Xiaomi has significantly improved its autopilot by training it on 10 million real videos.

Xiaomi has released a major update to its proprietary driver assistance system (ADAS) for its electric vehicles. The new version is already available for the SU7 Pro, Max, and Ultra models, and will also be pre-installed in the YU7 sedan. The main innovation is an improved algorithm trained not on simulations, but on 10 million short clips of real trips recorded from cameras, lidars, and radars.

Each clip lasts from 30 to 60 seconds and captures real-life traffic situations: traffic jams, sudden braking, driving around parked cars, and complex intersections. Instead of following set rules, the AI learns by example — it watches how drivers behave and copies human behavior. Initially, the database consisted of 3 million such videos, but now it has grown more than threefold.

According to Xiaomi, after training, the system has become noticeably better: smoothness of driving has increased by 57%, obstacle avoidance by 67%, and navigation accuracy through intersections by 23%. The result is fewer jerks and strange maneuvers, as well as more predictable behavior on the roads.
The ADAS update is part of Xiaomi's large-scale strategy in the field of smart cars. The company has invested 5.79 billion yuan (about $807 million) in the development of this area, with more than 1,800 engineers working on the ADAS development team and 400 test cars collecting data. The company's MiMo neural network is responsible for training.