Performance-based data-driven optimal tracking control of shape memory alloy actuated manipulator through reinforcement learning

• Published in: Engineering applications of artificial intelligence Vol. 114; p. 105060
• Main Authors: Liu, Hongshuai, Cheng, Qiang, Xiao, Jichun, Hao, Lina
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.09.2022
• Subjects: Data-driven control; Prescribed constraints; Reinforcement learning; SMA manipulator; Prescribed constraints; Data-driven control; Reinforcement learning; SMA manipulator
• ISSN: 0952-1976
• DOI: 10.1016/j.engappai.2022.105060

This article focuses on the continuous-time optimal tracking control problem of a shape memory alloy (SMA) actuated manipulator subject to prescribed error constraints and completely unknown nonlinear dynamics. Firstly, prespecified error constraints imposed by the prescribed performance function (PPF) are transformed into an equivalent unconstrained task by adopting a transformation function, where the newly designed PPF is irrelated to the initial condition. An unconstrained augmented system and a long-term discounted performance considering tracking errors and input cost are constructed for the convenience of designing the optimal controller. Then, a model-based optimal control algorithm is derived to solve the Hamilton–Jacobi–Bellman equation (HJBE). Next, data-driven reinforcement learning (RL) is employed to approximate the solution of the HJBE iteratively to obviate requiring an SMA manipulator model. Moreover, critic neural networks (NNs) and actor NNs are introduced to the RL-based optimal controller, and the least-squares method is used to find the parameters for the NNs-based RL algorithm. Rigorous theoretical analyses demonstrate that the proposed controller can stable the SMA manipulator system, and the tracking errors are always constrained within the prescribed region. Finally, experiments are conducted on the established SMA actuated manipulator platform, and the results illustrate that the proposed controller is feasible and effective. •A data-driven RL control algorithm is proposed for the SMA manipulator.•The CTOTCP and prescribed error constraints are addressed simultaneously.•The designed PPF is irrelated to the initial error condition.•The proposed scheme can overcome the inadequate exploration in the training process.•Experiment results illustrate the effectiveness and superiority of our approach.