XPeng CEO: We'll See L5 Self-Driving Cars in 3 to 5 Years — and AI Is Why

XPeng CEO He Xiaopeng has a prediction that would have sounded delusional two years ago: fully driverless cars, capable of handling any road in any condition, will be on sale within five years. He’s not hedging.

Speaking after the MONA L03 China debut event on Thursday, Xiaopeng told journalists that artificial intelligence has fundamentally changed the math on autonomous driving. “If you’d asked me two years ago, I wouldn’t have believed L4 and L5 could truly land in our lifetime,” he said. “But in the last two years, AI has had a massive推动作用 on autonomous driving and even physical AI. I firmly believe that in the next 3 to 5 years, we won’t just see L4 deployment — we’ll see L5 too, operating safely and smoothly in every scenario. The pace will exceed everyone’s imagination.”

The difference this time, Xiaopeng argues, is regulatory. In June, the United Nations formally adopted a global autonomous driving regulation called ADS GTR — and China was one of its principal authors. The regulation covers everything from L2 driver assistance through L5 full autonomy. For L2 systems, the rule is comprehensive, meaning that by December of this year, China’s current-generation L2 systems will be legally deployable worldwide.

That matters for XPeng. The company has been quietly building its autonomous driving stack — the VLA (Vision-Language-Action) and VLM (Vision-Language Model) systems — with global deployment in mind. A UN-level framework removes the patchwork of country-by-country approvals that has slowed every automaker’s autonomous driving rollout.

Xiaopeng revealed that during China’s annual parliamentary meetings earlier this year, he submitted a formal proposal urging the country to skip L3 certification entirely and jump directly to L4 and L5 regulation. It’s an unconventional take — most automakers have treated L3 as a necessary stepping stone — but it reflects his view that the technology curve is steepening, not flattening.

China’s smart driving regulations and the actual performance of its domestic systems are already ahead of the rest of the world, he said. The UN’s new framework effectively codifies that lead.

The timeline he’s describing — L4 and L5 in 3 to 5 years — is aggressive by any measure. Most industry analysts expect L4 deployment in limited geofenced areas (robotaxis in specific cities, for example) within that window, not across all scenarios. L5, which requires a car to drive itself anywhere a human could, under any conditions including snow, construction, and unmapped roads, remains a technical challenge that even Waymo and Tesla haven’t claimed to have solved.

Xiaopeng isn’t dismissing the difficulty. He’s saying AI has made the hard part easier than it looked. The question is whether the rest of the industry — and regulators — move as fast as he believes the technology will.