XPeng is known for electric vehicles. IRON is the company’s argument that the technology stack behind autonomous driving, sensor fusion, spatial computing, large neural networks, is directly applicable to humanoid robots. Unveiled in May 2025 with 82 degrees of freedom (22 DOF per hand alone), IRON is the highest-DOF Chinese humanoid robot publicly announced. Mass production preparation is targeted for 2027. This article examines IRON’s engineering, XPeng’s rationale for entering humanoid robotics, and what 82 DOF actually means for manipulation capability.
XPeng’s Rationale
XPeng CEO He Xiaopeng stated at IRON’s announcement that autonomous driving and humanoid robots are “essentially the same technology problem.” Both require fusing multiple camera streams into a 3D world model, predicting the motion of agents in that world, and executing physical actions based on those predictions. XPeng has spent seven years and billions of yuan building this stack for vehicles. Applying it to a bipedal robot is an extension of existing investment rather than a new capability build. XPeng also manufactures at scale, hundreds of thousands of vehicles per year. The manufacturing infrastructure, supply chain relationships, and production engineering expertise needed to mass-produce humanoid robots are analogous to those needed for EVs, with necessary modifications for the different product category.
82 DOF: What It Means
A standard industrial robot arm has 6 DOF. The Unitree G1 has 23 total DOF. UBTECH Walker S2 has 41 DOF. IRON has 82 DOF, 22 per hand, 7 per arm, 6 per leg, plus torso and neck. The 22-DOF hands are the defining specification. Human hands have approximately 27 DOF (21 in the fingers, 6 in the wrist), IRON’s 22 DOF hands approach human hand kinematics more closely than any other Chinese humanoid. What this enables: fully independent finger control (rather than coupled finger joints), variable grip topology (can form any grip pattern a human hand can), and in-hand object manipulation (rotating and repositioning objects within the palm without releasing and re-grasping). For comparison, UBTECH Walker S2 has 5 kg payload per hand but its hand DOF is not separately published. Galbot G1 has 12 DOF per hand, roughly half IRON’s hand kinematics.
AI Integration: Turing AI Chip and XPeng ADAS Stack
IRON runs on XPeng’s Turing AI chip, an automotive-grade silicon platform designed for ADAS and now dual-purposed for humanoid robot inference. The chip processes 6 camera streams simultaneously at low latency, enabling IRON’s visual perception pipeline. The ADAS software stack, perception, prediction, planning, is adapted for bipedal locomotion and manipulation rather than vehicle navigation. XPeng has not disclosed the details of this adaptation, but the architectural similarity allows reuse of trained neural networks for visual scene understanding, which is the most compute-intensive element of robot AI.
XNGP Spatial Intelligence
XPeng’s XNGP (XPeng Navigation Guided Pilot), the vehicle ADAS system, includes advanced spatial mapping using camera-only perception. No LiDAR required. This camera-only spatial understanding is highly relevant to humanoid robotics: a robot that can build and maintain an accurate 3D world model from cameras alone is significantly cheaper to produce than one requiring LiDAR. XPeng has applied this camera-based spatial intelligence to IRON, enabling environment mapping and navigation without LiDAR hardware costs. This is a direct cost advantage from the automotive development investment.
Production Timeline
Mass production preparation target: 2027. XPeng has not disclosed unit pricing for IRON. Given XPeng’s automotive manufacturing experience and stated mass-market intent, pricing below CNY 300,000 ($41,000) at scale appears plausible, comparable to or below current premium humanoids. First customer deliveries are expected in research and pilot deployment roles before mass production readiness. XPeng’s existing dealer and service network covers China, providing a distribution infrastructure that pure robotics companies must build from scratch.
Competitive Context
IRON competes with AgiBot A2 (bipedal industrial), UBTECH Walker S2 (industrial with Siemens integration), and Unitree G1 (research and light commercial) in the bipedal humanoid space. IRON’s differentiation: highest DOF count, automotive AI chip, XPeng EV manufacturing infrastructure for mass production. The 2027 production target places IRON behind AgiBot and Unitree in commercialization timeline but potentially ahead if XPeng’s manufacturing scale delivers on its mass-market pricing potential.
Full China humanoid field: Top 40 China Robot Rankings 2025. Dexterity comparison with Galbot 12-DOF hands: Galbot G1 Spring Festival.
Sources
- XPeng: IRON humanoid robot announcement (May 2025)
- Interesting Engineering: XPeng IRON 82 DOF humanoid unveiled (May 2025)
- The Robot Report: XPeng AI Robot Iron 82 DOF features (2025)
- Pandaily: XPeng unveils IRON humanoid robot (May 2025)
