Related Experiment Video
Updated: Jul 25, 2025
07:15
Tactile Vibrating Toolkit and Driving Simulation Platform for Driving-Related Research
Published on: December 18, 2020
4.5K
Data generation for connected and automated vehicle tests using deep learning models.
1School of Traffic and Transportation Engineering, Central South University, Changsha, Hunan 410075, China; Hunan Key Laboratory of Smart Roadway and Cooperative Vehicle-Infrastructure Systems, Changsha University of Science & Technology, Changsha 410114, Hunan, China.
Accident; Analysis and Prevention
|June 28, 2023
Summary
Generative models like WGAN-GP and VAE-GAN enhance connected and automated vehicle (CAV) testing by creating diverse trajectory data. WGAN-GP proves superior in generating critical driving scenarios for improved CAV safety performance.
Area of Science:
- Artificial Intelligence
- Robotics
- Transportation Engineering
Background:
- Connected and automated vehicles (CAVs) require extensive testing for safety validation.
- Real-world trajectory data for CAVs is often limited in sample size and diversity.
- Critical scenarios crucial for CAV testing may be absent in collected datasets.
Purpose of the Study:
- To develop advanced generative models for creating realistic and diverse background vehicle trajectory data.
- To evaluate the effectiveness of generated trajectory data in improving CAV safety performance assessment.
- To compare the performance of Wasserstein generative adversarial network with gradient penalty (WGAN-GP) and variational autoencoder-generative adversarial network (VAE-GAN) models.
Main Methods:
- Developed and implemented WGAN-GP and VAE-GAN models for trajectory data generation.
- Learned compressed representations of trajectory data in latent space.
- Generated new trajectory data by sampling from the latent space and mapping back.
- Integrated real and generated data into a cooperative adaptive cruise control (CACC) car-following model for CAVs.
- Evaluated safety performance using the time-to-collision (TTC) index.
Main Results:
- Generated trajectory data exhibited both differences and similarities to real data samples.
- The use of generated trajectory data increased the occurrence of critical fragments (low TTC) in CAV simulations.
- WGAN-GP demonstrated superior performance over VAE-GAN in generating critical fragments, indicated by a higher ratio.
- Both models successfully expanded the diversity of trajectory data for testing.
Conclusions:
- Generative models like WGAN-GP and VAE-GAN are effective for augmenting CAV trajectory datasets.
- Generated data, particularly from WGAN-GP, can reveal more critical safety scenarios for CAVs.
- This approach enhances the robustness and comprehensiveness of simulation-based testing for CAVs.
- Findings support improved safety performance evaluation and development of CAVs.
