Speed tracking and precise stop control of metro trains play a key role in train operation. Reinforcement learning (RL) is one of the methods that can solve train control challenges and adapt to various environments. However, when applying RL to train control, the input time delay of the train presents particular difficulties. This paper proposes a RL design method that integrates prediction techniques and the deep deterministic policy gradient algorithm to overcome the input time delay. The training and performance evaluation of the proposed RL is conducted using the validated Automatic Train Operation (ATO) simulator for Seoul Metro Line 5. The results demonstrate that prediction-based RL (PRL) controllers outperform RL without prediction controllers. Additionally, the PRL controller robustness is verified through performance comparisons with a classical model-based approach across various realistic scenarios.

Truck platooning is one of the most promising autonomous driving technologies with its commercialization on the horizon. However, to deploy heavy-duty platooning trucks on highways mixed with other vehicles, the platoon should be able to change lanes while guaranteeing road safety. However, most studies that have explored this challenge are simulation-based and thus do not deal with practical issues like the limitations of perception systems and the lane change control algorithms with actual prototype trucks. For a more advanced proof of concept, we present an implementation of scale trucks capable of lane changes while maintaining a platoon. Among recently proposed lane change protocols, a rear-first protocol is selected that can effectively block approaching vehicles for ensuring safe lane changes. The system is composed of (i) cooperative perception, (ii) Responsibility-Sensitive Safety (RSS)-based decision, and (iii) two-phase lane change control. The implementation is validated by realistic lane change scenarios on our proving ground.

Feedback control systems employing disturbance observers (DOB) have been shown to be effective in a wide range of applications. Motivated by the structure of conventional disturbance observers, we reformulate the design of disturbance observers as that of finding a solution to a specific system equation. The solution is derived using Willems’ fundamental lemma. Theoretical conditions are established to guarantee the validity of the formulation for linear time-invariant (LTI) systems. The effectiveness of the method is validated through numerical simulations and a DC motor experiment, demonstrating comparable performance to model-based disturbance observers. The proposed framework enables direct data-driven disturbance compensation without requiring explicit system identification.

This paper investigates the effect of output measurement noise on the robustness of one-step predictors in Subspace Predictive Control (SPC), a data-driven framework that constructs predictors directly from input-output trajectories. When measurement noise is present, the data matrix is corrupted in such a way that its rank is increased from that of the clean data matrix, leading to degraded prediction accuracy. To mitigate this, we examine two rank-reduction techniques—Low-Rank Approximation (LRA) and Generalized Total Least Squares (GTLS). Through simulations on six representative systems and 5000 randomly generated systems, we show that GTLS consistently reduces predictor error, especially when the minimum nonzero singular value of the clean data matrix is moderately small. Furthermore, the same trend holds for higher-order systems, indicating that the observed relationship generalizes beyond the second-order case. These findings reveal a strong empirical correlation between the smallest nonzero singular value and predictor sensitivity to noise, and demonstrate that GTLS enhances predictor robustness to measurement noise.

This paper presents a fault detection and toleration scheme for Unmanned Ground Vehicles (UGVs) with two position sensors and orientation sensors. Four representative types of sensor faults are considered: complete fault, bias fault, drift fault, and precision degradation. The proposed detection method consists of a Long Short-Term Memory (LSTM) Network Module, an Amplitude Difference Thresholding Module, and an Actuation Motion Coherence Module. A Husarion Rosbot 2.0 and VICON motion capture system compose a platform that is used to collect motion data for network training and experimental validation of the proposed scheme. Sensor fault detection performance is experimentally validated using a trajectory that was not included in the training data set. The fault detection accuracy is compared to other learning-based fault detection methods. Based on the fault detection result, we propose the fault toleration method.

Most of the existing studies on analyzing the productivity of serial production lines focus on cases where the coefficient of variation ( $CV$ ) for both uptime and downtime is less than 1. Hardly any result is available when $CV>1$ , i.e., uptime and downtime of machines exhibit high variability. The improvement of the production lines with high variable uptime and downtime depends on heuristic trial and error due to the lack of analysis method. This article suggests a neural network that approximates the throughput of serial production lines from machine and buffer parameters. Four neural network architectures (multilayer perceptron, recurrent neural network, long short-term memory (LSTM), and gated recurrent unit) are compared to determine the most effective architecture for the throughput approximation task. Training data are obtained from discrete-event simulations, encompassing a wide range of parameters. The results indicate that the LSTM model outperforms the other architecture considered. Furthermore, we present bottleneck identification and continuous improvement scenarios utilizing the model.

Over the past two decades, there has been a growing interest in control systems research to transition from model-based methods to data-driven approaches. In this study, we aim to bridge a divide between conventional model-based control and emerging data-driven paradigms grounded in Willem's fundamental lemma. Specifically, we study how input/output data from two separate systems can be manipulated to represent the behavior of interconnected systems, either connected in series or through feedback. Using these results, this paper introduces the Internal Behavior Control (IBC), a new control strategy based on the well-known Internal Model Control (IMC) but viewed under the lens of Behavioral System Theory. Similar to IMC, the IBC is easy to tune and results in perfect tracking and disturbance rejection but, unlike IMC, does not require a parametric model of the dynamics. We present two approaches for IBC implementation: a component-by-component one and a unified one. We compare the two approaches in terms of filter design, computations, and memory requirements.

Multijob production (MJP) is a class of flexible manufacturing systems, which produces different products within the same production system. MJP is widely used in product assembly, and efficient MJP scheduling is crucial for productivity. Most of the existing MJP scheduling methods are inefficient for multijob serial lines with practical constraints. We propose a deep reinforcement learning (DRL)-driven scheduling framework for multijob serial lines by properly considering the practical constraints of identical machines, finite buffers, machine breakdown, and delayed reward. We analyze the starvation and the blockage time, and derive a DRL-driven scheduling strategy to reduce the blockage time and balance the loads. We validate the proposed framework by using real-world factory data collected over six months from a tier-one vendor of a world top-three automobile company. Our case study shows that the proposed scheduling framework improves the average throughput by 24.2% compared with the conventional approach.

A widely used quadrotor model for designing controllers is based on rigid body dynamics, and it does not include the influence of wind, leading to several limitations. Namely, the quadrotor does not tilt when undergoing uniform linear motion; and it is not possible to account for energy dissipation due to wind and drag. To solve these two issues, we have added two types of aerodynamic drag due to wind into the quadrotor dynamics: propeller drag, and body drag. We conducted equilibria analyses of this model and observed that bifurcation occurs as the quadrotor body size is diminished. Through simulation, it was revealed that the model we propose overcomes the limitations of the rigid body model. Additionally, it has been demonstrated that the model we propose enables the calculation of energy dissipation due to wind, and the energy dissipation varies depending on the wind field distribution, even if the quadrotor’s trajectory is the same. This metric can serve as a cost function for a new path-finding problem, aiding in the more efficient operation of quadrotors.

An automated material handling system (AMHS) is a production line component responsible for transporting products from one machine to another for manufacturing processes. The AMHS also acts as a buffer that enhances overall productivity by reducing the dependency on individual machine operations. This paper introduces a buffer parameter optimization algorithm designed for advanced AMHS with the capability to control the speed of individual products. The buffer parameters targeted for optimization are buffer length (distance between machines) and transfer speed. The algorithm addresses each parameter separately through two distinct optimization problems. The buffer length optimization problem is formulated with the constraint of limited space assigned to the production system. On the other hand, the transfer speed optimization problem is formulated based on the constraints of network resources and hardware limitations. The proposed algorithm employs an aggregation method to evaluate the performance of the production systems analytically.

In-line metrology provides critical information for feedback and feedforward process control. In high-volume manufacturing, the fundamental question is: how fast, how frequent, and how accurate measurements should be made to satisfy control requirements. This paper develops a framework to study the tradeoff among the sampling rate, the delay, and the quality of measurements and the effect of these canonical factors on the variability of processes. As a consequence of this fundamental tradeoff, the relative value of virtual metrology with respect to real metrology can be quantified in the context of advanced process control.

This paper proposes a task reallocation method for multi Unmanned Grounded Vehicle (UGV) systems in sensor and actuator fault situations. The proposed method formulates the task reallocation problem as a Mixed-Integer Linear Programming (MILP) using the distance between UGVs and task priority. When a UGV sensor or actuator fault occurs, the proposed method solves the MILP and finds the optimal task reallocation solution. The solution ensures continuing high-priority tasks for the multi-UGV systems. The proposed method is validated in simulation and experiment.

Recent publication [1] presents modified quadrotor dynamics in the presence of wind based on blade element momentum theory. It focuses on modeling the effect of wind on the blades. However, interpretation of the modified dynamics has not been presented anywhere. This paper intends to provide insights of the modified dynamics through equilibrium studies and uniform velocity linear motion. Results are obtained by simulating the dynamics presented in [1] under the said scenarios.

This work develops a data-based construction of inverse dynamics for LTI systems. Specifically, the problem addressed here is to find an input sequence from the corresponding output sequence based on pre-collected input and output data. The problem can be considered as a reverse of the recent use of the behavioral approach, in which the output sequence is obtained for a given input sequence by solving an equation formed by pre-collected data. The condition under which the problem gives a solution is investigated and turns out to be L-delay invertibility of the plant and a certain degree of persistent excitation of the data input. The result is applied to form a data-driven disturbance observer. The plant dynamics augmented by the data-driven disturbance observer exhibits disturbance rejection without the model knowledge of the plant.

Data-enabled Predictive Control (DeePC) allows controlling dynamic systems soley based on its input/output data. This approach is based on behavioral theory, which guarantees precise prediction of the output for given input as long as the collected input data satisfy Persistency of Excitation (PE) condition and the system is linear time invariant. In practice, however, DeePC faces to control nonlinear dynamics and it is necessary to investigate whether there is a preferred way of collecting input and output data for DeePC besides the PE condition. This paper investigate the issue using an Automatic Train Operation (ATO) simulator that represents existing metro train control systems including time delays and nonlinearities. We implement DeePC using two different datasets to control metro train. Comparison and discussion are provided.

In ultrasound shear wave elastography (USWE), the elasticity of a small lesion is underestimated due to the wave reflection inside the lesion. This paper proposes using a deep neural network to compensate for the size effect without explicit size information. The deep neural network corrects the underestimation of the elastic modulus. The dataset for the training process consists of 4000 images with lesions of random sizes and elastic moduli, obtained from the k-Wave MATLAB Toolbox. A mean absolute error (MAE) is calculated between the ground truth and the image generated by the generator network for 500 test datasets with the unitary elasticity background. The maximum value of MAE is 0.0082, indicating that the generator network generates images that are similar to the ground truth in size and modulus. When the unitary elasticity background is replaced by the image of the breast phantom, the neural network proves to be effective in size effect compensation.

Velocity and position control for metro trains is typically achieved by classical control methods (PID, etc). Challenges in this control problem include imprecise position sensing, time delay, and external disturbances due to weight changes, curves, and slopes of the rails. In order to achieve acceptable stop position of the trains at each station, the controller design often involves individual gain tuning for each sections in the route, which consumes much time and effort. As a means to reduce the effort, reinforcement learning approach is looked into for train control. Automatic Train Operation (ATO) simulator capable of realistic simulation of train dynamics along the Line 5 in Seoul Metro is used to investigate the feasibility of this approach. Results are discussed from the perspective of practicality.

The evaluation of cardiac anisotropic mechanics is important in the diagnosis of heart disease. However, other representative ultrasound imaging-based metrics, which are capable of quantitatively evaluating anisotropic cardiac mechanics, are insufficient for accurately diagnosing heart disease due to the influence of viscosity and geometry of cardiac tissues. In this study, we propose a new ultrasound imaging-based metric, maximum cosine similarity (MaxCosim), for quantifying anisotropic mechanics of cardiac tissues by evaluating the periodicity of the transverse wave speeds depending on the measurement directions using ultrasound imaging. We developed a high-frequency ultrasound-based directional transverse wave imaging system to measure the transverse wave speed in multiple directions. The ultrasound imaging-based metric was validated by performing experiments on 40 rats randomly assigned to four groups; three doxorubicin (DOX) treatment groups received 10, 15, or 20 mg/kg DOX, while the control group received 0.2 mL/kg saline. In each heart sample, the developed ultrasound imaging system allowed measuring transverse wave speeds in multiple directions, and the new metric was then calculated from 3-D ultrasound transverse wave images to evaluate the degree of anisotropic mechanics of the heart sample. The results of the metric were compared with histopathological changes for validation. A decrease in the MaxCosim value was observed in the DOX treatment groups, with the degree of decrease depending on the dose. These results are consistent with the histopathological features, suggesting that our ultrasound imaging-based metric can quantify the anisotropic mechanics of cardiac tissues and potentially be used for the early diagnosis of heart disease.

An automated material handling system (AMHS) is part of a production system that transports products from one machine to another for manufacturing processes. While conveyor belts have been commonly employed as AMHS, the increasing demand for enhanced performance has led to the emergence of advanced AMHS based on linear motors. A feature in these advanced AMHS is that they often allow non-uniform transfer speed setup, i.e., transfer speeds are allowed to be different by individual sections within the production system. This characteristic introduces an opportunity in determining the optimal transfer speed vector for the maximum productivity. In this paper, we propose an algorithm for optimizing the transfer speed vector using an analytical throughput evaluation method. The effectiveness of the algorithm is demonstrated through an example involving five machines.