He Xiaopeng predicts that Robotaxi (autonomous taxi) will enter a turning point in two years and suggests that major car companies transition to end-to-end.
On July 11th, He Xiaopeng, Chairman of Xiaopeng Motors, posted on Weibo: "In the past two days, Robotaxi has been on fire, and I have communicated with several AI/travel/L4 industry friends about robotaxi. The general conclusion is that the ability, regulations, and commercial closed loop have not yet arrived. The reason why it can be quickly balanced is because many accounts have not been calculated. However, with acceleration, I believe that we will start to enter a turning point in about two years
He Xiaopeng added that the ability requirements for Robotaxi are significantly higher than L4 (note: driving automation is divided into six levels from L0 to L5, with L4 being highly automated driving), and there are comprehensive issues such as operational regulations and social factors, making development difficult. In addition, many of L4 Company's technological routes are still a combination of algorithms and small AI models, and they are hesitating whether to switch to end-to-end. His advice on this is: "Don't hesitate, change it quickly, the one behind is the big one
Source: Weibo
Recently, with the introduction of relevant favorable policies, more companies have begun to embrace Robotaxi. The self driving travel platform "Carrot Run" under Baidu has received particularly prominent attention. On July 10th, "Carrot Run orders skyrocketed" made it to the hot search on Weibo. According to official data from Baidu, as of April 19th, the cumulative number of service orders for Carrot Run has exceeded 6 million; About 826000 ride hailing services were provided in the first quarter, a year-on-year increase of 25%.
As early as the Xiaopeng Automotive Technology Day in October 2022, Wu Xinzhou, who was then the Vice President of Xiaopeng Automotive's autonomous driving, mentioned that he hoped to start Xiaopeng's own Robotaxi testing operation in Guangzhou in 2023 or 2024. At the same time, he hoped that by 2025, the development of Robotaxi and Xiaopeng Automotive's hardware capabilities could achieve a "meeting" and enter the advanced autonomous driving stage.
Taking this opportunity to evaluate Robotaxi, He Xiaopeng also expressed his optimism about the end-to-end technology roadmap. Unlike traditional autonomous driving routes, end-to-end large models have replaced multiple modules previously used for perception, description, prediction, and planning, allowing autonomous driving to directly input image data from one end and output operational control from the other end, closer to real human driving. This can greatly improve the anthropomorphism and development efficiency of intelligent driving.
In March of this year, Tesla introduced "end-to-end neural network" technology to the FSDV 12.3.1 version pushed to North American car owners. By deeply learning from over one million driving data and using artificial intelligence technology to influence vehicle control, such as operating steering wheel pedals, turn signals, etc., instead of hard coding every driving behavior, this transformation is hailed as a "game changing technology".