XPeng began rolling out OTA version 6.2.0 on May 31, 2026, with a target completion date of June 15. Simply calling it a software update would be underselling it. The headline feature is VLA 2.0, a second-generation Vision-Language-Action architecture that collapses the decision chain between what the car sees and what it does into a single end-to-end process, bringing latency down to 80 milliseconds. For context, previous-generation systems passed information between separate perception, planning, and execution modules, each one introducing delay. VLA 2.0 cuts straight from vision to action.
XPeng’s CEO, He Xiaopeng, has been publicly self-critical about what the old approach produced. He called it a “feature monster,” a system that accumulated capabilities without ever becoming coherent. VLA 2.0 is the company’s answer to that problem, and OTA 6.2.0 is the first time it ships to customers at scale.
What Is Actually New
Four things in this update are genuinely new rather than refinements of existing features.
The first is the no-navigation NGP roaming. XPeng’s Navigate on Pilot system previously required a destination before it would activate, as do most competitors. That requirement is gone. Drivers can now engage NGP without setting any route, and the car will handle urban roads, campus environments, and underground car parks on its own. It avoids obstacles, yields to other road users, and supports smooth transitions between automated and manual control. It also addresses the problem of underground garage maneuvering, where GPS dropout, absent lane markings, and chaotic mixed traffic in enclosed spaces have beaten most ADAS systems for years. XPeng is claiming it has solved the problem well enough to ship it.
The second change is the activation of P-gear NGP. Until now, the driver had to shift out of Park before any smart driving feature would engage. OTA 6.2.0 changes that. NGP can be activated while the car is stationary, with the car transitioning to movement without additional driver input. It is a small interaction change, but the direction it points is not small.
Third is destination parking memory. If NGP has previously learned a parking space at a regularly visited location, the driver can select that space during the approach, and the car will navigate to it and park without further input. If the preferred space is taken, the system can switch to roaming mode and find an alternative.
Fourth is a lane-level rendering upgrade that combines navigation path data with real-time lane perception, giving the driver a clearer picture of the car’s intentions. Alongside this, a “familiar route” recommendation mode learns daily habits and surfaces common routes without manual setup, a narrow map window reduces screen intrusion during driving, and intersection displays now show road names prominently. The AR-HUD gains its own map mode, either full-screen or a lighter overlay. Adjustable intermittent wiper sensitivity and customizable wing mirror deployment timing round out the cabin updates.
Where XPeng Sits Right Now
Western coverage tends to benchmark Chinese smart driving against Tesla FSD. That frame is gradually becoming less useful.
Tesla’s FSD has not received regulatory approval for rollout in China. XPeng operates without that constraint, and Huawei’s ADS system, deployed across AITO and Avatr vehicles, is XPeng’s most direct domestic competitor and arguably more aggressive in straightforward urban environments. Li Auto’s ADAS is well regarded for its human-like feel while BYD’s “God’s Eye” system has the widest deployment across its model range, prioritizing coverage over depth. XPeng’s position has always been technical ambition: higher compute, and a more aggressive architecture.
The Volkswagen partnership gives that position commercial credibility. VW has agreed to deploy XPeng’s smart driving software in new electric models for the Chinese market. A legacy European manufacturer, after years of claiming it could develop autonomous capability in-house, has contracted it out to a Chinese startup. That is worth paying attention to, regardless of where you stand on the technology.
VLA 2.0’s stated design goals address the weaknesses of the previous generation. The old system was conservative in ambiguous situations, dependent on high-definition maps that covered only pre-surveyed roads, and prone to hesitation in scenarios where a human driver would not. XPeng has named the pattern it is trying to break: the assumption that hesitation is a safe default. Whether VLA 2.0 has actually broken it will take real-world validation to confirm.
The OTA Track Record Is Not Clean
XPeng has shipped software that annoyed and, in some cases, genuinely harmed its customers, and that history belongs in any fair account of this update.
In 2025, an OTA update pushed the Tianji system to older P7 models, removing the App Lab feature and USB video playback without notifying owners beforehand. Complaints on XPeng’s own community forums went largely unaddressed. When MONA M03 owners reported camera failures and navigation errors after a subsequent Tianji update, XPeng confirmed in its official response that rolling back to a previous version was not technically possible. Owners were stuck with a buggy system and no recourse.
The demo car situation is a separate issue with its own paper trail. Multiple G7 owners who purchased official used or demo vehicles through XPeng-authorized channels discovered after the transaction that their cars did not qualify for OTA updates. XPeng explained that the hardware and software on demo units differed from those on retail stock. Those complaints are documented on China’s 12365auto consumer platform from early 2026 and remain unresolved publicly.
In May 2026, a wider story circulated claiming that eight EV brands, including XPeng, were under regulatory investigation for OTA-based battery locking, which remotely reduces battery capacity after purchase without the owner’s consent. XPeng’s legal department specifically denied the investigation, stating that no regulatory contact had occurred and that the company was pursuing action against those spreading the reports. Independent complaints about post-OTA range-reduction on older P7 models remain on public platforms, regardless of what the company’s legal team says about the investigation itself.
Editor’s Take
XPeng makes ambitious technical claims and, more often than its critics expect, ships them on schedule. The P-gear NGP activation reads as a gimmick. In practice, removing a manual step from a process that users perform dozens of times a week is not a gimmick. No-navigation roaming sounds niche until you consider that underground parking and gated residential compounds are daily realities for a substantial portion of XPeng’s customer base. The 80ms latency figure is the kind of number a company publishes when it expects third parties to test it.
The uncomfortable part is that the same company shipping this update has a documented history of quietly removing features, refusing rollbacks when updates break things, and leaving certain categories of owners without support. Those are not rumors. They are logged complaints with official responses attached. VLA 2.0 may be the most technically coherent smart driving system XPeng has produced, but it is being delivered by a company whose customer service instincts are yet to keep pace with its engineering.
Sources
Autohome, Yiche, XPeng
Dollars loves following the latest innovations in automobile technology and sharing insights.
When she’s not writing, you can find her playing badminton or diving into a new opera piece.