Xiaomi EV said it has started rolling out an enhanced version of its Hyper Autonomous Driving (HAD) driver assistance system to eligible vehicles, adding reinforcement learning algorithms and a so-called world model to refine driving behavior.
In a post on Weibo, the company said the upgraded system is designed to continuously explore more optimal driving strategies through simulation and large-scale data training. Xiaomi said the world model enables simultaneous exploration of more than 100 driving routes, allowing the system to rapidly accumulate operational experience across a wide range of scenarios.
See also: Xiaomi Hits 500,000-Vehicle Production Milestone, Sees Over 400,000 Deliveries in 2025
“By incorporating reinforcement learning algorithms, the system can repeatedly explore within a virtual environment constructed by the world model, receiving rewards or penalties to derive optimal driving strategies,” Xiaomi EV said on Weibo.
According to the company, the enhanced HAD now delivers smoother acceleration and deceleration, more decisive lane changes and more precise recognition of road conditions. Xiaomi EV reiterated that the system remains a driver assistance feature rather than fully autonomous driving and that drivers must stay attentive at all times.
See also: Xiaomi’s EV and AI Businesses Turn Profitable for First Time as Quarterly Deliveries Hit Record
Xiaomi first unveiled the enhanced HAD system on Nov. 21 at the opening of the 2025 Guangzhou Auto Show. The company had previously begun rolling out its end-to-end technology-based HAD to eligible vehicles in late February, positioning the feature as a free selling point. By late July, the system had been upgraded using training based on 10 million video clips of real-world driving behavior.