Machine learning based iterative learning control for non‐repetitive time‐varying systems

Resource-efficient ILC for LTI/LTV systems through LQ tracking and stable inversion: Enabling large feedforward tasks on a position-dependent printer

Iterative learning control approach for linear discrete time-varying systems with different initial time points

The problem of iterative learning control (ILC) is considered for a class of time-varying systems with random packet dropouts. It is assumed that an ILC scheme is implemented via a remote control system and that packet dropout occurs during the packet transmission between the ILC controller and the actuator of remote plant. The packet dropout is viewed as a binary switching sequence which is subject to the Bernoulli distribution. In order to eliminate the effect of packet dropouts on the convergence property of output error, the hold-input scheme is adopted to compensate the packet dropout at the actuator. It is shown that the hold-input scheme with average ILC can achieve asymptotical convergence along the iteration axis for discrete time-varying linear system. Numerical examples are provided to show the effectiveness of the proposed method.

Highly dynamic tasks that require large accelerations and precise tracking usually rely on precise models and/or high gain feedback. While movement primitives allow for efficient representation of such tasks from demonstrations, the optimization of the required motor commands for systems with inaccurate dynamic models remains an open problem. To achieve accurate tracking for such tasks, we investigate two related Iterative Learning Control update laws and present a variant suited for optimizing hitting movement primitives. The resulting algorithm generalizes well to different initial conditions and naturally addresses striking movements where reaching specific velocities at certain positions is crucial. We evaluate the performance of our approach in a simulated putting task as well as in robotic table tennis, where we show how the striking performance of a seven degree of freedom anthropomorphic arm can be optimized. Our final implemented algorithm compares favorably with two state-of-the-art approaches.

This paper investigates robustness and convergence of iterative learning control (ILC) for mecanum-wheeled mobile robots operating under plant uncertainty and nonrepetitive disturbances. A Model-Based Corrector (MBC) ILC is introduced, which employs a fixed nominal inverse model to update the control input each iteration, thereby eliminating the need for online parameter estimation. Theoretical analysis based on the lifted-system formulation shows that convergence depends solely on the nominal model’s fidelity, ensuring stability under moderate modeling errors. The proposed method is experimentally compared with a baseline P-type ILC and an adaptive parameter-estimation ILC on a lemniscate trajectory tracking task subject to Gaussian measurement noise, a one-time disturbance, and a permanent plant change. Quantitative results demonstrate that MBC-ILC achieves the desired noise-floor accuracy (RMSE ≈ 0.02) within a few iterations, rejects nonrepetitive disturbances in a single iteration, and re-converges within six iterations following a plant change. Results show that a fixed-model corrector achieves a strong tradeoff of convergence speed, steady accuracy, and robustness at low computational cost, making it practical control architecture for robust trajectory learning in mobile robots.

In this article, we propose a data-driven iterative learning control (ILC) framework for unknown nonlinear nonaffine repetitive discrete-time single-input-single-output systems by applying the dynamic linearization (DL) technique. The ILC law is constructed based on the equivalent DL expression of an unknown ideal learning controller in the iteration and time domains. The learning control gain vector is adaptively updated by using a Newton-type optimization method. The monotonic convergence on the tracking errors of the controlled plant is theoretically guaranteed with respect to the 2-norm under some conditions. In the proposed ILC framework, existing proportional, integral, and derivative type ILC, and high-order ILC can be considered as special cases. The proposed ILC framework is a pure data-driven ILC, that is, the ILC law is independent of the physical dynamics of the controlled plant, and the learning control gain updating algorithm is formulated using only the measured input-output data of the nonlinear system. The proposed ILC framework is effectively verified by two illustrative examples on a complicated unknown nonlinear system and on a linear time-varying system.

Since they were first introduced by Arimoto, iterative learning control (ILC) methods have received lots of attention because of their effectiveness and briefness. In this paper, an open-closed-loop PID-type ILC scheme with bounded learning matrixes for the control of nonlinear time-varying systems are studied. Under a few reasonable restrictions that are very common, as in other ILC research papers on the convergence problem, sufficient conditions for guaranteeing the convergence of ILC systems are given and then proved with linear operator theory. Both the system and the ILC scheme described have very general forms. It is shown that the convergence condition is weaker than some known results for a similar ILC scheme.

Iterative learning control for nonlinear stochastic systems with variable pass length

Hydro-power production capacity prediction based on machine learning regression techniques

The classical problem setup of iterative learning control (ILC) is to enforce tracking of a reference profile specified at all time points in the fixed task duration. The removal of the time specification releases significant design freedom in how the path is followed but has not been fully exploited in the literature. This article unlocks this extra design freedom by formulating the ILC task description to handle repeated path-following tasks, e.g., welding and laser cutting, which aim at following a given “spatial” path defined in the output space without any temporal information. The general ILC problem is reformulated for ILC design with the inclusion of an additional performance index, and the class of piecewise linear paths is characterized for the reformulated problem setup. A two-stage design framework is proposed to solve the characterized problem and yields a comprehensive algorithm based on an ILC update and a gradient projection update. This algorithm is verified on a gantry robot experimental platform to demonstrate its practical efficacy and robustness against model uncertainty.

A framework is developed to construct computational models of the human motor system (HMS) using various iterative learning control (ILC) update structures. Optimal models of movement are introduced using a general cost function (involving both tracking objective and an additional constraint term), and its parameters are fitted to observed limiting solutions corresponding to learned human motion obtained from experiments. Three general ILC update structures are considered which each generate the required limiting solution using different forms of experimental data. It is shown how the parameters in each which govern convergence may also be fitted to experimental learning data, with the different ILC structures permitting varying degrees of freedom in capturing the observed learning transients. Experimental results in which a participant uses a planar robot to perform reaching tasks confirm the ability of the proposed ILC structures to accurately model the learning ability of the human motor system.

This paper deals with the Iterative Learning Control (ILC) of a Flexible Joint Experiment. The paper presents a short overview of the field of Iterative Learning Control and some basic ILC strategies (P – type and D – type) are highlighted. The design of ILC for linear systems is briefly presented. The ILC is implemented on a Flexible Joint experiment, working in repetitive manner, where the control goal is the trajectory tracking. Even the Flexible Joint is a nonlinear system, in order to tune the controller parameters a linear model is derived. The controller performances are evaluated using a real plant experiment. The main experimental results are presented and the performance of the control system is analyzed.

Iterative Learning Control (ILC) has recently emerged as a powerful control strategy that iteratively achieves a higher accuracy for systems with repetitive tasks. The basic idea of ILC is to construct a compensation signal based on the tracking error in each repetition so as to reduce the tracking error in the next repetition. In this paper, particle swarm optimization (PSO) is proposed to optimize the input of iterative learning controller. The experimental results confirm that the proposed method not only has higher tracking accuracy than that of Improved Genetic Algorithm (IGA) and traditional Genetic Algorithm based elisit strategy (EGA), but also has the advantages of simple algorithm and good flexibility. And compared with conventional iterative learning control methods, it is easy to solve the optimal input for non-linear plant models.

An iterative learning controller (ILC) is proposed for the tracking control of functional neuromuscular stimulation (FNS) system performing the given task repeatedly. A P-type ILC updating law assisted by a PD closed-loop controller is suggested for a simpler implementation. This kind of learning from repetitions of control strategy supplies strong robustness in the tracking control of uncertain time-varying FNS systems, which is essential for the adaptation and customization of FNS applications. Nonlinear muscle recruitment, linear muscle dynamics in force generation, and multiplicative nonlinear torque-angle and torque-velocity scaling factors are considered in the electrically stimulated muscle model used for the simulation studies. A one-segment planar system with passive constraints on joint movement is taken as the skeletal model. Simulation results indicate that the control scheme of this paper is promising for FNS system control.

Multiphase flow metering is a challenging task because of the complexity of multiphase flow. In this paper, nonintrusive multiphase flow metering techniques, including machine learning (ML) / artificial intelligence models for the identification of flow regimes and estimation of flow parameters of a two-phase flow in a horizontal pipe are proposed that use data from Electrical Capacitance Tomography (ECT) and conventional measurements such as differential pressure in the pipe. The flow regimes are classified into five types, namely plug, slug, annular, wavy and stratified. Two-phase air/water flow experimental data from ECT are collected by running extensive experiments using the horizontal section of the multiphase flow rig at the University of South-Eastern Norway (USN). Exploratory data analysis (EDA) is performed on these data to extract features for use in classification and regression algorithms. Time series of normalized capacitance data from ECT sensors are used to classify flow regimes and identify flow parameters. ML techniques of Artificial Neural Network, Support Vector Machine (SVM), K-Nearest Neighbors (KNN) and Decision Tree (DT) are used to classify flow regimes by using features extracted from ECT data. The cross-correlation technique is used to estimate flow velocity using data from a twinplane ECT module. ML regression techniques are used to estimate phase fractions. Fusing data from differential pressure sensors enhances the flow regime classification. An overall system performance is given with suggestions for designing dedicated control algorithms for actuators used in multiphase flow control.