Xiaomi's assisted driving team welcomes another key figure: Chen Guang, the former CTO of FAW Nanjing, joins | Exclusive from 36Kr

Text | Li Anqi

Editor | Li Qin

36Kr Auto has learned from multiple channels that Chen Guang, the former CTO of FAW Nanjing Research Institute, has joined Xiaomi Auto and taken up the position of the person in charge of assisted driving perception, reporting to Ye Hangjun, the person in charge of Xiaomi's assisted driving. Cai Rui, the former person in charge of Xiaomi's assisted driving perception, has transferred to the robot - related department.

Before joining Xiaomi Auto, Chen Guang was the CTO of FAW Nanjing Research Institute, responsible for the overall technological R & D and framework design of the autonomous driving business. He led the team to develop the third - generation L4 - level fully unmanned Robotaxi (autonomous driving taxi) of FAW Hongqi. FAW Nanjing Research Institute is a subsidiary of FAW Group, but there have recently been rumors of the team's disbandment.

Dr. Chen Guang graduated from the Department of Electrical and Computer Science at the University of Missouri in the United States and has many years of R & D experience in artificial intelligence and computer vision. He has published multiple papers at top - tier artificial intelligence conferences such as CVPR. Chen Guang also worked at the US R & D center of Baidu Apollo for many years and once served as the person in charge of the perception system technology.

According to 36Kr Auto, currently, Xiaomi's assisted driving team is divided into two major groups: "end - to - end" algorithm and function, and technology pre - research. After joining Xiaomi, Chen Guang's main task is to continuously promote the implementation of the "end - to - end" function.

Different from the traditional rule - based solutions, the "end - to - end" solution integrates modules such as perception, prediction, and planning control into a unified deep - learning model. Therefore, in the "end - to - end" organizational structure of automobile manufacturers, there are usually two model teams for perception and planning control.

In February this year, Xiaomi officially pushed the "end - to - end" full - scenario assisted driving function to all users. This is a version trained by Xiaomi based on 3 million Clips (video segment materials). Xiaomi said that it will also launch a version based on 10 million Clips in the future and continue to evolve.

In terms of technology pre - research, industry insiders told 36Kr that Xiaomi is also pre - researching the currently popular VLA (Vision - Language - Action) model, etc., which is led by Chen Long, a former principal scientist recruited personally by Lei Jun from the British autonomous driving company Wayve. "Xiaomi started the R & D of VLA relatively early, and its progress may be similar to that of Li Auto now."

As a new car - making force that officially launched its first car in 2024, Xiaomi itself needs to invest more resources to catch up with the average level of the assisted driving industry.

Xiaomi is also constantly recruiting high - end technical talents. Whether it is Wang Naiyan, the former CTO of TuSimple, who joined last year, or Chen Long, the former principal scientist of Wayve, recruited personally by Lei Jun, or the newly joined Chen Guang, the density of high - end talents in Xiaomi's assisted driving is already quite high.

Previously, according to 36Kr Auto, the scale of Xiaomi's assisted driving team has reached 1,200 people.

However, the engineering challenges and experience - accumulation process of mass - producing assisted driving are difficult to compress.

From 2023 to the present, Li Auto has switched routes several times and adjusted its departmental organizational structure multiple times, successively betting on the "end - to - end" and VLA (Vision - Language - Action) large - model departments before achieving the current phased results.

As Li Xiang, the CEO of Li Auto, said, a company can't directly eat the tenth bun. "If you can't do well in the rule - based algorithm, you simply don't know how to do end - to - end; if you haven't achieved an extremely high level in end - to - end, you don't even know how to train the VLA (Vision - Language - Action model)."

Xiaomi's assisted driving is also experiencing a similar route switch.

In June 2024, three years after starting car - making, Xiaomi gradually pushed the rule - based urban NOA solution to users. However, as the assisted driving technology route has evolved, the rule - based solution has difficulty coping with complex urban roads and has a limited ability ceiling. Therefore, the industry has shifted to the "end - to - end" solution led by Tesla's FSD.

Xiaomi also officially established the "end - to - end" department in September last year. After half a year of R & D, Xiaomi's "end - to - end" solution was officially pushed to users in February this year. However, from the actual user experience, Xiaomi's assisted driving level still lags behind the industry's first - tier players.

People familiar with the matter said that Xiaomi is intensively developing the next - generation assisted driving solution, VLA, "which may be launched within this year."

In other words, Xiaomi Auto is trying to achieve a "catch - up" effect similar to that of Li Auto in a shorter time.

However, Xiaomi's assisted driving is currently facing huge public - opinion pressure. At the end of March this year, a Xiaomi SU7 with entry - level assisted driving enabled was involved in a traffic accident on the expressway in Tongling, Anhui, resulting in the death of three people in the car, which triggered an uproar in public opinion.

Xiaomi has fallen into a serious crisis of trust regarding the safety of assisted driving. Lei Jun said that this is "the most difficult period since the founding of Xiaomi."

Both Xiaomi's assisted driving and Xiaomi itself need a comeback.

Xiaomi's second car, the YU7, undertakes this important task. At the new - car unveiling press conference on May 22, Lei Jun tried to dispel users' safety concerns by standardizing assisted - driving hardware across the board.

For this model, Xiaomi has standardized the Nvidia Thor in - vehicle chip with a 4nm process and 700 TOPS computing power, as well as 1 lidar, 1 4D millimeter - wave radar, 11 high - definition cameras, and 12 ultrasonic radars.

However, after recruiting many high - level talents and maximizing the configuration of the new car, whether Xiaomi's assisted driving can truly "catch up" still needs strong proof.