Unitree H2 Plus Review: Specs, Price and Verdict
Unitree H2 Plus pairs 75 degrees of freedom with NVIDIA GR00T, but no public price or independent task results support a broader buying case.
This Unitree H2 Plus review finds a credible research platform for well-funded labs, not a proven worker. Unitree supplies a human-scale H2 chassis, Sharpa supplies tactile five-finger hands, and NVIDIA adds Jetson Thor with Isaac GR00T. No independent H2 Plus task benchmark or public price supports a broader buying case.[1][2][3]
NVIDIA says the system will be available from Unitree in late 2026. This is a documentary review, not a hands-on test. We found detailed component specifications and public developer material.[1][2][5] None of the reviewed sources provides an independent H2 Plus task benchmark, duty-cycle test or service record.[1][2][6]
What are the Unitree H2 Plus specs?
The Unitree H2 Plus is 1,820 mm tall, weighs about 70 kg with its battery and has 31 powered body joints. Two Sharpa Wave hands add 44 active degrees of freedom, taking the manufacturer's total to 75. Jetson Thor T5000 supplies onboard AI compute.[1][2][5]
Each leg has six powered joints, and each arm has seven.[2][5] The waist has three. The head has two.[2][5] The joints differ in role and load capacity, so the total alone says little about task ability.
| Specification | Published value | Evidence status |
|---|---|---|
| Standing size | 1,820 by 456 by 218 mm | Unitree specification.[2] |
| Weight with battery | About 70 kg | Unitree specification.[2] |
| Body joints | 31 | Unitree specification and developer documentation.[2][5] |
| Total body and hand degrees of freedom | 75 | Unitree and NVIDIA specification.[1][2] |
| Arm payload | About 7 kg rated, 15 kg peak | Manufacturer figures that vary by configuration and scenario.[1][2] |
| Peak joint torque | 120 Nm in an arm joint, 360 Nm in a leg joint | Manufacturer maximums, not sustained output figures.[1][2] |
| Main AI computer | Jetson AGX Thor T5000, 2,070 FP4 TFLOPS, 128 GB unified memory | NVIDIA and Unitree specification.[1][2] |
| Sensors | Head stereo camera, wrist cameras and an inertial measurement unit | NVIDIA and Unitree specification.[1][2] |
| Battery | 15 Ah, 0.972 kWh, about three hours | Manufacturer estimate without a published workload or test method.[1][2] |
| Availability | Planned for late 2026 | NVIDIA announcement dated 1 June 2026.[1] |
| Price | Contact sales | Unitree store listing.[3] |
The base H2 developer guide identifies the rear electrical interfaces. It separates motion control from user computers and documents optional Jetson Thor and Orin NX modules.[5] That should help with setup, although delivered H2 Plus units may differ.
Our Jetson Thor vs AGX Orin guide explains why teams should compare memory, power and representative workloads instead of dividing the headline AI figures.
Our view: H2 Plus earns attention because its documented hardware and software meet in one package, not because 75 is a magical number.[1][2][5]
How does NVIDIA Isaac GR00T work on H2 Plus?
NVIDIA's Isaac GR00T workflow lets researchers record demonstrations in Isaac Teleop, build and test policies in Isaac Sim and Isaac Lab, then run them through Isaac ROS and Jetson Thor. The public example uses Unitree G1, not H2 Plus.[1][6]
NVIDIA published the public GR00T 1.7 workflow in July 2026.[6] It covers scene setup, collecting teleoperation data, converting it to LeRobot, post-training and evaluating a policy.[6]
Buyers should keep that distinction clear. A working G1 tutorial supports confidence in the software path. It does not test H2 Plus. The larger platform has different hands, sensors, mass and joint limits.[1][2][4]
How much does Unitree H2 Plus cost?
The Unitree H2 Plus does not have a public price on its product page. Unitree asks buyers to contact sales and pay customs duties separately. The page also omits setup, spares and support costs, so prospective labs need an itemized quote before comparing its total cost with another research humanoid.[3]
For context, Unitree lists the base H2 at $29,900 before tax and shipping. Unitree says buyers cannot develop on that model. H2 EDU requires a quote.[7] The smaller base G1 costs $13,500 before tax and shipping, but Unitree also bars buyers from developing on it. G1 EDU requires a quote.[4]
Those are not direct alternatives to H2 Plus. A public H2 Plus price still matters. The package adds tactile hands, wrist cameras and developer compute that the headline chassis prices do not capture.[1][2][7]
Safety disclosure is basic. H2 Plus has a remote emergency stop.[1][2]
Unitree warns users to keep people at a sufficient distance from the robot.[2] The reviewed product pages name no safety certification or completed conformity review for the full system.[2][3]
Our legged robot safety standards guide separates checks for the robot, its application and its movements.
Reliability evidence is thinner. A quick-release battery and documented interfaces help teams plan repairs, but neither predicts uptime or repair turnaround.[2][5]
*Scores are Robotsider editorial calls based on the cited documents, not measurements from tests. Low evidence scores reflect the missing public price, independent H2 Plus benchmarks and service history.*
Pros
- The package joins a 31-joint human-scale chassis, tactile hands and Jetson Thor.[1][2]
- Unitree publishes a useful H2 interface and onboard computer guide.[5]
- NVIDIA's GR00T stack captures data, simulates tasks, trains models, evaluates policies and deploys them.[1][6]
Cons
- Unitree publishes no H2 Plus price or total cost information.[3]
- The cited test figures come from the manufacturers, without an independent H2 Plus benchmark.[1][2]
- The late 2026 delivery date is a vendor forecast.[1]
Unitree H2 Plus vs H2 vs G1: which fits?
Choose H2 Plus for a human-scale NVIDIA research stack.[1][2] The base H2 suits motion demonstrations, but Unitree bars buyers from developing on it.[7]
Base G1 costs less and works at a smaller scale. Unitree bars buyers from developing on it too.[4] These models are not direct alternatives.
| Decision point | H2 Plus | Base H2 | Base G1 |
|---|---|---|---|
| Best fit | Human-scale hand and arm research on the NVIDIA stack.[1][2] | Full-size motion demonstrations without user development access.[7] | Lower-cost humanoid hardware without user development access.[4] |
| Body size and mass | 1,820 mm, about 70 kg.[2] | 1,820 mm, about 70 kg.[7] | 1,320 mm, about 35 kg.[4] |
| Body joints | 31.[2] | 31.[7] | 23.[4] |
| Hands | Two Sharpa Wave tactile hands, 22 active degrees of freedom each.[1][2] | No dexterous hand listed.[7] | No dexterous hand listed.[4] |
| Developer compute | Intel Core i7 plus Jetson Thor T5000 in the system design.[1][2][5] | Unitree bars buyers from developing on it.[7] | Unitree bars buyers from developing on it.[4] |
| Published price | None.[3] | $29,900 before tax and shipping.[7] | $13,500 before tax and shipping.[4] |
| Current purchase status | Contact sales, with late 2026 availability announced.[1][3] | Listed for purchase.[7] | Listed at $13,500.[4] |
A research team that needs this hardware should ask Unitree for an itemized quote and an exact list of included parts. It should also agree warranty terms and a delivery test before committing funds.[1][2][3]
Teams that mainly want to learn the GR00T workflow can simulate it or use NVIDIA's documented G1 path.[6] Buying a base H2 or base G1 for that purpose would be a mistake. Unitree says buyers cannot develop on either base model.[4][7]
Is Unitree H2 Plus worth buying?
Unitree H2 Plus is worth shortlisting for controlled hand and arm research with NVIDIA GR00T. It is not ready for an evidence-based factory purchase. Buyers still lack an independent task benchmark, a public price and a field reliability record for delivered H2 Plus units.[1][2][3]
The 75-degree-of-freedom system has human-scale reach, tactile hands, substantial onboard compute and software that researchers can inspect.[1][2][6]
The buying case is incomplete. There is no public price, no independent H2 Plus task benchmark and no field reliability record in the reviewed evidence.[1][2][3]
Make the purchase conditional on a written list of included parts, delivery criteria and support terms. Anyone buying for immediate factory work should wait for delivered units and repeatable task results.[1][2][3] Our humanoid robots in warehouses explainer shows how narrow the documented jobs remain in 2026.
Unitree H2 Plus: price, specs and buying questions
Is the Unitree H2 Plus worth buying?
It is worth shortlisting for a funded research program that needs human-scale hand and arm work with NVIDIA GR00T. Factory buyers should wait for independent task results, service records and firm commercial terms.[1][2][3]
How much does the Unitree H2 Plus cost?
Unitree does not publish a price on the H2 Plus product page. It asks buyers to contact sales and says customs duties are extra. Request an itemized quote covering the robot, included components, setup, spares and support.[3]
What are the main Unitree H2 Plus specs?
Unitree lists a height of 1,820 mm and weight of about 70 kg. It has 31 body joints, 75 total degrees of freedom and Jetson Thor T5000 compute. Unitree estimates about three hours of battery life but does not state the workload.[1][2]
Does Unitree H2 Plus support NVIDIA Isaac GR00T?
Yes. NVIDIA and Unitree present H2 Plus as an Isaac GR00T system with Isaac Teleop, Isaac Sim, Isaac Lab, GR00T models, Isaac ROS and Jetson Thor. The public workflow reviewed here demonstrates Unitree G1, not H2 Plus.[1][6]
How does Unitree H2 Plus compare with H2 and G1?
H2 Plus combines the human-scale H2 body with tactile hands, Jetson Thor and the NVIDIA software path. Base H2 and base G1 cost less, but Unitree bars buyers from developing on them. G1 is also smaller and lighter.[1][2][4][7]
Sources
[1] https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Announces-NVIDIA-Isaac-GR00T-Reference-Humanoid-Robot-for-Academic-Research/ [2] https://www.unitree.com/H2plus/ [3] https://shop.unitree.com/products/unitree-h2-plus [4] https://www.unitree.com/g1 [5] https://support.unitree.com/home/en/H2_developer/About_H2 [6] https://developer.nvidia.com/blog/develop-humanoid-robot-policies-end-to-end-with-nvidia-isaac-gr00t/ [7] https://www.unitree.com/H2/
Sources
- https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Announces-NVIDIA-Isaac-GR00T-Reference-Humanoid-Robot-for-Academic-Research/
- https://www.unitree.com/H2plus/
- https://shop.unitree.com/products/unitree-h2-plus
- https://www.unitree.com/g1
- https://support.unitree.com/home/en/H2_developer/About_H2
- https://developer.nvidia.com/blog/develop-humanoid-robot-policies-end-to-end-with-nvidia-isaac-gr00t/
- https://www.unitree.com/H2/