Data-Driven Optimal Output Feedback Control of Unknown System Model via Adaptive Dynamic Programming

This paper investigates the linear quadratic optimal output feedback control problem for an unknown linear continuous-time system. Combined with adaptive dynamic programming and optimal control theory, an online data-driven iteration learning algorithm is developed to learn an optimal controller from system data. The main advantage of the proposed algorithm is that it does not require an initial stabilizing control policy, a full-rank condition, or historical data storage to guarantee algorithm convergence. This is fundamentally different from the existing results based on the least-squares method, which requires these conditions. Moreover, the developed algorithm uses only the input and output data of the system, which solves the problem of unmeasurable system states. The simulation results demonstrate the efficacy of the proposed algorithm, and its superiority is demonstrated by comparison with the existing algorithms. Note to Practitioners—In the actual control systems, stability is only the basic condition that should be met. Besides, the control systems usually need to optimize certain indices to achieve the optimal control performance. Therefore, there is a practical research demand for the optimal control problem of the system. The traditional optimal control method requires the knowledge of the accurate system model, which is difficult to establish in actual control system. To this end, combined with the adaptive dynamic programming and optimal control theory, this paper proposes a data-based iteration learning algorithm to adaptively learn an optimal output feedback control policy from system data. Simulation results related to autonomous ground vehicle underscore the effectiveness of our proposed algorithm, and its superiority is further demonstrated through comparisons with existing methodologies.