This technique aims to employ generative AI to produce a variety of complex and diverse road scenarios that simulate the kinds of situations likely to be encountered in the real world. To evaluate the performance of autonomous driving systems under different road conditions, we generate primarily urban road scenes and simulate multiple weather conditions such as sunny, cloudy, and rainy, as well as different lighting conditions including daytime, dusk, and nighttime. Test scenarios cover stationary target vehicles, slow-moving target vehicles, pedestrians crossing the road, and other common traffic situations to assess the systems’ adaptability. In addition, we generate uncommon yet potentially challenging situations for autonomous systems during normal driving—such as a bicycle suddenly appearing. This technique is based on diffusion models as the core framework, integrating multimodal control signals enriched with physical information, and employs verification tools to ensure the accuracy of the generated scenarios.

1.Input text prompt format: String 

2.Input control video format: mp4 

3.Input control video resolution: 720p 

4.Output video format: mp4 

5.Output video resolution: 720p 

6.Generated scenarios: various weather, lighting conditions, and testing situations

This technology can be applied to autonomous vehicle system simulation testing.

This image generation technology can create driving scenarios, enhance the completeness and safety of model training, and accelerate the verification of autonomous vehicle systems.