Optimal Control and Reinforcement Learning for Robot: A Survey

Along with the development of systems and their applications, conventional control approaches are limited by system complexity and functions. The development of reinforcement learning and optimal control has become an impetus of engineering, which has show large potentials on automation. Currently, the optimization applications on robot are facing challenges caused by model bias, high dimensional systems, and computational complexity. To solve these issues, several researches proposed available data-driven optimization approaches. This survey aims to review the achievements on optimal control and reinforcement learning approaches for robots. This is not a complete and exhaustive survey, but provides some latest and remarkable achievements for optimal control of robots. It introduces the background and facing problem statement at the beginning. The developments of the solutions to existed issues for robot control and some notable control methods in these areas are reviewed briefly. In addition, the survey discusses the future development prospects from four aspects as research directions to achieve improving the efficiency of control, the artificial assistant learning, the applications in extreme environment and related subjects. The interdisciplinary researches are essential for engineering fields based on optimal control methods according to the perspective; which would not only promote engineering equipment to be more intelligent, but extend applications of optimal control approaches.