Edit the search image.
There is no shortcut to the research and development of intelligent driving. Author |Wang Yajun.
Cover |Photo.com.
Li Xiang, who has always been careful in budgeting, no longer hesitates to invest in intelligent driving. On November 25, Li Auto announced that its L series models have been updated with the largest OTA 50, all users will be pushed by the end of the year. For the upgraded intelligent driving experience, Ideal declares: All-Scenario Intelligent Driving (NOA) "can be opened nationwide", and All-Scene Assisted Driving (LCC) "can be opened globally". At the Q3 financial report meeting in the first half of this month, Ma Donghui, the chief engineer of Ideal, said that he would take the leading position of intelligent driving as the core strategic goal and continue to expand the size of the team: the autonomous driving department has about 900 employees, and will exceed 2,000 next year and more than 2,500 in 2025. This means that the expected size of the ideal autonomous driving team in the next two years will be about twice that of the current NIO. A former NIO employee told Snow Leopard Finance that before the layoffs in November, NIO's autonomous driving team was about 1,300 people.
Two years ago, Li Xiang once made a bold statement, "The company's self-developed autonomous driving system can compete head-on with Huawei and Tesla in 2022." However, some ideal employees bluntly said that intelligent driving is still the company's shortcoming.
8 years after its establishment, the ideal began to "make up lessons" for intelligent driving, is it still too late?
Recommend our annual meeting. On December 19, the 2023 Global Autonomous Driving Summit will be held in Shenzhen. The main venue has an opening ceremony, as well as three special sessions: high-end intelligent driving, large model, and computing power. The sub-venues will hold the Shenzhen Nanshan Intelligent Connected Vehicle Government-Enterprise Exchange Meeting, the Autonomous Driving Analyst Forum, and the Autonomous Driving BEV Perception Technology Forum. At present, 20+ guests, including Professor Deng Zhidong of Tsinghua University, Li Hongyang, author of UNIAD, Zhang Fang, co-founder of IDRIVERPLUS, Huang Guan, founder of Jijia Technology, and Li Zhiqi, a work of Bevformer, have confirmed their participation in the speech and high-end dialogue. Registration for free or ticket purchase is open, and we look forward to your participation
Edit the search image.
01.Expanding enrollment and expanding recruitment, Lin Feng (pseudonym), a researcher at intelligent driving and automatic driving, noticed that from September to now, he has two friends who have joined the ideal intelligent driving team. They didn't do autonomous driving before, but engaged in computer vision industry research. In his past Xi and industry experience, Lin Feng has never seen an AI practitioner from a "non-science class" join the intelligent driving team of an automobile company. He believes that the ideal of including pan-AI practitioners in the intelligent driving team means that the company's intelligent driving recruitment efforts have begun to increase. "Ideal has been making efforts in the recruitment of intelligent driving teams this year. Headhunter Jiang Nan (pseudonym) told Snow Leopard Finance Agency, "Especially in terms of recruitment budget, it is much higher than that of NIO, which is also a new force." "This is reflected in the ideal for headhunters to recruit more relaxed ranks, last year benchmarking Ali P8 and above to go headhunting, this year P7 can go headhunting. Although the overall salary is not much different from last year, there is more room for people with excellent ability and resume to negotiate salary, "a conservative estimate of 50% more". Within the ideal intelligent driving team, some people also feel the company's strength. "Since the company's decision to launch urban NOA at the beginning of the year, the intelligent driving team has continued to expand. Yang Lin (pseudonym), a member of the Ideal Intelligent Driving team, told Snow Leopard Finance Agency, "I can clearly feel that there are more and more returnees among my colleagues." After the Spring Festival this year, Li Xiang issued a letter to all employees, saying that based on BEV perception and Transformer models, urban NOA navigation assisted driving (which does not rely on high-precision maps) for end-to-end training will be implemented from the end of 2023. In March this year, the company's intelligent driving team conducted a joint debugging experience (observing the operation of multiple functions or multiple components together), and by June, the intelligent driving ability has been significantly improved. Yang Lin told Snow Leopard Finance Agency, "This level of improvement may take one or two years to complete in traditional car companies." Another member of the Ideal Intelligent Driving team told Snow Leopard Finance that his current focus is on conducting on-board tests, which is in the "ability acceleration stage" in the intelligent driving research and development process. The company's senior management also said at the Q3 financial report that the company is confident that it will become the "first echelon verified by the market" in the first half of next year. "After Huawei announced again in September that it was far ahead of intelligent driving, the company's senior management took the intelligent driving strategy to a new level. Yang Lin told Snow Leopard Finance Agency, "After entering October, the company continues to expand its recruitment efforts, and the team is also optimizing personnel." ”02.To make up for the shortcomings, it is good to sell the body as the ideal of the new force to sell the championship, and its performance in the field of intelligent driving is not outstanding. Cheng Jun (pseudonym), a practitioner of intelligent driving, told Snow Leopard Finance that compared with Huawei, Xiaopeng, and other leading intelligent driving manufacturers, the ideal intelligent driving function has a significant gap in planning ability and horizontal control, "positioning at the L2 level, not reaching the L2++ function". The number of urban NOAs taken over per 100 kilometers is a key indicator to measure the intelligent driving capability of an OEM. According to the information officially disclosed by Xpeng, the number of emergency takeovers per 100 kilometers of the city's NOA (Xpeng NGP) is 065 times, the efficiency is close to the level of 90% of human drivers. But the ideal has not yet come up with a plan for this. Yang Lin's judgment is that the reason why Ideal expands the intelligent driving team is to "make up for the shortcomings". The backwardness of the ideal in the intelligent driving track has laid a clue at the beginning of the company's establishment. Li Xiang once admitted on Weibo that in the early days of the company's establishment, due to his limited financing ability, the company's funds could not support the self-research of intelligent driving technology. "It's not that I don't want to engage in self-research, but the money I have raised barely meets the construction of product research and development, self-built factories, first-class chains and service systems, and I have to pay salaries to the team. Although the lack of money has improved after the IPO of Li Auto's U.S. stocks, Li Xiang, who has been focusing on cost control, has been quite conservative in his investment in research and development for a long time. A former employee of Ideal told Snow Leopard Finance that during his tenure in the Ideal Intelligent Driving team, his salary was 20% lower than that of his peers at NIO. From 2020 to the first half of 2023, the proportion of ideal R&D expenses to revenue is % and 9% respectively, ranking at the bottom of Wei Xiaoli. In the past, the intelligent driving function was not a rigid need of consumers, but urban NOA has become a flashpoint in the industry, which has changed this situation. At this year's China Electric Vehicle 100 Forum, Gao Guorong, vice president of the Intelligent Driving Business Group, said that intelligent driving has become a major factor in consumers' purchase decisions. In order to sell more cars, improving the intelligent driving function has naturally become an ideal top priority. On October 31, Li Xiang said on his personal Weibo, "If the average monthly (delivery) is only 50,000 vehicles next year, I will go out to find a job." In addition to sales, the make-up class of intelligent driving is also closely related to Li Xiang's other grand vision. In an internal letter in early 2023, Li Xiang said that the company's vision is to "become the world's leading artificial intelligence company by 2030." "Intelligent driving is similar to artificial intelligence in terms of technical architecture, and intelligent driving will be more complex at the algorithm level, with higher requirements for integration and delicacy. Lin Feng told Snow Leopard Finance Agency, "Being able to lead in intelligent driving means leading in general artificial intelligence." ”03.Conclusion: There is no shortcut to the research and development of intelligent drivingAlthough the investment in intelligent driving has been increased, the ideal has not yet won a big victory in this year's intelligent driving competition. At the Shanghai Auto Show in April, Ideal announced that it would land in 100 cities this year. By the Chengdu Auto Show in August, the ideal changed the publicity caliber from "urban NOA" to "commuter NOA". "The scope of the commuter NOA is on a limited route, while the urban NOA has to face more complex sections and more uncertainties. Cheng Jun told Snow Leopard Finance Agency, "The difference in the difficulty of achieving the two is exponential." "Even if it is a less difficult commuting NOA, Ideal is not open to all car owners in the landing city, but to specially recruited "early bird users" who can only reach 10,000 people by the end of the year. However, with the continuous development of Ideal since the beginning of this year, the optimization and iteration of intelligent driving R&D and product experience may be better than expected, which is why there is the "OTA 5." mentioned at the beginning of this article0 will push all users before the end of the year, and all scenarios of intelligent driving (NOA) can be opened nationwide". There is no shortage of "tuition" for the ideal of making up lessons for smart driving. As of the end of Q3 2023, the ideal free cash flow was 132200 million yuan. What is lacking in the ideal is time. Huaxi ** said in a research report that intelligent driving has effectively influenced consumers' car purchase decisions, and in the future, the leading car companies that can identify differentiated selling points and leading technology will form differentiation in the era of "standard intelligent driving" from the end of this year to the beginning of next year, and maintain a leading edge for more than two years. In the face of a narrower and narrower window of time, it is difficult to find shortcuts to ideals. Influenced by Tesla's vision system-based 3D perception autonomous driving system (BEV+Transformer) released in 2021, the development path of intelligent driving in China this year has switched from relying on high-precision maps to build vehicle road information to the mode of "heavy perception and light map". There is a view that the switch of the technical path gives Ideal the opportunity to return to the same running line with Huawei and Xiaopeng. "It is certainly a trend to focus on perception and light maps, but the premise is that key cities will switch after they have decided on extreme road conditions by collecting high-precision land. Lin Feng told Snow Leopard Finance that just like Xiaopeng in Guangzhou and Huawei in Shanghai collecting high-precision maps, there is no shortcut to the research and development of intelligent driving. He Xiaopeng said in a recent interview that in the field of intelligent driving, if you want to catch up, you must have long-term accumulation, be willing to work hard, and work hard. With a fairly abundant cash flow, how much time can you buy for an ideal corner overtaking?