Chinese tech giant Xiaomi has officially released and open-sourced its new Xiaomi OneVL framework. It is a system designed to improve how autonomous driving models understand, reason, and predict road situations.
According to the company, Xiaomi OneVL is the first framework in the industry to combine multiple major technologies into a single system. These include VLA (Vision-Language-Action), world models, and latent space inference. The company says the framework builds on the reasoning abilities of the XLA model while improving both inference speed and accuracy.
In autonomous driving research, VLA and world models have usually been treated as separate approaches. VLA systems primarily focus on understanding traffic scenes and generating driving actions, whereas world models predict how a scene may evolve. Xiaomi says OneVL is the first framework to unify both methods through latent space reasoning.
The company also claims that the framework performs strongly across several mainstream benchmarks for perception, reasoning, and planning. Xiaomi says OneVL pushes the limits of latent inference methods and delivers higher accuracy than explicit Chain-of-Thought (CoT) reasoning while maintaining speeds similar to latent space CoT systems that focus only on predicting final answers.
Xiaomi OneVL is also built around interoperability
Another major focus of the framework is interpretability. Xiaomi says OneVL can explain its decision-making process in both language and visual form. Practically, this means the system can describe why a vehicle should take a certain driving action while also showing predictions of what could happen next on the road.
The OneVL comes just after Xiaomi recently open-sourced its audio generation model called Omnivoice. By open-sourcing OneVL, Xiaomi is also positioning itself more aggressively in the broader AI and smart mobility space, areas where competition has continued to increase among major technology companies.
Don’t miss a thing! Join our Telegram community for instant updates and grab our free daily newsletter for the best tech stories!
For more daily updates, please visit our News Section.
Comments