Alibaba releases Qwen-Drive, a single model for autonomous driving

Alibaba's research arm released Qwen-Drive 1.0, a model that handles environmental perception, traffic Q&A, and route planning for driving in one system. The researchers found that text-image models don't automatically understand three-dimensional space, so spatial awareness had to be trained on purpose. The model also generates natural-language explanations for its driving actions, such as braking, though these explanations don't always match the actual maneuver.

Why it matters: This adds evidence that general vision-language models need dedicated training for physical and spatial reasoning before they can be trusted in embodied tasks like driving, rather than picking it up incidentally. It also flags a practical safety issue relevant beyond autonomous vehicles: an AI's stated explanation for an action isn't guaranteed to reflect its actual reasoning, which matters as more systems are asked to narrate their own decisions.