To investigate motion sickness caused by mismatched visual and operational information in the integration of autonomous driving and virtual reality technologies, this paper simultaneously collects electroencephalogram (EEG) signals from participants using a dual-task paradigm that combines active driving and autonomous driving on a simulated driving platform. This approach is complemented by the Go/No-go behavioral paradigm and standardized motion sickness questionnaires to explore the impact of different driving modes on the allocation of brain cognitive resources. Results indicate that autonomous driving scenarios significantly exacerbate motion sickness symptoms due to visual-vestibular conflict. Autonomous driving based on virtual reality is particularly prone to inducing motion sickness. The underlying neural mechanisms are characterized by increased power spectral density in the Pz, Cz, and Fz EEG channels (p<0.05), as well as decreased amplitude and shortened latency of the N200 and P300 components (p<0.05). Furthermore, a convolutional neural network classification model is constructed that integrates time-domain ERP, frequency-domain PSD, and nonlinear complexity features. The model achieves an accuracy of 92.7%, which provides a scientific basis for real-time monitoring and the optimization of human-computer interaction design.

In order to explore the underlying mechanisms of driver distraction in turning and straight driving scenarios, this study uses a driving simulator to create straight-driving and turning virtual scenarios. It also collects driving performance and eye-movement data of drivers in different driving states. The KNNImputer algorithm is employed to handle missing data during data collection. Then, a paired samples T test is used to analyze significant differences and extract significant difference feature indexes from sample data with a time window of 1 s length and 75% overlap. Based on these features, an XGBoost classifier is used to build cognitive distraction recognition models for different scenarios. The results show that compared with straight driving, drivers in turning scenarios have higher mental workload, indicated by lower pupil diameter change frequency, higher saccade speed and higher fixation duration percentage. The built cognitive distraction recognition model achieves an accuracy of 91.30% for straight-driving and 83.28% for turning scenarios. This suggests that cognitive distraction behavior in turning scenarios is more dangerous and harder to recognize.

With regard to the driving environment and safety of sanitation vehicle drivers, this paper proposes a driver fatigue detection method based on an enhanced YOLOv8n algorithm. Specifically, FasterNet is employed to replace the backbone network of the YOLOv8 object detection algorithm, resulting in the design of a lightweight FasterNet-YOLO network model. To preserve critical feature information from the input feature map, Squeeze-and-Excitation (SE) modules are integrated into the backbone network, while Convolutional Block Attention Modules (CBAM) are added to the neck network. Additionally, the Zero-DCE++ algorithm is introduced to enhance the brightness of video streams captured by cameras, addressing the issue of insufficient brightness in the driver’s face that hinders accurate detection. Experimental results demonstrate that the proposed method achieves an average precision of 98% (mAP@0.5) at an intersection over union ratio of 0.5, with an average inference time per frame reduced to 6.95 ms. This approach can effectively monitor the driver’s fatigue state in real-time under varying lighting conditions.

In order to enhance accuracy of driving fatigue detection, this paper takes drivers’ physiological signal pulse wave as the data source, introduces hemodynamic-based blood pressure waveform features based on the extraction of Heart Rate Variability (HRV) features. Moreover, a feature indicator set that can effectively characterize driving fatigue is constructed, and a three-classification model of driver fatigue is constructed based on ensemble learning. Then, a resampling method is introduced in the data preprocessing stage, and the effects of different sampling methods on the detection performance of the model are contrasted. Test results show that multidimensional feature fusion of pulse signals can significantly improve the detection accuracy of driver fatigue by 24.68 percentage points on average in all scenarios compared with the method of using only HRV features; resampling can further enhance the detection performance of the ensemble learning model, and the model achieves the best detection performance in a scenario with a sampling window width of 2 min, a sampling window overlap of 80%, and a fusion of HRV features with pulse waveform features.

In order to effectively extract interaction features among vehicles in high-speed traffic scenarios, thus accurately predict the trajectories of dynamic obstacles, this paper proposes a multi-vehicle interaction trajectory prediction model using the coding-decoding framework based on the graph spatial-temporal attention mechanism. The vehicle-to-vehicle graph interaction field is established by combining the repulsive force field and the graph model, the node feature matrix and the adjacency feature matrix are used to characterize the dynamic interaction between the vehicle and the surrounding vehicles, and the deep spatial-temporal interaction features are extracted by the graph spatial attention and temporal polytope attention to obtain the graph spatial-temporal fusion coding features. The one-hot encoding of the longitudinal and lateral behavior intentions of the vehicles is concatenated with the encoding to achieve multimodal trajectory prediction for the target vehicles. Validation using the NGSIM dataset shows that, compared with 6 other models, the proposed model achieves the lowest RMSE and NLL values. Ablation experiments further validate the effectiveness of the graph interaction field, demonstrating that the model can significantly improve the accuracy of vehicle trajectory prediction.

In order to solve the problem of inaccurate subjective evaluation of vehicle driving and the inability of objective evaluation to reflect subjective feelings, an evaluation model based on Stacking ensemble learning method is proposed. First, the acceleration condition characteristics of vehicles are studied and objective evaluation indicators of driverability are defined. Then the evaluation indicators are used as input features to train the Stacking ensemble model. Moreover, the Improved Harris Hawk Optimization algorithm was used to optimize the hyperparameters in the Stacking ensemble model to improve the model prediction performance. Finally, the road test proves that the performance of the HHO-Stacking ensemble model is superior to that of a single machine learning model. The qualification rate of the HHO-Stacking ensemble model is 95%. The HHO-Stacking ensemble model can complete the drivability evaluation more effectively.

Aiming at the issues of low high-speed tracking accuracy and weak robustness in the trajectory tracking control of distributed electric drive bearing platforms, a hierarchical lateral motion control strategy based on desired front wheel angle tracking is designed. Taking the distributed electric drive bearing platform as the research object, a comprehensive dynamic model integrating rubber wheels, vehicle body, and electric drive modules is constructed based on the dynamic analysis of each subsystem and the interaction relationship between the wheels and the ground. By constructing a hierarchical motion control strategy with upper-level Model Predictive Control (MPC) trajectory tracking and lower-level steering motor angle control, high-precision control of the platform’s lateral position can be achieved. A holistic dynamic simulation model of the bearing platform is built using Simulink. The simulation results show that the lateral motion control strategy designed in this research can achieve multi-scenario trajectory tracking with high precision at various speeds. Compared with the sliding mode controller, the control accuracy of this strategy is improved by 33%, and the control stability is significantly enhanced.

In order to solve the problem of clutch overheating during the launching of a Dual Clutch Transmission (DCT) vehicle in the Drive mode of D1 gear under extreme working conditions, this paper proposes a dual-clutch launching control method to improve the climbing and escape performance of the vehicle. Based on the analysis of the hardware structural conditions, the clutch plate temperature model and the specific process of DCT launching, this paper proposes to use the D2 gear clutch to undertake the useless slipping power at the beginning of the steady speed stage of launching, and then switch back to the D1 gear clutch when the engine torque gradually increases to exceed the threshold, the paper then elaborates in detail the activation conditions and specific implementation process of the control method. Vehicle test verification shows that the dual-clutch launching has obvious advantages in clutch temperature control and driving smoothness compared with the single-clutch launching, and the vehicle climbing and escape performance has been significantly improved.