Research on Takeover Safety of Intelligent Vehicles Based on Accident Scenarios in Real-Vehicle Testing

Pingfei Li, Meiling Zhou, Chang Xu, Li He, Wenhao Hu, Zhengping Tan, Lingyun Xiao, Xiaojun Mou, Hao FengSensors 2025 · DOI

TL;DR

This paper investigates the critical safety risks of intelligent vehicles during takeover maneuvers in challenging scenarios like sharp curves and static obstacles. Through real-vehicle testing, it quantifies the minimum driver reaction times needed for safe control, highlighting the urgent need for enhanced driver engagement and robust early warning systems in current autonomous driving technology.

What Problem Does This Paper Address?

The rapid advancement of intelligent vehicles, particularly the transition to Level 3 (conditional automation), introduces a complex human-machine cooperative driving (HMCD) phase. While these systems can handle driving for extended periods, they exhibit functional limitations in specific scenarios. When combined with driver distraction, often stemming from over-reliance on automation, these limitations can lead to novel accident patterns. Traditional accident investigation methods struggle with these new complexities. This research directly addresses the critical safety challenge of ensuring drivers can safely and effectively retake control of the vehicle when the autonomous system encounters its operational limits, especially in high-risk situations that mirror real-world accident scenarios.

Methods

The study employs a rigorous real-vehicle testing methodology to simulate two representative accident scenarios:

  • Lateral Control Challenge: Large-curvature curves.
  • Longitudinal Control Challenge: Static obstacles.

These scenarios were designed based on analyses of actual traffic accidents involving Level 2 advanced driver-assistance systems (ADAS). The experiments were conducted at a closed test facility using two different Level 2 combined assisted driving vehicles. Data collection involved a high-precision acquisition system capturing vehicle kinematic data, IMU/RTK for motion status and position, and synchronized video from onboard cameras. The core of the analysis focused on measuring driver takeover time (TOT) and takeover reaction time (TORT) under various conditions, including different speeds and curve radii for lateral control, and various obstacle types for longitudinal control. The study also defined specific criteria for when a driver takeover was considered to have occurred, particularly in cases where the system failed to issue a warning.

Datasets

The study did not introduce new datasets for training. Instead, it utilized real-world accident data to inform the design of its experimental scenarios. The experimental data itself was collected during the real-vehicle tests conducted at the Chengdu Sino-German Intelligent Connected Vehicle Sichuan Test Base.

Key Insights and Ideas

  • "Passive Vigilance" is a Major Hazard: The research underscores how drivers, lulled into a false sense of security by automation, can become passively vigilant, reducing their situational awareness and readiness to intervene. This mental complacency, coupled with physical distractions, significantly delays takeover responses.
  • Current Takeover Prompts Are Insufficient: The study found that the warning times provided by the tested vehicles in critical lateral control scenarios (lane departure) were extremely short (around 0.47-0.56 seconds), making it nearly impossible for a driver to safely regain control.
  • System Limitations in Complex Scenarios: Both vehicles struggled with large-curvature curves at higher speeds, exhibiting lane departure and significant oscillation. Similarly, in longitudinal scenarios, they showed limitations in recognizing irregularly shaped or non-standard static obstacles.
  • Driver Engagement is Key for Level 2/3: For current HMCD systems (Level 2/3), the design must prioritize keeping the driver actively engaged and prepared for emergencies, rather than assuming passive monitoring is sufficient.
  • Early Warning is Crucial for Level 3+: As automation levels increase, the emphasis must shift to strengthening the vehicle's early warning capabilities to provide ample time for drivers to take over safely.

Key Results

  • Lateral Control:
    • At 60 km/h on a 250m radius curve, both vehicles performed well.
  • At 90 km/h on a 250m radius curve, Vehicle A showed lane departure, while Vehicle B maintained better control but with large oscillations.
  • At 120 km/h on a 250m radius curve, both vehicles lost control.
  • In scenarios where lane departure warnings were issued, the time to departure was very short (0.47-0.79 s), indicating insufficient warning time.
  • The minimum required driver reaction time for safe lateral control takeover was calculated to be 1.87 seconds.
  • Longitudinal Control:
    • Vehicles showed varying success in recognizing static obstacles. Vehicle A generally performed better than Vehicle B in recognizing standard obstacles like stationary vehicles and pedestrians.
  • Both vehicles failed to recognize irregularly shaped targets (traffic cones, cardboard boxes) at 60 km/h.
  • The minimum required driver reaction time for safe longitudinal control takeover was calculated to be 4.12 seconds.

Limitations and Future Work

The authors acknowledge several limitations:

  • Lateral Control Scenarios: Due to test site constraints, only right-turn curves were tested, and the parameter settings had large gradients, limiting a full analysis of curvature and speed effects.
  • Longitudinal Control Targets: The diversity of target objects used was based on ISO standards but could be expanded further.
  • Driver State: For safety reasons, tests did not include scenarios with driver distraction or fatigue.

Future work suggested includes:

  • Enriching the variety of road types and target objects in real-vehicle tests.
  • Introducing physiological monitoring devices (EEG, ECG) to analyze takeover capabilities across different driver demographics and states.
  • Developing more adaptive and hierarchical takeover systems, potentially integrating driver monitoring systems (DMS).
  • Incorporating multi-sensor fusion perception schemes (including LiDAR) to enhance obstacle detection capabilities.
  • Exploring advanced predictive control and optimization algorithms for more resilient takeover strategies.

Summary updated

Abstract

With the increasing emergence of intelligent vehicles, novel accident patterns have gradually emerged. In human-machine cooperative driving (HMCD) states, despite driving automation systems being capable of controlling lateral and longitudinal vehicle motions over extended periods, functional limitations persist in specific scenarios due to insufficient expected functionalities. When combined with risk factors, such as driver distraction, these limitations significantly elevate accident risks. This study investigated takeover safety through real vehicle testing in two typical accident scenarios: large-curvature curves and static obstacles. The key findings include the following: (1) in scenarios involving large curvature curves and static obstacles, vehicles are prone to lane departure and missed target detection, which are typical dangerous scenarios; (2) during the human-machine cooperative driving phase, the design of the driving automation system should focus on enhancing driver engagement in driving tasks, while in the autonomous driving phase, the vehicle's early warning capabilities should be strengthened; (3) the takeover request for longitudinal control requires at least 4.12 s of driver reaction time, while the takeover request for lateral control requires at least 1.87 s. This study provides important theoretical and practical references for safety in designing assisted driving systems and the testing of hazardous scenarios.
BibTeX
@inproceedings{li2025research,
  author = {Pingfei Li and Meiling Zhou and Chang Xu and Li He and Wenhao Hu and Zhengping Tan and Lingyun Xiao and Xiaojun Mou and Hao Feng},
  title = {{Research on Takeover Safety of Intelligent Vehicles Based on Accident Scenarios in Real-Vehicle Testing}},
  booktitle = {Sensors},
  year = {2025},
  doi = {10.3390/s25175589},
}
Figure