Infinite Traffic Scenarios for AV Testing
Using large language models to generate diverse simulation environments
DriveGen introduces a novel traffic simulation framework that leverages large models to create unlimited diverse traffic scenarios for autonomous vehicle testing and training.
- Overcomes diversity limitations of single-dataset approaches
- Supports customized scenario generation through text prompts
- Integrates specialized modules for initialization and realistic traffic rollout
- Enables generation of challenging corner cases for safety testing
This research advances engineering capabilities for AV development by providing richer simulation environments that better prepare autonomous systems for real-world complexity and edge cases, ultimately accelerating safe deployment.
DriveGen: Towards Infinite Diverse Traffic Scenarios with Large Models