To systematically review the safety risks of autonomous driving and promote its safe market-oriented application, this paper summarizes and analyzes the characteristics of current autonomous vehicle accidents and relevant research from selected developed countries, identifying the core factors and deep-seated challenges affecting the safety of autoromous driving. The study reveals that the safety landscape encompasses 4 key domains: data security, technological safety, transfer of control between human and mahine, and risk control. Furthermore, future development faces profound challenges including the algorithm "long-tail problem", a crisis of trust in human-machine interaction, data security vulnerabilities, and regulatory lag. The research proposes that coordinated advancement across 4 dimensions—technological integration, regulatory framework construction, human-centered design, and data protection—is essential to establish a collaborative governance framework. This provides a systematic reference for the safe development and application of autonomous driving technologies.

Autonomous Emergency Braking (AEB), as a core safety function of Advanced Driver-Assistance Systems (ADAS), faces 3 major certification challenges: perception limitations, verification complexity, and human-machine collaboration. This paper systematically analyzes the technical architecture of AEB, including multi-sensor fusion for robust environmental perception, conservative decision-making logic to ensure intervention safety, and Operational Design Domain (ODD) definition for user expectation management. In terms of certification frameworks, the study explores the dual-pillar model based on ISO 26262 Functional Safety (FS) and ISO 21448 Safety of the Intended Functionality (SOTIF), highlighting the shift from regulatory testing to scenario-library-driven virtual validation. For human-machine collaboration, it examines mechanisms for establishing calibrated driver trust through transparent interaction design, and prospects the potential of neuro-symbolic AI in enhancing interpretability and localized innovation for China's complex traffic scenarios. This research provides a systematic reference for the industrial safe deployment of AEB systems.

To gain a deeper understanding of driving behavior and its influencing factors, experiments are conducted under various environmental conditions. Data such as driving videos and vehicle status are collected, and the NASA-TLX scale is used to obtain drivers'subjective evaluations. The study analyzes the variations in vehicle speed and driver responses under different driving conditions, as well as the influence of drivers'subjective cognition on the execution of driving behaviors. The results indicate that when vehicles pass through intersections during a yellow light or during overtaking, the speed variation exceeds 50%, while during car-following, the vehicle speed continuously fluctuates within the range of 30~50 km/h. These conditions tend to cause drivers to feel tense and irritable, with subjective ratings reaching up to 500 points. A significant negative correlation was found between vehicle speed and subjective evaluation scores, indicating that both can serve as effective indicators for assessing the state of drivers during the execution of driving behaviors.

In order to improve the testing efficiency and comprehensiveness of the Adaptive Cruise Control (ACC) system, this paper proposes a test scenario generation method based on the Deep Deterministic Policy Gradient (DDPG) algorithm. Firstly, typical driving scenarios are designed, and a test scenario model is constructed, which includes scenario initialization, the definition of state and control variables, the agent-environment interaction mechanism, and a reward function. Then, an agent is designed based on a reinforcement learning algorithm, and generate challenging test scenarios. a reward function integrating exposure rate, adversarial factors, and unreasonable collision penalties is developed, guiding the agent to generate scenarios that are not only highly risky but also consistent with real traffic rules. Finally, simulation experiments on left/right cut-in and cut-out scenarios are conducted to validate the effectiveness of the proposed method. The results demonstrate that this approach can automatically generate critical hazardous scenarios for ACC, improve test coverage and specificity, and provide significant value for ensuring system safety testing.

To improve the fusion accuracy of multi-modal data in the field of autonomous driving this paper proposes a spatiotemporal synchronization technology techinal scheme for Light Detection and Ranging (LiDAR) and camera. By adopt a method of estimating and adjusting the offset, it calulates the time of the trigger for the camera exposure, and adjusts the offset time of the trigger for the camera exposure thereby controlling LiDAR and camera to scan and the same target at the same time. The results show that the proposed scheme significantly improves the fusion accuraly of poirt cloud data and images, verifying the feasibility and effectiveness of the technical scheme.

This paper aims to enhance the effectiveness of risk management in the entire vehicle development process. Based on a Sino-foreign cooperative entire vehicle research and development project, the study analyzes the characteristics of the low-voltage harness in vehicles, identifies and categorizes risk factors, and utilizes fault tree analysis to model the causal relationships of various risks. These relationships are then mapped into a Bayesian Network (BN). By employing the fuzzy set theory, collected expert opinions are transformed into prior and conditional probabilities of node events, establishing a risk assessment model for vehicle low-voltage harness development based on fault tree-Bayesian network. This model quantifies the probability and impact of risk factors. For key and sensitive risk factors, corresponding risk response strategies are proposed, providing effective risk management for harness development.

To clarify the regulatory mechanism of temperature on the performance of narrow-range oxygen sensors, this paper takes the thickness of the heater resistor as the regulatory carrier, systematically investigates its influence on the operating internal resistance and operating temperature of the sensor, and explores the correlation characteristics between temperature and performance by combining EDS, ICP composition analysis and performance tests such as response time and output voltage. The results show that there is a positive correlation between the thickness of the heater resistor and the operating internal resistance of the sensor; as the resistor thickness increases, the operating internal resistance gradually decreases, which in turn causes the operating temperature of the sensor to rise. Elevated operating temperature is the core inducement leading to the performance degradation of narrow-range oxygen sensors. On the contrary, reducing the thickness of the heater resistor can optimize the sensor performance, which is specifically manifested in the significant improvement of the rich-burn voltage output and the obvious shortening of the response time under rich-lean combustion and lean-rich combustion conditions.

To achieve lifetime prediction of power modules under actual operating conditions, a power module lifetime prediction method based on a temperature model is proposed. Simulation parameters such as voltage, phase current, power factor, and switching frequency are obtained from the vehicle driving cycle data to calculate module losses. First, the Foster thermal network model is then used to calculate the junction temperature, thereby obtaining the module temperature variation sequence. The rainflow counting method is applied to extract the junction temperature fluctuation sequence. Next, the damage degree of the power module under the corresponding driving cycle is evaluated by combining the lifetime model fitted from power cycling tests. Additionally, the damage degrees of the same module under power generation and electric driving conditions are compared to analyze the performance requirements of the module in different application scenarios. The results show that under a 240 000 km driving cycle, the damage degree in electric driving conditions is significantly higher than in power generation conditions, and an increase in switching frequency exacerbates the damage. This method can effectively support rapid reliability assessment of power modules and provide a basis for module selection and optimization in different application scenarios.