During the 18th Xuan Yuan Automotive Blue Book Forum held in Guangzhou from May 15 to 17, Han Xu, founder and CEO of WeRide, emphasized the importance of measurable performance metrics over buzzwords when discussing the company's Robotaxi technology progress.
Han Xu noted that he prefers to rely on data rather than jargon. He expressed skepticism towards terms like "large models," "world models," "VLA," "end-to-end," and "reinforcement learning," pointing out that many who use these terms have not even read a single research paper, with their presentations often prepared by subordinates.
He highlighted a significant reliability gap between advanced driver-assistance systems and fully autonomous driving, stating that the latter requires a reliability level 1000 times, or three orders of magnitude, higher than the former.
According to Han Xu, companies working on advanced driver-assistance systems often claim targets of achieving 1000 kilometers per intervention (MPI) by year-end. However, current performance in urban driving conditions typically ranges between 100 to 200 kilometers per intervention, which is considered decent. Some claim 1000 kilometers or even tenfold annual improvements, yet their capabilities remain at the thousand-kilometer level.
For Level 4 autonomous driving, Han Xu provided WeRide's benchmark: one million kilometers per intervention. He asserted that without reaching a million-kilometer MPI, companies should not claim to be engaged in driverless technology, as they are still operating at the thousand-kilometer tier.
Han Xu stressed that in China, only three companies truly qualify to make such claims: WeRide, Pony.ai, and Baidu. Didi is making efforts but has not yet reached this level. Other companies may purchase WeRide's technology or vehicles but have not achieved this capability independently.
In conclusion, Han Xu stated, "I have the confidence to present these numbers. Let's avoid throwing around terms and instead show me the data."
Comments