Motion control of autonomous electric vehicles (AEVs) faces severe challenges due to significant uncertainties introduced by dynamic environments, which may lead to potential safety issues. To address this problem, this paper proposes a meta-learning-enhanced online adaptive control method for AEVs to realize robust and high-precision motion control. First, a novel Meta-Learning-based Online Adaptive (MLOA) modeling approach is introduced, which enables rapid online adaptation of vehicle dynamics through few-shot learning combined with real-time operational data. This approach effectively captures dynamic behaviors in previously unseen tasks. Furthermore, the MLOA model is integrated into a Stochastic Model Predictive Control to enhance control adaptability and responsiveness under various conditions. Meanwhile, chance constraints are incorporated to handle random disturbances, thereby strengthening the robustness of the control strategy. The proposed method is validated through both simulations and real-vehicle experiments. Results show that the controller adapts within 1.8 s in previously unseen tasks and achieves up to 72.6% reduction in lateral tracking errors compared to the baseline method, and maintains an average computation time of only 0.0148 s per control step. These findings confirm the proposed method’s ability to maintain high trajectory tracking accuracy, fast response, and real-time feasibility under uncertainties and highlight the effectiveness of combining meta-learning with optimal control, providing a robust and adaptive control framework for autonomous driving in diverse and complex environments.

The malicious cyberattack in connected automated vehicles leads to a major threat to the safety and security of autonomous driving. Different from traditional approaches which generally require perfect knowledge of system models and state measurements, in this article, a novel learning-based adaptive resilient control framework is proposed to defend against various false data injection attacks on the steering system to improve safe driving for automated vehicles. First, a robust nonlinear state estimation method is developed to provide accurate observation of unmeasured state variables for feedback control with limited onboard sensors. Then, an active model learning approach is proposed to present the vehicle driving behavior under different attacks to improve the dynamic model for state prediction over receding horizon. Finally, a data-driven attack-resilient method is designed to optimize the vehicle motion for autonomous driving. The derived control policy can be adapted to different scenarios for safety. MATLAB/Simulink and CarSim co-simulation platform is applied to evaluate the effectiveness and robustness of the proposed method on state estimation, model learning and accurate tracking control under various attack conditions.

Presents corrections to the paper, (Corrections to “Learning-Based Optimal Adaptive Resilient Control for Safe Autonomous Driving Under Cyberattacks”).

Accurate and stable localization system plays a significant role in safe driving for connected automated vehicles (CAVs). However, the vulnerability from GPS spoofing attacks undermines the security of the localization system, posing great challenges for autonomous driving. In this article, a security-critical study for anomaly detection and defense against GPS attacks for CAVs using vehicle-to-vehicle (V2V) technology is explored to improve the localization for driving safety under cyber-attack. First, a robust learning-based GPS stealthy attack model is designed to generate spoofing GPS signals, which can evade currently widely applied Kalman filter-based localization with$\chi ^{2}$anomaly detector and result in vehicle positioning errors, leading to more potential driving hazards than traditional models. Then a novel detection method for GPS anomaly with V2V communication based on density clustering algorithm is proposed to detect the wrong GPS data effectively. When the GPS attack is detected, the vehicle position is estimated accurately by an innovative cooperative localization (CL) approach with multi-information fusion from neighboring vehicles to defend against the GPS attack. The proposed framework is evaluated with three real-world driving data sets in closed-loop simulation. The results show that the developed attack detection method has the best performance compared to the state-of-the-art methods in terms of detection accuracy and detection timeliness. Furthermore, the CL for attack defense can provide accurate position estimation for the victim vehicle against GPS attack to realize safe autonomous driving, illustrating the effectiveness and robustness in various driving scenarios.

Current trajectory prediction methods for autonomous vehicles commonly rely on deep neural networks, which are vulnerable to adversarial attacks. To enhance the security of trajectory prediction, this paper proposes an anomaly detection method based on generative adversarial networks. Firstly, a novel unsupervised anomaly detection model is proposed, taking into account both temporal and spatial features of trajectories with Long Short-Term Memory. The networks are trained using max-min game theory between the generator and the discriminator to capture the normal driving feature distribution. Furthermore, trajectory data is mapped to the latent space, and the generator reconstructs data from the latent space to compute reconstruction loss, while the discriminator detects trajectory data to calculate discrimination loss. Finally, anomalies are detected using an anomaly score that represents the extent to which the data point deviates from normal behavior and determines whether the trajectory of this segment is anomalous within the time window. We evaluate the method on three public datasets, and experimental results demonstrate its excellent performance under adversarial attacks.

Accurate trajectory prediction is crucial for autonomous vehicles to realize safe driving. Current trajectory prediction approaches generally rely on deep neural networks, which are susceptible to adversarial attacks. To evaluate the adversarial robustness and security of deep-learning-based trajectory prediction models, this paper proposes an adversarial attack method on trajectory prediction using generative adversarial networks (GANs). First, a novel LSTM-based attack trajectory model named Adv-GAN is proposed considering both the temporal and spatial driving features. The networks in Adv-GAN are trained through game learning between the generator and the discriminator to obtain the adversarial trajectories with real driving feature distribution. Furthermore, the generated trajectory is optimized with the vehicle kinematics model for driving feasibility on roads. The derived adversarial attack can lead to considerable deviations in trajectory prediction which affects driving safety for autonomous vehicles. We evaluate the proposed Adv-GAN on three public datasets, and experimental results show the effectiveness with better attack performance compared to a state-of-the-art adversarial attack model.

Connected automated vehicles (CAVs) share information through vehicular networks; however, cyber-attacks on GPS may cause significant challenges to compromise vehicle security and driving safety. In this paper, a novel approach for GPS attack detection and mitigation is proposed using vehicle-to-vehicle (V2V) communication, which enables vehicles to access and utilize accurate location information for autonomous driving. Instead of directly fusing the location data received from other vehicles, a trust evaluation process with a $\chi$2-detector is developed to identify and isolate potential malicious surrounding vehicles that may send erroneous information into the V2V network. Subsequently, a Bayesian approach is employed to fuse data from GPS, inter-vehicle distance, and bearing angle measurements. A real-time Robust-Random-Cut-Forest based detector is constructed to identify possible GPS attacks for an ego vehicle. When a malicious attack is detected, a novel cooperative positioning method is used to mitigate the impact of the GPS attack based on V2V information. Simulation results demonstrate the performance of the proposed approach in detecting GPS attacks timely and improving the positioning accuracy and robustness of CAVs under different attacks.

In this paper an ecological adaptive cruise controller to reduce the fuel consumption and ensure the safe inter-vehicle distance for vehicles with step-gear transmissions is presented. An optimal control strategy using reinforcement learning with a novel actor-gear-critic architecture is proposed to obtain the continuous traction force trajectory and the discrete gear shift schedule. The traction force is determined from an actor network to maintain a desired inter-vehicle distance which improves the driving safety in a car-following process. The gear shift schedule is derived from a gear network to reduce the fuel consumption. The control strategy is model-free and allows continuous online learning for different driving situations without look-ahead velocity information. Particularly the nonlinear vehicle dynamics, the nonlinear transmission efficiency map for different gear ratios, and the nonlinear fuel consumption map are learned for fuel consumption reduction. The proposed controller is evaluated for different driving scenarios to demonstrate its robustness. Furthermore simulation comparisons for different gear shift schedules and velocity trajectories are given underling the advantages in terms of fuel economy and driving safety.

In this paper, an ecological adaptive cruise controller (ECO-ACC) for parallel hybrid electric vehicles (HEVs) in a car-following scenario is presented to improve the fuel economy and to maintain a desired inter-vehicle distance from the preceding vehicle. An ACC based on action dependent heuristic dynamic programming (ADHDP) is proposed to obtain an ecological velocity profile and realize an active distance control in normal driving situations. ADHDP is able to adapt internal parameters online and can thus deal with systems with disturbances. Furthermore, an adaptive energy management strategy for HEVs is introduced to control the gear shift and power split for fuel consumption optimization. The gear shift command is designed by enumeration, and the power distribution between the engine and the electric motor is performed by ADHDP. The developed ACC and energy management strategy are finally combined to an ECO-ACC to achieve a multi-objective optimization. Only the current velocity and acceleration of the preceding vehicle are used while knowledge about the future velocity is not needed. The simulations of different driving cycles indicate that the ECO-ACC can lead to near-optimal fuel economy and comfortable driving.