Photo Credit: iStockThere are plenty of ways to earn press attention in the fraught world of autonomous driving. Distilling a brand's philosophy into a phrase like "language is poison" is certainly one way to go about it. Xpeng's head of autonomous driving, Dr. Xianming Liu, continues to draw attention for his go-to phrase in a wide-ranging interview with Electrek that also covered the brand's lofty artificial intelligence budget. What happened? In the interview, Liu claimed that Xpeng has quickly caught up to Tesla's Full Self-Driving v13, helped by an AI training budget of roughly $41 million a month.AdvertisementAdvertisementLiu later clarified his famous "language is poison" phrase in the Electrek interview — Xpeng still takes language input from drivers, but it no longer wants the vehicle to produce language-like representations during driving.As Electrek described it, the updated setup skips the extra step of turning what the car sees into language-style tokens before converting that into driving commands.Liu told Electrek that removing that middle step reduces computational waste and latency. He also said Xpeng is pairing that approach with a "world model" meant to anticipate what is about to happen on the road, with a production rollout expected later this year.Liu said Xpeng's main driving stack is vision-only, though the company still includes radar and ultrasonic sensors on some vehicles for separate emergency safety systems such as automatic braking and steering. AdvertisementAdvertisementThat puts Xpeng somewhere between Tesla's camera-only philosophy and Waymo's more sensor-heavy approach, the publication noted.Liu also touted Chinese roads as a superior training ground for autonomous driving. "In China, you have a larger chance to get corner cases and data," Liu told Electrek. "This is one advantage."Why does it matter?Xpeng's spending shows how much money automakers are pouring into software alongside factories and batteries.How companies best deploy their money for safety purposes is somewhat up for debate. As Electrek alluded to, Waymo and Tesla are at opposite ends of the camera vs. sensor debate.AdvertisementAdvertisementXpeng offers another path, using the sensors as a fail-safe to help prevent some of the high-profile incidents that have plagued Tesla."You're not talking about chatting with ChatGPT and making a mistake — you just say, 'hey, this is so silly, redo,'" Liu told Electrek. "You're talking about lives."What are people saying?Liu has some personal skin in the game when it comes to matching Tesla's FSD. There's a now-public bet that he'd have to streak across the Golden Gate Bridge if Xpeng failed to meet its Tesla benchmark, as Electrek noted. "I'm pretty confident I don't need to run," he told the outlet. "Based on test drives, we already hit the goal."AdvertisementAdvertisementHe's also pouring a ton of resources into the AI systems and capitalizing on China as a prime training ground. A commenter on Electrek opined that Liu "was simply being humble" in his comparisons to Tesla."I've watched the videos — Xpeng VLA2.0 is well ahead of Tesla Assisted Driving in China," they wrote. "I expect Xpeng to leapfrog Tesla quickly when the software update is released in Europe."Get TCD's free newsletters for easy tips, smart advice, and a chance to earn $5,000 toward home upgrades. To see more stories like this one, change your Google preferences here.
+15 others
