news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

text｜xu caiyu, li anqi

edited by li qin and yang xuan

another star car company has adjusted its team around the "end-to-end" technology route.

36kr learned from multiple insiders that xiaomi's smart driving team has recently completed a new round of organizational restructuring. 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 solutions.

as a result, xiaomi's smart driving has entered a new round of acceleration, and will shift from traditional modular development to an end-to-end large-scale model technology route. in the new stage, xiaomi's core goal will be to quickly implement new solutions and expand experience scenarios.

there are many new cars on the market, and many brands are pursuing smart driving in urban areas. if xiaomi wants to capture the hearts of users and maintain high sales, it must fill the gap in smart driving.the switch from organizational adjustments to technical routes is a signal that intelligent driving is accelerating.

"xiaomi auto's goal is to deliver end-to-end intelligent driving within the year," a person familiar with the matter revealed.

in addition, according to 36kr auto, the "autonomous driving" team that was originally responsible for pre-research work has also become a key development department, simultaneously promoting the research and development and implementation of l3/l4 autonomous driving products.

anchored on the two cards of "end-to-end" and "autonomous driving", xiaomi auto's intention of "all in intelligent driving" is obvious.

xiaomi smart driving needs to be faster

36kr learned that xiaomi has smart driving teams in beijing, shanghai and wuhan, with a total size of about 1,200 people.

the reorganized "end-to-end algorithm and function department" will be responsible for the development of mass production solutions, including two sets of technical solutions with and without lidar. the team is led by ye hangjun, chairman of xiaomi's technology committee and head of xiaomi automotive autonomous driving.

the reason behind the adjustment is the intensified competition in the industry, and the progress of xiaomi's smart driving mass production plan needs to be further accelerated.

xiaomi auto's smart driving team was one of the first teams to start recruiting after xiaomi officially announced its car manufacturing in 2021. in august 2022, the team submitted its first report to the public: a real-life video of an l2+ assisted driving system on the road covering three major scenarios: city, highway and parking.

this technical demonstration was officially delivered to users two years later. in june this year, xiaomi began to push the city pilot assisted driving function (hereinafter referred to as city noa) to some users who passed the review.

at the end of august, xiaomi's "national driving" version of smart driving began to be pushed. after the fourth ota upgrade in september, users in ten cities including beijing, shanghai, and guangzhou will have their city noa activated in real time after meeting the 1,000km smart driving safety mileage.

it took xiaomi only three and a half years from the establishment of the team to the delivery of noa in urban areas, while huawei and xiaopeng, the current first-tier companies in the industry, took five years. judging from the speed of development, xiaomi auto is indeed catching up.

but obviously, xiaomi intelligent driving is still far from the first echelon in terms of experience. at present, huawei has realized the full-scenario urban nca from parking space to parking space on its new car xiangjie s9, and xiaopeng city ngp has achieved "all-country driving" regardless of cities, routes, and road conditions, and has even implemented a pure visual solution on the new car mona 03. in comparison, xiaomi, which currently only provides urban noa functions for major roads in ten cities in models with laser radar, has made a slightly slower progress.

after the integration of the intelligent driving team, the "end-to-end algorithm and function department" responsible for mass production took on the task of accelerating the opening of urban noa. two sets of high-end and low-end solutions were developed simultaneously. lei jun's goal for this is to enter the first echelon within this year.

another "autonomous driving" team that emerged in this adjustment will focus on the research and development of l3/l4 high-level autonomous driving solutions.

although xiaomi can only catch up with the industry pace in terms of mass production, it still has a chance to seize the technological high ground in l3/l4 autonomous driving, which has not yet been implemented.

use organizational strength to ensure end-to-end supply

the industry has gradually reached a consensus: compared with the traditional multi-module solution, the end-to-end intelligent driving solution can achieve more lossless information transmission, and at the same time rely on massive data, large cloud computing power, and larger models for training, with a higher ceiling for intelligent driving. xiaomi is not the first car company to make an all-out bet on end-to-end technology through organizational adjustments.

in the middle of this year, nio merged its intelligent driving perception and regulation and control teams into a large model team; ideal also established an end-to-end autonomous driving department; and xiaopeng adjusted the technical development department of the intelligent driving team into ai end-to-end, ai application and other departments. although huawei's intelligent driving organizational structure has not undergone major changes recently, its end-to-end intelligent driving products have evolved from ads2.0 to ads3.0, and more advanced versions are already in pre-research.

as new technologies disrupt old routes and leading players make preparations one after another, traditional car companies cannot sit idly by.byd has announced a partnership with huawei on advanced intelligent driving. great wall motors has also introduced the intelligent driving company yuanrong qixing to pursue end-to-end mass production. end-to-end has begun to transform from a technological undercurrent into a key factor affecting automakers' technological heights, brand image, and even the commercialization of intelligent driving.

examples of accelerating commercialization through intelligent driving include ideal auto. in the second quarter earnings conference call this year, ideal auto said that its orders for ad max (i.e. intelligent driving version) models priced above rmb 300,000 accounted for nearly 70%. ideal believes that an end-to-end solution is the true ai approach to intelligent driving, which will actually boost sales.

in other words, with huawei's sales and smart driving commercialization soaring, and new smart driving forces extremely inward-looking, if xiaomi auto fails to catch up with the end-to-end wave, the past halo of its internet technology brand may find it difficult to continue to influence users' minds and support subsequent larger-scale sales.

however, if xiaomi auto can successfully establish an end-to-end intelligent driving system, its endless stream of massive data can be transformed into continuous momentum for xiaomi's intelligent driving evolution. investing manpower and resources in the form of organizational adjustments is an important guarantee for xiaomi to expand from end-to-end parking scenarios to end-to-end full scenarios. and autonomous driving pre-research is also an indispensable link in exploring a higher ceiling for intelligent driving.

with xiaomi auto's all-in efforts, the industry's end-to-end mass production torch will be even brighter.