Xiaomi Automobile's Intelligent Driving Updates and Advancements

TapTechNews September 20th news, 36Kr reported that the intelligent driving team of Xiaomi Automobile has recently completed a new round of organizational structure adjustment.

The original two secondary departments of 'perception' and 'regulation and control' have been reorganized and merged into the 'end-to-end algorithm and function department', which is responsible for the development of mass production plans. People familiar with the matter disclosed that Xiaomi Automobile's goal is to deliver end-to-end intelligent driving within the year.

At the technical conference in December last year, Lei Jun said: Although Xiaomi Automobile started relatively late in intelligent driving, Xiaomi's goal is still to be the number one in the industry in 2024. It is said that the relevant investment has increased from 3.3 billion to 4.7 billion, and the AI laboratory established in 2016 by Xiaomi has been called, and currently has 3000 AI engineers.

On the hardware side, TapTechNews found that Xiaomi Automobile is equipped with 2 NVIDIA Orin chips, which can achieve 508 TOPS computing power, and also has 1 laser radar, 11 high-definition cameras, 3 millimeter-wave radars, and 12 ultrasonic radars.

Xiaomi Automobile said that SU7 has comprehensive empowerment by AI, fully self-developed underlying algorithms, fully self-developed for the AD three-piece set (BEV, Transformer, OCC occupancy network), and is fully integrated into the large model.

Xiaomi Automobile adopts zoom BEV technology, which can switch the recognition mode according to the vehicle driving scene; it also has super-resolution occupancy network technology, which can improve the recognition accuracy of obstacles, and also supports intelligent noise reduction in rainy and snowy weather.

In addition, Xiaomi Automobile has also self-developed a road large model without map plan. The large model generates the road topology in real time and generates high-precision maps, with an effect comparable to high-precision maps, thereby realizing functions such as intersection recognition and automatic bypass of obstacles.

Lei Jun said that the intelligent driving ability of Xiaomi Automobile can deal with complex road conditions, and the driver does not take over throughout the whole process. Xiaomi's urban pilot assisted NOA will achieve 100-city coverage support by the end of next year.

In addition, Lei Jun also demonstrated the valet parking ability of Xiaomi Automobile, which can automatically pass through the gate, give way to pedestrians, park in the parking space and intelligent summon, etc., and can even deal with very narrow mechanical garage parking scenarios. Behind this is Xiaomi's end-to-end perception decision-making large model, perception decision-making algorithm two-in-one, large model real-time observation, dynamic adjustment.