Reinforcement Learning for Industrial Applications

Programming desired task solutions for modern complex systems is often challenging, since it relies on detailed system understanding. In such cases, learning from data can be a useful alternative. Reinforcement learning (RL) is a general approach to learn policies while interacting with the system. In this thesis, we investigate the use of RL for several industrial applications, such as control of a robot arm and a throttle valve, and propose RL approaches while addressing practical constraints.