In this article, we discuss a near-optimal tracking control problem (NOTCP) of robots used for inspecting aircraft skin with partially unknown systems, unmeasurable states, unknown disturbances, and unknown output delay. A novel observer based on an augmented neural network is designed to overcome the unknown disturbances, unknown output delay, and unknown internal states. An augmented system state, composed of the tracking error and reference system state, is proposed to introduce a new nonquadratic discounted performance function for the NOTCP. Due to the complexity in solving the Hamilton-Jacobi-Bellman equation, an online policy iteration is presented under the adaptive dynamic programming (ADP) framework. Unlike the traditional ADP, the event-driven algorithm updates the control input only when the event is triggered, which reduces the computational cost and transmission load. Both the control policy and the observer are updated according to the developed triggering condition. Convergence to a near-optimal control solution and the stability analysis of the proposed algorithm are shown through the Lyapunov candidate function for both the continuous and jump dynamics. The performance of the proposed algorithm is demonstrated by simulation.
K E Y W O R D Sadaptive dynamic programming, augmented neural network observer, event driven, nonlinear partially unknown systems, online learning and control
INTRODUCTIONWith the continuous advancement of new technologies, service robots have begun to enter into people's work and life to replace tedious, repetitive, and onerous tasks. 1 Among them, the climbing robots have attracted increasing attention as their applications are found in industry, 2 dangerous areas, 3 and domestic needs. 4 Damaged skins of aircrafts have led to a number of air crashes. Nevertheless, there are many shortcomings, such as high cost and low efficiency. An oversight of the staff is also a potential threat. Such circumstances motivate people to design robots for replacing/reducing human involvement. This article introduces a robot for inspecting skins of aircrafts. 5 In general, there always exist some problems in physical systems, for instance, unmeasurable system states, disturbances, and system uncertainties between mathematical models and practical systems. Such issues may lead to degrade the controlling performance. Hence, the designed controlling policy should be robust enough.