Learning‐based robust control methodologies under information constraints

Abstract

This editorial is dealing with the collection and report of some recent advances in learning-based robust control methodologies under information constraints. Both theoretical and practical contributions focusing on this theme are partly addressed in this special issue. Particularly, the latest progress of learning-based control in autonomous systems, large-scale systems, interconnected systems, robotics, industrial mechatronics, transportation and variously broad applications are introduced to the literature through this special issue. Within the past decade, various learning-based control technologies have prosperously emerged in both academic and industrial communities, and have expectantly performed remarkable superiority in terms of intelligence, autonomy, conciseness, reliability, resilience, and so forth. At the early stage, neural/fuzzy learning architectures have been widely deployed to online capture complex unknowns including unmodeled dynamics, uncertainties, and disturbances pertaining to the plant which might be a nonlinear system addressing the vehicles, robotics, transportations, mechatronics, informatics, circuits, and so forth. Recently, fruitful machine learning-based approaches, for example, reinforcement learning, deep learning, brain-inspired learning, have been incrementally promoted to innovate traditional learning-based intelligent control methodologies in both theoretical and practical sides. Especially, promising applications have also been developed to autonomous systems and robotics. In addition to booming advances in learning-based control philosophy, within a complex system, unexpected constraints would be inevitably involved, for instance, communication delays, sensor failures/noises, actuator nonlinearities, nonholonomic/underactuated dynamics, and so forth, especially within distributed systems. With a stringent peer review process, there are 34 papers finally included in this Special Issue, which are covering the following aspects: (1) Learning-based control methodologies over network; (2) Estimation and fault detection under information constraints; and (3) Case studies. A brief summary of the accepted papers is discussed in the following. The authors in Reference 1 studied the parameter learning problem for stochastic Boolean networks. Then, a numerical experiment is presented to show the usefulness of the designed parameter learning algorithm. In Reference 2, the authors investigated a distributed online learning problem with privacy preservation, in which the learning nodes in a distributed network aims to minimize the sum of local loss functions over finite-time horizon. The authors in Reference 3 studied the sliding mode control of Markovian jump systems under communication constraints subject to unavailable states and deception attack and a dynamic event-triggered scheme. In Reference 4, the authors designed a sliding mode control for the discrete-time interval type-2 fuzzy singularly perturbed systems under a component-based dynamic event-triggering scheme. The authors in Reference 5 developed an observer-based proportional integral derivative security control for a networked system subject to aperiodic denial-of-service attacks. In Reference 6, the authors investigated the finite-time tracking control problem for nonstrict feedback nonlinear systems subject to full state constraints. Then, an example of the electromechanical system is given to verify the applicability of the purposed method. In Reference 7, the problem of stability analysis is investigated for a class of switched nonlinear systems whose control inputs include time delay and sampling. The authors in Reference 8 studied the finite-iteration tracking control problem for discrete-time linear systems in repetitive process setting with external disturbances based on the robust iterative learning control scheme. In Reference 9, the authors designed a set of dual-mode feedback controllers for a class of Markovian jump Lur'e systems in the framework of efficient model predictive control to make a nice tradeoff among the online computation burden, the initial feasible region, and the control performance. In Reference 10, the authors studied the optimal control problem via reinforcement learning for a class of nonlinear discrete-time systems. Then, a new learning-based algorithm, T-step heuristic dynamic programming with eligibility traces (T-sHDP( λ )) is proposed to tackle the optimal control problem. The authors in Reference 11, the authors proposed a method to compute a control Lyapunov function for nonlinear dynamics based on a deep learning robust neuro-control strategy. An estimation of the region of attraction is produced for advanced stability analysis as well. In Reference 12, a reinforcement learning based approach is proposed for model-free robust optimal regulation of continuous-time nonlinear systems. Specifically, incremental adaptive dynamic programming is utilized to allow the design of the approximate optimal incremental control strategy, stabilizing the controlled system incrementally under model uncertainties, environmental disturbances, and input saturation. In addition, the authors in Reference 13 developed a reinforcement learning-based robust control strategy for uncertain heterogeneous multi-agent systems based on a fully distributed adaptive observer. On the other hand, in Reference 14, the authors developed a new proportional-integral-derivative-based tracking control for uncertain Euler–Lagrange systems subject to actuation failures and saturation. In Reference 15, the authors proposed a novel adaptive neural network control scheme to resolve the tracking control problem for flexible-joint robots with random noises. In addition, the command filtered technology is applied to the adaptive neural network design framework. Moreover, the authors in Reference 16 developed a control scheme with deferred and asymmetric full-state constraints based on an asymmetric barrier Lyapunov function. In addition, radial basis function neural network and an antiwindup compensator are adopted for the problems of uncertainties and input saturation. Meanwhile, the authors in Reference 17 proposed a neural network-based control for unknown discrete-time nonlinear systems with a denial-of-service attack and an adaptive event-triggered mechanism. The authors in Reference 18 investigated the problem of simultaneous estimation of states and uncertainties by using limited information based on a sampled-data learning observer scheme. It is shown that the uncertainty estimation is performed by a learning equation with only simple addition operations, which is particularly suitable for actual digital system scenarios. Moreover, the authors in Reference 19 proposed a distributed security filtering scheme for networked switched systems under the round-robin protocol and a cyber attack scenario. The authors in Reference 20 proposed a data-stream-driven event-triggered control strategy using evolving fuzzy models learned by granulation of input-output samples for nonlinear systems with unknown time-varying dynamics. The estimation of the parameters of the fuzzy information granules by using a novel Fuzzy Weighted Recursive Least Squares with Variable-Direction Forgetting. In Reference 21, the authors proposed a single-hidden-layer feedforward network aided fault-tolerant control scheme for a class of nonlinear systems subject to actuator faults. In addition, a sliding mode control approach is presented to allow for prompt corrective reactions, with explicit consideration of multivariable conditions and control input constraints. Finally, a nonlinear Boeing 747 aircraft model is used to demonstrate the effectiveness of the proposed fault-tolerant control scheme. The authors in Reference 22 proposed an adaptive second-order sliding mode control for attitude regulation of spacecraft. In the presence of actuation fault, inertial uncertainties, external disturbances, and input saturation, the problem of finite time attitude stabilization is ensured. Moreover, the authors in Reference 23 proposed a cooperative time-varying formation control scheme for heterogeneous multi-agent systems with unknown actuator faults and external disturbances. The authors in Reference 24 proposed the finite-time formation tracking control for a multi-agent system with obstacle avoidance based on the artificial potential field. The results are validated by simulating multiple autonomous underwater vehicles system and it is shown that the obstacle avoidance with high-precision tracking and formation performance will be achieved under the formation trajectory tracking controller. Moreover, in Reference 25, the authors addressed the problem of cooperative learning from adaptive neural formation control for a group of underactuated unmanned surface vehicles with modeling uncertainties. On the other hand, the authors in Reference 26 designed the finite-time global trajectory tracking control for autonomous underwater vehicles in presence of input saturation constraints, actuator faults, unknown dynamics, and external disturbances. The authors in Reference 27 proposed a self-learning-based optimal tracking control for an unmanned surface vehicle based on the actor-critic reinforcement learning mechanism and backstepping technique. Then, applicability of the proposed method is illustrated through a prototype of unmanned surface vehicle. In Reference 28, the authors developed a neural-network based backstepping control for autonomous marine vehicles perturbed by external disturbances. Moreover, in Reference 29, the authors proposed a neural-network-based adaptive finite-time output constraint control scheme for attitude stabilization of rigid spacecrafts based on a singularity-free terminal sliding mode variable. Moreover, in Reference 30, the problem of leader-following consensus is addressed for a group of spacecraft systems with uncertain inertia matrices subject to unknown bounded disturbance over switching communication networks. In Reference 31, the authors developed an adaptive backstepping flight controller under angle of attack constraints for hypersonic flight vehicles based on the asymmetric integral barrier Lyapunov function. Then, the neural networks and disturbance observer are applied in the controller to deal with the model uncertainty and external disturbance. Moreover, in Reference 32, the authors proposed a novel asynchronous advantage actor-critic learning-based dynamic event-triggered mechanism for the decentralized load frequency regulation to alleviate the local-area communication burden and influence of the load fluctuations. In the proposed algorithm framework, the long short-term memory network is used to estimate the policy function and value function. In Reference 33, the authors proposed an adaptive robust barrier-based control for a three-dimensional riser system subject to system uncertainties and output constraints. In addition, in Reference 34, the authors developed an adaptive fuzzy fixed-time controller for the permanent magnet synchronous motors based on the fixed-time stability theory, adaptive control, fuzzy control, and backstepping algorithm. We really appreciate all the authors and anonymous reviewers who contributed to this Special Issue. Meanwhile, we would like to thank the supports from the Editor-in-Chief and Editorial Staff to our Special Issue. This work was partially supported by the Italian Ministry of Education, University and Research through the Project “Department of Excellence LIS4.0-Lightweight and Smart Structures for Industry 4.0,” the Equipment Pre-Research Fund of Key Laboratory (6142215200106), and the Liaoning Revitalization Talents Program (under Grant XLYC1807013). The authors declare that they have no conflicts of interest.