Reinforcement learning (RL) is a machine learning method used to find an optimal policy in an uncertain environment. Although reinforcement learning algorithms achieve great success in model-free learning of the state-feedback control, it is rarely used in the observer-based control. In this paper, we develop a design scheme of the observer-based control integrated with residual generation in the framework of RL, which has the ability to detect the system fault. An actor-critic RL algorithm is designed to train the observe-based control, which proved a parameter-free way to design the observer-based control with residual generator. The performance and effectiveness of the proposed scheme are demonstrated through a robot test rig. After a short period of learning, the robot is controlled only with the measured joint angle, and meanwhile the residual generator can be used for fault detection to improve the system reliability. The proposed control strategy and the learning scheme can be u...

Observer-based control has been widely used in mechatronic systems. In this article, an observer-based control integrated with a residual generator is designed in the framework of actor–critic reinforcement learning, which has been applied to robot systems. In the learning process, a critic function is constructed by the state of the original system and its twin system. Thus, the system parameters and control gain can be obtained simultaneously through trial-and-error learning. To achieve system stability and reliability, the observer-based control with the residual generator is designed based on the learned results. The performance and effectiveness of the proposed scheme are demonstrated through a robot test rig. After a short period of learning, the robot is controlled only with the measured joint angle, and meanwhile, the residual generator can be used for fault detection to improve the system reliability.