Review of Deep Reinforcement Learning for Robot Manipulation

Reinforcement learning combined with neural networks has recently led to a wide range of successes in learning policies in different domains. For robot manipulation, reinforcement learning algorithms bring the hope for machines to have the human-like abilities by directly learning dexterous manipulation from raw pixels. In this review paper, we address the current status of reinforcement learning algorithms used in the field. We also cover essential theoretical background and main issues with current algorithms, which are limiting their applications of reinforcement learning algorithms in solving practical problems in robotics. We also share our thoughts on a number of future directions for reinforcement learning research.