Qwen Steps Into Autonomous Driving with Qwen-Drive-1.0-4B
Key Info
Qwen's new Qwen-Drive-1.0-4B is an Apache-2.0 vision-language foundation model for autonomous driving, combining 3D perception, driving VQA, and motion planning in a single framework while keeping the Qwen3.5-4B backbone untouched.
Highlights
- Uses two plug-in modules: a BEV head for 3D perception and a flow-matching Planning Expert for trajectory generation.
- Reports strong driving VQA results, including 77.8 on LingoQA and the lowest Ego3D distance error among compared models.