfully learned by the agents include pole swing-up, quadruped locomotion, planar biped walking, balancing, 2D target reaching, and 3D manipulation. Paper link – http://arxiv.org/pdf/1602.01783.pdf