XPENG Robotaxi

XPENG's L4 ride-hailing vehicle on the GX platform, not a trim in the GX configurator. Internal testing began in July 2026.

Changes (1)

XPENG Robotaxi is XPENG’s L4 autonomous ride-hailing vehicle - a fleet variant built on the GX platform, not a trim in the retail GX configurator. This page covers the service vehicle and its China rollout. For the consumer SUV, see XPENG GX; for European models, see the Lineup hub.

China roadmap. Robotaxi is a commercial fleet programme in China - not a consumer product in Europe or Denmark, and not a promise of EU type approval for L4 operation here.

XPENG GX Robotaxi rollout - Guangzhou factory team with 3000 TOPS banner (May 2026)

Images: XPENG

Jump to: Development platforms · GX fleet vehicle · Mass production · Employee testing · Technology · Timeline · Europe · Brand Day Munich · vs retail GX · Sources

Development over time: multiple platforms

XPENG’s current GX-based mass-production fleet is one chapter in a longer Robotaxi story. Official and semi-official material shows several retail model lines used as test beds, demos or earlier permit programmes - not all of them destined to become the Guangzhou fleet vehicle.

PlatformRole in Robotaxi workWhen
G9First zero-hardware-modification Guangzhou road-test permit; later passenger-carrying Robotaxi permit - retail G9 with software upgrades onlyOct 2022 road test · May 2023 passenger permit
X9Official “Intent made visible” demo - sun-visor external displays, pedestrian yield cues; camouflaged MPV in XPENG’s own YouTube / Weibo preview before AI Day 20254 Nov 2025
P7+Appeared in XPENG’s pre-AI Day 2025 Robotaxi preview material (press reported a P7+ still labelled Robotaxi alongside the X9 video). No dedicated xpeng.com fleet announcement - distinct from the later Robo consumer trim planNov 2025 preview
GXMass-production fleet platform - factory-built L4 Robotaxi, employee beta and planned pilot operations2025-2026

At AI Day 2025, XPENG said it would launch three Robotaxi models in 2026 (typically reported as 5 / 6 / 7-seat layouts) without naming G9, X9 or P7+ as those products. The May 2026 mass-production press release names GX only. Treat G9 / X9 / P7+ entries above as development and demonstration history, not a confirmed “three-model fleet lineup”.

Current fleet vehicle: GX platform

XPENG states that its mass-produced Robotaxi is built on the XPENG GX platform - the same architecture family as the retail flagship SUV (SEPA 3.0, 800 V, Turing compute). The fleet vehicle is engineered to L4 autonomous driving standards with a separate product definition from showroom GX trims.

XPENG GX Robotaxi fleet vehicle - Robotaxi-003 door decal (May 2026 press)

First mass-produced unit

On 18 May 2026 in Guangzhou, XPENG rolled the GX-based fleet vehicle off the line and called it China’s first production-ready Robotaxi built full-stack in-house - pre-assembled at the factory rather than converted from a showroom SUV.

XPENG GX Robotaxi units on the mass-production assembly line
XPENG GX Robotaxi at the Guangzhou production line - commercialization arch
XPENG GX Robotaxi commercialization milestone - team at Guangzhou plant

Internal employee testing (July 2026)

On 10 July 2026, about eight months after AI Day 2025, Guangzhou staff began taking closed beta rides. Chairman He Xiaopeng completed the first end-to-end trip (order, pickup, destination) on the Robotaxi platform - XPENG’s proof that ride hailing, dispatch, transport and completion are wired together.

XPENG’s fleet livery on the test vehicles matches the large GX-based SUV shown at AI Day 2025 and in May 2026 press photos: wrapped body, rear privacy treatment and a ceiling-mounted screen for passengers showing a bird’s-eye autonomy view. This is still a closed employee beta, not public ride-hailing in Denmark or the EU.

Compute, perception and VLA

At AI Day 2025 (5 November), XPENG unveiled Robotaxi as part of its Physical AI roadmap: four Turing chips, up to 3,000 TOPS, pure vision without LiDAR or HD maps, dual-redundancy hardware, and an integrated in-cabin and external interaction system. According to XPENG’s technical presentations and May 2026 production release, the total compute of 3,000 TOPS is delivered across four in-house Turing AI chips (750 TOPS per chip):

Turing chip allocation breakdown

  • Chip 1 (750 TOPS) - Primary ADAS / VLA 2.0 perception: Real-time vision perception (high-resolution camera suite without LiDAR or HD maps) and spatial 3D environmental modeling.
  • Chip 2 (750 TOPS) - VLA 2.0 End-to-End planning & control: Executes the large VLA 2.0 foundation model (Vision-Language-Action) for trajectory planning, road-user intent prediction, and direct vehicle control with sub-80 ms response latency.
  • Chip 3 (750 TOPS) - Smart Cabin & Passenger AI: Powers the in-cabin passenger experience: local language models (LLM/VLM) for voice interactions, passenger climate/comfort controls, and ceiling/rear displays rendering real-time bird’s-eye autonomy views.
  • Chip 4 (750 TOPS) - Safety redundancy (Fail-Operational) & External interaction:
    • Hardware Dual-Redundancy (L4 safety requirement): To operate fully driverless without an on-board safety driver, functional safety standards (ISO 26262 ASIL-D) require complete compute redundancy. If a primary driving chip or power channel faults, this backup chip immediately executes a Minimal Risk Maneuver / Safe Stop.
    • External pedestrian interaction (External Intent Display): Drives exterior communication lighting and screens (including the sun-visor display behind the windshield that signals the vehicle’s driving intent to pedestrians and traffic).
XPENG GX Turing AI - up to four chips / 3,000 TOPS (Robotaxi variant)

Retail GX marketing render: up to four Turing AI chips / 3,000 TOPS on the Robotaxi variant (Robotaxi最高算力3000TOPS; gx.html).

XPENG says the vehicle operates without LiDAR or high-definition maps, using a pure vision solution with decision-making driven by the VLA 2.0 end-to-end large model. XPENG claims the model removes the language-translation step in traditional three-stage Vision-Language-Action architectures, compressing system response latency to under 80 milliseconds, with improved urban generalization and cross-city deployment.

Fleet cabin equipment highlighted in the release includes privacy glass, comfort gravity seats, rear in-car entertainment screens and voice-assistant control of in-car settings during rides.

Rollout timeline (XPENG statements)

WhenWhat
Oct 2022G9 obtains Guangzhou autonomous road-test permit - zero-modification retail vehicle
May 2023G9 obtains Guangzhou Robotaxi passenger test permit - earlier programme
4 Nov 2025Official X9-based Intent made visible demo (YouTube / Weibo) - external sun-visor interaction
5 Nov 2025AI Day - dedicated Robotaxi unveiled; three fleet models in 2026 announced; Amap SDK partner named
January 2026Guangzhou road testing permit - routine L4 public road testing (GX programme)
2 Mar 2026VLA 2.0 media day - Robotaxi public road testing; Volkswagen named first VLA 2.0 customer in China
March 2026Robotaxi business unit established
18 May 2026First mass-produced GX-based Robotaxi rolls off the line in Guangzhou
10 Jul 2026Internal employee testing begins; He Xiaopeng completes first end-to-end ride
16 Jul 2026Brand Day Munich - Robotaxi / VLA 2.0 slides; European partner invite; VLA adapting for Europe
Second half 2026Pilot / trial operations planned - technical viability, user acceptance, business model
Early 2027Target: fully autonomous operations without an on-site safety officer

XPENG also plans to open its Robotaxi SDK, with Amap named as the first global ecosystem partner.

Europe

There is no announced Robotaxi service timeline for Denmark or the EU - no operating permit, no public hail-to-ride launch date, and no EU type approval for L4 fleet operation as of mid-2026. XPENG Denmark’s May 2026 release repeats the China pilot plan; that is marketing context, not a Danish deployment promise.

Brand Day Munich (16 July 2026)

On the same Munich stage as the European L03 premiere, XPENG put several slides on Robotaxi and VLA 2.0. Archive stream: YouTube.

  • XPENG Robotaxi (GX) - stage reveal of the mass-production fleet vehicle on the GX platform, labelled with Dual-Redundancy Design, 3,000 TOPS and Global Deployment.
  • Robotaxi in Europe - wrapped GX-based fleet livery on stage; slide text: “We invite European partners to shape the future of mobility with us, here in Europe.”
  • Employee beta (July 2026) - slide headline tied to the Guangzhou closed beta that started 10 July; bullet line: Full-Stack In-House Development | Mass-Produced | Vision Only; VLA 2.0 framed as one model bridging L2 and L4.
  • VLA 2.0 in Europe - separate slide: One Model Bridges China and Europe · Now Adapting for Mass Production in Europe (stack adaptation for European roads - not the same as launching a Robotaxi service).

Two days earlier, on 14 July 2026, He Xiaopeng posted on X that NGP with VLA 2.0 would reach Europe in early 2027 (after China rollout from March 2026), and that XPENG looked forward to working with European partners on a Robotaxi fleet powered by the same foundation model. That is CEO-level intent - still not an EU Robotaxi operating date.

The 2 March 2026 VLA release names Volkswagen as the first VLA 2.0 customer in China, with global VLA delivery targeted from 2027 - primarily consumer ADAS, not a Danish Robotaxi launch. XPENG has also described exploring Robotaxi partnerships in Europe, the Middle East and Southeast Asia as a technology provider, not necessarily as its own ride-hailing operator.

Fleet product vs retail GX

Retail consumer models across XPENG’s lineup (such as L03 Ultra, G9L Ultra, and GX Ultra) top out at three Turing chips (2,250 TOPS), where two chips handle driving/NGP and one powers the intelligent cabin. The mass-produced Robotaxi on the GX platform is a dedicated fleet variant that adds the fourth Turing chip (3,000 TOPS total) specifically to fulfill the regulatory and functional requirements for complete L4 hardware redundancy.

Sources

Related: XPENG GX · Compare specs · Lineup hub

Changes to this page (1)

All changes across the site