The effectiveness of existing testing methods is under scrutiny due to significant limitations, notably the discrepancy between action distributions and actual distributions caused by inadequate environmental understanding. Additionally, the lack of a scheme prioritizing fidelity while coordinating testing efficiency results in considerable divergence between test and natural scenarios. To address these issues, this paper proposes a Directed Graph Reinforcement Learning approach with action constraint optimization (DGRL) to generate critical scenarios, balancing efficiency and fidelity. By incorporating directed graph convolutional networks, the model encodes environmental states within the observation interval, providing spatiotemporal insights and reducing action distribution discrepancies. Furthermore, it constructs an unbiased estimation reward considering action constraints by sampling action confidence intervals, filtering out distorted actions, thus balancing efficiency and fidelity. DGRL was trained using the highD dataset, demonstrating robust acceleration performance, with interquartile range ($IQR$) of 3.1 and first quartile ($Q_{1}$) of 75.6 in the acceleration ratio distribution, representing the bandwidth and baseline, respectively. The model also achieved high fidelity, with scenario discrepancies compared to natural scenarios reduced by 85.4% ($Q_{1}$) and 46.5% ($IQR$) relative to GAIL, which considers the rationality of driving behavior. Here,$Q_{1}$represents the baseline of scenario discrepancy, and$IQR$denotes the distribution bandwidth of scenario discrepancy. Additionally, there was 69.7% reduction in the upper limit of the 95% confidence interval, indicating a significant decrease in maximum scenario discrepancy. Deployment on an intelligent connected hardware-in-the-loop testing platform validated DGRL’s effectiveness and applicability in real-world.

Precise control of vehicle path tracking is essential for ensuring the safe operation of autonomous vehicles (AV s), AV s equipped with four-wheel independent steering and drive (4WIS-4WID) experience continuous uncertainties during path tracking, such as parameter fluctuations, modeling errors, and external interference. Inadequate management of these uncertainties can result in deviations from the intended path or instability in AV s. This study employs backstepping and optimization method to develop a tailored nonlinear and robust path tracking controller. The controller coordinates the steering of four wheels and applies an additional yaw moment, facilitating stable tracking of the reference path and yaw angle. Such coordination facilitates the gradual convergence of the tracking error to zero, minimizes overshoot, and enhances vehicle driving comfort. Furthermore, by obtaining the feedback matrix through solving a quadratic programming (QP) problem, the controller can explicitly account for the executor constraints, ensuring Lyapunov stability. The simulation results confirm the efficacy of the proposed method and illustrate its superiority over similar approaches.

Advanced control and safety systems are crucial for electric vehicles, and the accurate estimation of the tire-road friction coefficient (TRFC) is crucial for developing effective safety control strategies. The hybrid data-mechanism model (HDMM), introduced in this paper, addresses the performance challenges posed by the inaccuracies of physical models and the limited interpretability of data-driven models in tire force estimation for TRFC estimation. Tire dynamics often exhibit transient responses, while mechanism-based models(MBM) typically reflect steady-state characteristics. Neglecting transient characteristics leads to a decrease in model accuracy. A neural network is used to learn the transient response characteristics of tire dynamics. These characteristics are then integrated with the steady-state tire forces from MBM to estimate the lateral and vertical forces acting on the wheel. The estimated tire forces serve as virtual measurements to calibrate parameters in the TRFC estimator, based on the Unscented Kalman Filter (UKF). During real-world vehicle tests, the proposed method reduced the Mean Error (ME) in lateral and vertical forces by 1271.85 N and 996.7 N, respectively, compared to the estimated tire forces from MBM. Additionally, the estimated TRFC converged to the reference value approximately 40ms earlier than the result from the MBM, with an estimated deviation within 0.1.

Motion prediction of surrounding vehicles in complex environments is essential for autonomous vehicle trajectory planning. Accurate motion prediction requires accurately estimating the state information of the surrounding vehicles. For this purpose, a moving horizon estimation with interacting multiple model (IMM-MHE) algorithm is first proposed here. The algorithm can match multiple vehicle maneuvers, but also fully utilizes the historical information obtained during the driving process, achieving a high estimation accuracy. Second, a moving horizon estimation with variable structure interacting multiple model (VSIMM-MHE) framework is designed. Time-domain adaptation is proposed to solve the problem that the fixed time domain of some models cannot be filled due to model activation and elimination. A new interaction method is proposed to solve the problem that models cannot interact because the starting timesteps of their time domains are different. The proposed framework reduces not only the computational burden, but also the final estimation error caused by the model not matching the current maneuver. Third, based on a model set consisting of different kinds of intention models, a VSIMM-MHE algorithm is proposed. This algorithm introduces residual information into the model classification method, reducing the dependence on the accuracy of the model probabilities. It can not only accurately estimate the state information of surrounding vehicles in a complex environment, but also identify the model that best matches the current maneuver and effectively predict the motion trajectories of surrounding vehicles through model probabilities. Finally, joint simulation with SCANeR studio, Carsim and Simulink and hardware-in-the-loop experiment demonstrate the effectiveness of not only the two proposed estimation algorithms but also the motion prediction of surrounding vehicles using the model probabilities in the VSIMM-MHE algorithm.

Driver distraction is a significant factor leading to traffic accidents. Detecting driver distraction is crucial for the development of advanced driver assistance systems (ADAS). With the development of deep learning techniques, advanced computer vision technologies have been continuously applied for driver distraction detection. To date, most distraction detection approaches cannot well be adapted to the distraction behaviors that are not included in the training dataset. To address this problem, we propose a lightweight driver distraction detection method using semisupervised contrastive learning. Unlike other studies that rely on large-scale models, a lightweight vision transformer with convolutional neural network (CNN) obtained by knowledge distillation is adapted to extract features, and the design of the dual-stream backbone network increases the generalization ability without increasing the computational burden. Furthermore, the combination of three-level attention mechanisms (i.e., channel-level, spatial-level, and batch-level) enhances the representative power of the model. Both depth and RGB datasets are used to train and test our proposed method. The experimental results show that our method shows superior performance in comparison with other state-of-the-art methods. Its lightweight architecture is suitable for practical applications. This study contributes to the development of ADAS and provides a new perspective on driver distraction detection.

This paper proposes an integrated planning and control method for autonomous ground vehicles in dynamic environments. This method takes the dynamic characteristics of large vehicles in consider and is able to autonomously avoid obstacle vehicles during global path tracking without the need for additional planning modules. Firstly, we establish a vehicle dynamics model and tire model to describe the complex dynamic response characteristics of the vehicle. Then, the planning and control problem is formulated as a multiple-constraint model predictive control (MPC) problem to achieve optimal decisions within the predictive horizon. A control barrier function (CBF) is designed as a constraint for the optimization problem to achieve safety-critical obstacle avoidance. Additionally, for dynamic traffic scenarios, an adaptive rule for control barrier function is designed to reduce the impact on surrounding vehicles. Finally, the effectiveness of this method is verified in multiple simulation scenarios.

Under extreme conditions, such as low friction surfaces, violent steering and urgent acceleration/deceleration, vehicle states change rapidly and are influenced by nonlinear and coupled vehicle dynamics. To improve vehicle stability under extreme conditions, a hierarchical control strategy is proposed for DDEVs. In the upper layer controller, a combined-slip tire model is adopted to improve the model accuracy under extreme conditions. A nonlinear model predictive control based controller is then designed to generate the desired tire slip ratios with the main objectives of tracking the desired yaw rate and suppressing the lateral velocity and tire slip ratios. In the lower layer controller, the disturbance on the driver’s torque requirement, which is disregarded by existing studies, is taken into account. Next, a linear predictive controller is designed to track the desired tire slip ratios by adjusting the motor torques. To improve the computational efficiency of the nonlinear predictive controller, a PMP-based, fast solving algorithm is proposed. The effectiveness of the proposed solving algorithm is checked by comparing the control performance with IPOPT. The proposed control strategy is evaluated by a series of HIL experiments. The HIL results show better performance in overall stability improvement and minimize the disturbance on the driver’s torque requirement.

Existing hierarchical planning and control architectures can cause the upper-level target trajectory passed to the lower-level tracking controller to be too conservative or impossible to track on slippery roads. To solve this problem, this paper proposes a new simultaneous planning and control scheme that determines the control inputs without explicit path planning and requires only information about the control objectives and safety constraints. First, we establish the vehicle stability boundary and the safety distance constraints to ensure that the vehicle avoids drifting and collisions on slippery roads. Moreover, real-time adaptive model predictive control (MPC) with online model linearization is designed to approximate the nonlinear programming as a quadratic program (QP), which allows the use of fast convex optimization tools. The steering angle and longitudinal acceleration are thus obtained. Finally, we design the controller to convert the longitudinal acceleration into an actuatable drive torque to avoid tire skidding on slippery surfaces. The simulation results show that under the conditions of low adhesion road and$\mu $-split roads, the proposed algorithm makes the sideslip angle of the vehicle within 1 degree. In contrast, the sideslip angle of the hierarchical algorithm reaches 6 degrees, and the vehicle has a noticeable drift. The proposed algorithm dramatically improves stability and driving comfort.