Unitree H1-2 Universal Humanoid Robot
In stock
- BRAND:
- UNITREE ROBOTICS
- PART #:
- H1-2
- ORIGIN:
- China
- AVAILABILITY:
- SUBJECT TO AVAILABILITY
- SKU:
- Unitree-H1-2
Unitree H1-2 Universal Humanoid Robot
Continuous Physical Platform: 180 cm, 47 kg
The H1-2 retains the H1's physical body characteristics: approximately 180 centimeters tall, approximately 47 kilograms, and the same industrial skeletal aluminum alloy frame without cosmetic body panels. This continuity means all the H1's physical advantages apply equally to the H1-2: full human-scale proportions for deployment in human-environment infrastructure, the 189 N·m/kg torque density that enables 3.3+ m/s bipedal running, and the lightweight construction that makes the 35-kilogram lighter platform than the H2 feasible without sacrificing height.
7-DOF Arms: The Manipulation Upgrade
The H1-2's transition from 5-DOF to 7-DOF arms is the key design change that enables manipulation. The additional two DOF are at the wrist — adding pitch (±92.5° confirmed from the G1 EDU Ultimate tier, the same wrist specification Unitree uses across their platforms) and yaw to the H1's shoulder-elbow configuration. With only 5 DOF, the H1's arm has a kinematic singularity problem: there exist target hand orientations that the arm cannot reach, requiring the robot to significantly reposition its whole body to bring the hand to certain approach angles. The 7-DOF arm eliminates these singularities, enabling the arm to reach any target orientation within its workspace from the same body position. For manipulation tasks — grasping objects from a shelf, assembling components, handing objects to humans — this kinematic completeness is essential.
Dexterous Hand Options
The H1-2 supports multiple optional dexterous hand configurations across the same spectrum as the G1 EDU manipulation configurations:
Basic grippers: Two-finger force-controlled grippers for pick-and-place tasks.
Force-controlled three-finger hands: Dex3-1 configuration with 7 active DOF (thumb 3, index 2, middle 2) for three-finger grasping across power, precision, and tripod configurations.
Optional tactile three-finger hands: Dex3-1 with integrated tactile arrays for contact-aware manipulation, slip detection, and fragile object handling.
Five-finger dexterous hands: For the full human grasp taxonomy including tool use and complex multi-finger coordination.
Technology and Specifications
Full H1-2 Specifications
| Specification | Value |
|---|---|
| Height | ~180 cm |
| Weight | ~47 kg |
| Total DOF | Up to 27 (with 7-DOF arms) |
| Arm DOF | 7 per arm (bilateral) |
| Leg DOF | 6 per leg (bilateral) |
| Waist DOF | 3 |
| Maximum Knee Torque | 360 N·m |
| Hip/Waist Torque | 220 N·m |
| Ankle Torque | ~75 N·m |
| Arm Shoulder Torque | ~120 N·m |
| Arm Elbow Torque | ~120 N·m |
| Arm Wrist Torque | ~30 N·m |
| Platform CPU | Intel Core i5 |
| User Dev CPU | Intel Core i7 |
| Optional AI Module | NVIDIA Jetson Orin NX 16GB (100 TOPS) |
| Perception | 3D LiDAR + depth camera |
| Battery | Quick-swappable |
| Dexterous Hands | Optional (multiple configurations) |
| Maximum Running Speed | 3.3+ m/s |
| Locomotion Policy | RL-trained, sim-to-real (Isaac Gym) |
The Dual Intel Core CPU Architecture: Why It Matters
The H1-2's most consequential technical upgrade for research users is not the 7-DOF arms — it is the dual CPU design that separates platform computing from user development computing.
The problem with single-CPU humanoids for research: A robot's real-time control loops — joint position/velocity/torque commands, sensor fusion, balance computation — must run at fixed frequencies (typically 500 Hz to 1000 Hz) with hard timing guarantees. Any process that causes CPU scheduling latency exceeding the control loop period can cause the robot to lose balance and fall. Research AI models — neural network inference, vision processing, language model queries — are computationally expensive and non-deterministic in their timing, creating exactly the kind of scheduling interference that real-time control cannot tolerate.
The H1-2's solution: The Intel Core i5 handles all platform functions under hard real-time scheduling with dedicated CPU resources, maintaining control loop timing regardless of what the user compute environment is doing. The Intel Core i7 runs user development code under a standard Linux environment where researchers can run PyTorch models, ROS 2 nodes, vision pipelines, and custom AI policies without affecting control timing.
This dual-CPU separation is the same architectural pattern as the G1 EDU's 8-core CPU plus Jetson Orin design, elevated to Intel Core x86 architecture for the H1-2's larger user development requirements. For research teams implementing complex manipulation policies, multi-modal AI pipelines, and simulation-based testing alongside physical hardware, the dedicated i7 development environment is a meaningful practical improvement over the H1's single-CPU constraint.
The Jetson Orin NX option extends the H1-2's compute further for GPU-accelerated AI: when the optional Jetson Orin NX 16GB (100 TOPS) is added, the i7 serves as a CPU compute environment while the Jetson Orin handles neural network inference at GPU speeds — the full three-tier compute architecture of platform control, CPU AI, and GPU AI operating simultaneously.
Locomotion: Continuous M107 Motor Performance
The H1-2 retains the H1's M107 proprietary motor design with the same joint torque specifications: 360 N·m at the knee, 220 N·m at hip and waist, 75 N·m at ankle. The 7-DOF arm upgrade changes the arm torque distribution to include wrist joints at 30 N·m each, with shoulder and elbow maintained at 120 N·m.
This torque continuity means the H1-2's locomotion performance is equivalent to or better than the H1's — the 7-DOF arms add slightly more distal arm mass from the wrist actuators, but the platform's overall locomotion capability is maintained by the same leg and waist actuator specification.
Applications and Use Cases
Full-Spectrum Humanoid Research: Locomotion and Manipulation
The H1-2 is the appropriate platform for research programs that require high-performance full-size bipedal locomotion and dexterous arm manipulation in a single robot. Where the H1 constrains research to locomotion due to its limited arm configuration, and the G1 EDU constrains research to a smaller-scale platform, the H1-2 enables research that requires both full human-scale locomotion performance and 7-DOF arm manipulation capability simultaneously — loco-manipulation research at full human scale.
Advanced Industrial Deployment Prototyping
For manufacturing and logistics organizations evaluating full-size humanoid robots for industrial deployment, the H1-2's combination of world-record running locomotion capability, 7-DOF arms, optional dexterous hands, and dual-CPU user development architecture provides a research-grade pilot platform for developing and evaluating the manipulation and navigation policies that would be needed for production deployment.
AI Policy Development for Whole-Body Humanoid Control
The H1-2's Intel Core i7 user development CPU with optional Jetson Orin NX provides the development infrastructure for building whole-body humanoid AI policies — learning algorithms that coordinate the full 27-DOF body (arms, legs, waist) for tasks combining dynamic locomotion with simultaneous manipulation. The UnifoLM-WMA-0 world model AI framework, open-sourced by Unitree in September 2025, provides a foundation for this kind of whole-body policy development on the H1-2 platform.
Enterprise Research and Government Laboratory Programs
BotInfo.ai specifically identifies the H1-2 for "enterprise research labs and government agencies studying full-size humanoid manipulation" — the institutional segment that needs the H1's scale and locomotion reputation combined with the manipulation capability that smaller platforms like the G1 EDU cannot provide at full human scale.
Comparison: H1-2 vs. H1 vs. H2
| Feature | H1 | H1-2 | H2 Commercial |
|---|---|---|---|
| Arm DOF | 5 per arm | 7 per arm | 7 per arm |
| Compute | 8-core CPU | Intel Core i5 + i7 | 2,070 TOPS chip |
| Optional AI | None | Jetson Orin NX (100 TOPS) | N/A |
| Dexterous Hands | Optional (limited) | Optional (full range) | Separate purchase |
| Weight | 47 kg | 47 kg | ~70 kg |
| Bionic Face | No | No | Yes |
| Best For | Pure locomotion research | Locomotion + manipulation | Commercial deployment, HRI |
Frequently Asked Questions (FAQ)
What is the Unitree H1-2? The Unitree H1-2 is an enhanced version of the H1 full-size universal humanoid robot from Unitree Robotics, introducing 7-DOF arms (adding wrist pitch and yaw), a dual Intel Core CPU architecture (Intel Core i5 for platform function and Intel Core i7 for user development), and an optional NVIDIA Jetson Orin NX 16GB (100 TOPS) module. It retains the H1's 180-centimeter height, 47-kilogram weight, 360 Newton-meter knee torque, world-class locomotion capability, and quick-swappable battery. Optional dexterous hand configurations from two-finger grippers through five-finger hands are supported. Price is approximately $99,900 to $128,900.
What are the key improvements of the H1-2 over the H1? The H1-2 makes three key improvements: (1) 7-DOF arms replacing the H1's 5-DOF base arms, adding wrist pitch and yaw for full arm workspace coverage and manipulation capability; (2) dual Intel Core CPU architecture (Intel Core i5 for platform control, Intel Core i7 for user development) replacing the H1's single CPU, enabling complex AI model development without competing with real-time control timing; and (3) optional NVIDIA Jetson Orin NX 16GB (100 TOPS) for GPU-accelerated AI inference. Joint torques are also expanded to specify arm-level values: shoulder and elbow at 120 N·m, wrist at 30 N·m.
What is the Intel Core i7 user development CPU used for in the H1-2? The Intel Core i7 in the H1-2 provides a dedicated x86 computing environment for researcher development — running custom AI models, ROS 2 nodes, manipulation policy networks, vision pipelines, and application code without competing with the Intel Core i5's real-time platform control functions. This separation ensures the robot's locomotion and balance control (running at 500 to 1,000 Hz with hard timing guarantees) is not disrupted by computationally intensive user AI code. The Intel x86 architecture enables standard Linux development tools, Python, PyTorch, and ROS 2 to run without cross-compilation overhead.
How does the Unitree H1-2 compare to the H2 for research? The H1-2 ($99,900–$128,900) and H2 ($40,900 commercial) address different research profiles. The H2 is lighter (70 kg versus 47 kg), has a bionic face for HRI, uses a 2,070-TOPS main AI chip, and began shipping in April 2026 — targeting commercial deployment and social HRI. The H1-2 has a proven locomotion record (3.3 m/s world speed record), the specialized Intel Core i5/i7 dual-CPU architecture for research development workflows, is 23 kg lighter enabling higher dynamic performance, and has a longer institutional track record from the H1 family's 2024 launch. For locomotion research, manipulation development at full scale, and enterprise research requiring the H1 family's performance reputation, the H1-2 is the appropriate choice.
Summary
The Unitree H1-2 Universal Humanoid Robot represents the bridge between the H1's world-record locomotion platform and the full-capability research humanoid that advanced manipulation and embodied AI research programs require — upgrading to 7-DOF arms for complete wrist articulation, adding a dedicated Intel Core i7 user development CPU alongside the platform-function i5, supporting optional NVIDIA Jetson Orin NX (100 TOPS) for GPU-accelerated AI, and enabling the full spectrum of optional dexterous hand configurations from two-finger grippers through five-finger tactile systems. Priced at $99,900 to $128,900, qualifying for federal research grant funding, and backed by institutional procurement support through North American authorized dealers including BotInfo.ai, the H1-2 provides research institutions with a commercially mature, full-scale humanoid platform that combines the H1 locomotion legacy with the manipulation and development compute infrastructure that the next generation of embodied AI research demands.