An Adaptive-Gain Reinforcement Iterative Learning Control Scheme for Nonlinear Systems to Track Morphologically Similar Targets

This paper investigates the problem of tracking morphologically similar targets for nonlinear systems and proposes an adaptive-gain reinforcement iterative learning control (AG-RILC) scheme. Unlike existing control approaches, where gain schemes are predefined before the control process, this study incorporates a reinforcement learning (RL) mechanism into a novel iterative learning control (ILC) framework. Specifically, this novel ILC scheme can address the tracking problem for morphologically similar targets using the condition-performance matching algorithm and time-scale transformation. Further, an adaptive-gain scheme is designed to adapt to the control performance metrics by introducing a weighting factor that integrates the decreasing gain sequences of ILC with RL’s powerful exploratory capabilities. In addition, it is proved that under the proposed scheme, the control input error converges to zero with the number of iterations approaches infinity. Finally, illustrative simulations demonstrate that the AG-RILC scheme achieves a balance between high tracking accuracy and fast convergence. Experimental outcomes validate the performance benefits of the AG-RILC algorithm. Note to Practitioners—Many engineering systems, such as rail transportation systems, robotic systems, and chemical process systems, require tracking repetitive tasks with morphological similarity over finite time intervals. Iterative learning control (ILC) is a well-established approach for such repetitive control tasks. However, many existing ILC methods for nonlinear systems rely on constant or predefined decreasing learning gains before the learning process, which prevents real-time adaptation to control performance. As a result, these methods struggle to effectively balance fast convergence with high tracking accuracy. In this paper, we overcome the difficulty of learning parameter tuning in classical ILC algorithms, and draw inspiration from reinforcement learning (RL) algorithm to propose an adaptive-gain reinforcement iterative learning control (AG-RILC) scheme for nonlinear systems. The scheme strikes an optimal balance between fast convergence and accurate tracking performance, resulting in superior overall control performance.