Unitree H2 Dedicated NVIDIA Jetson AGX Orin Computing Board
Genuine Unitree H2 Dedicated NVIDIA Jetson AGX Orin Computing Board — internal 275 TOPS AI development board for the H2 and H2 EDU. The high-tier CUDA path for onboard perception and embodied-AI inference.
- Unitree H2
- Unitree H2 EDU
About the Unitree H2 Dedicated NVIDIA Jetson AGX Orin Computing Board
The Unitree H2 Dedicated NVIDIA Jetson AGX Orin Computing Board is the high-tier CUDA development computer for the H2 family: an NVIDIA Jetson AGX Orin module assembly that mounts inside the robot's chassis and brings 275 TOPS of AI performance to the perception, planning and embodied-AI inference running on the robot itself. It takes the H2's AGX-class development position, and every H2 — the standard robot and the H2 EDU alike — reserves its development slots for exactly this kind of add-on compute.
It is the board for the workloads an entry-tier Jetson has to spill: multi-camera fusion at full frame rate, vision-language-action models that need tens of gigabytes of memory resident on the robot, learned policies that train offline and run onboard. No tethered GPU box, no cart-mounted workstation sharing a Wi-Fi link with the motion controller — the model runs where the sensors are.
Unlike the AGX modules Unitree offers for the B2 and H1 families, this is not an external pack bolted to the robot's back: it is an internal board assembly, sold as a single unit, that drops into the H2's own compute bay alongside a Core i7 development unit for the x86 side of your stack — so the whole pipeline rides inside the chassis instead of across the bench.
PC1 to PC4: Where the AGX Orin Board Goes
The H2's computing architecture is layered by position. PC1 is the motion-control computer — an Intel Core i5 running the platform itself: locomotion, balance, housekeeping. Above it, the official H2 developer documentation provisions up to three development computing units reserved for user code, and it maps each position to a class of board: the H2 EDU leaves the factory with a Core i7 development unit at PC2, the Jetson Orin NX (16GB) module is designated for PC4, and PC3 is the AGX-class position — the slot the architecture map assigns to the Jetson Thor (T5000) module. On the standard H2, those development positions are open.
The Unitree H2 Dedicated NVIDIA Jetson AGX Orin Computing Board is the PC3 assembly built on the Jetson AGX Orin: the same AGX module family as Thor, in the tier most perception and embodied-AI stacks run on today. With the board installed, the chassis's expansion ports route through to the development units, so cameras, LiDAR and peripherals connect to the H2's own connectors and land on your compute. Unitree notes that modules may be upgraded to newer versions with equivalent or higher performance between production runs — the quote we send confirms the exact unit in stock.
AGX Orin, Orin NX and Thor Are Tiers, Not Variants
The H2's development slots take four classes of board, and they are not interchangeable. The dedicated Intel Core i7 board is the x86 path — standard PC software and CPU-heavy processing. The Jetson Orin NX board is the entry CUDA tier: 100 TOPS with 16GB of memory, enough for a single perception pipeline. The Unitree H2 Dedicated NVIDIA Jetson AGX Orin Computing Board is the 275 TOPS tier with 64GB of memory — the step up when a model no longer fits the smaller module, when several high-resolution camera streams must be fused at full rate, or when a vision-language-action policy has to run onboard rather than on a tethered workstation. Above it, the Jetson Thor class at 2,070 TOPS is the frontier-model tier. Match the board to the workload: PC software leans Core i7, lean GPU pipelines lean Orin NX, memory-hungry models lean this AGX Orin board, frontier runs lean Thor.
One Board for the H2 and the H2 EDU
The Unitree H2 Dedicated NVIDIA Jetson AGX Orin Computing Board fits the whole H2 family: the standard H2 and the H2 EDU reserve the same internal development positions, and the AGX-class PC3 slot is present on either robot. On both it is an addition rather than a replacement — the EDU's factory development unit is a Core i7, so this board brings GPU compute the robot does not ship with, in the same slot family on every H2.
It is not the AGX module for other Unitree robots. The B2 and H1 families take external AGX packs that mount on the robot's back and are sold in 275 and 550 TOPS versions; the H2 board is an internal, single-unit assembly, and the two are not cross-fitted. The Go2, G1 and R1 platforms run their own compute — same module names, different robots, different parts.
Specifications
| Type | Unitree H2 dedicated NVIDIA Jetson AGX Orin computing board assembly |
| Module | 1 × NVIDIA Jetson AGX Orin |
| AI performance | 275 TOPS |
| GPU | 2048-core NVIDIA Ampere architecture GPU with 64 Tensor Cores (NVIDIA module spec) |
| CPU | 12-core Arm Cortex-A78AE v8.2 64-bit CPU (NVIDIA module spec) |
| Memory | 64GB 256-bit LPDDR5, 204.8 GB/s (NVIDIA module spec) |
| Power | 15W–60W (NVIDIA module spec) |
| Position | Internal — built into the robot chassis |
| Slot fit | PC3 development slot — the H2's AGX-class position |
| Compute allowance | Up to three development computing units per robot |
| In the box | 1 × Unitree H2 Dedicated NVIDIA Jetson AGX Orin Computing Board Assembly |
Shipping, Quality Guarantee & Board Care
The board assembly contains no lithium cells, so it ships as standard express freight rather than battery-class cargo — worldwide via DHL or FedEx, with the transit time confirmed in your quote.
Every Unitree H2 Dedicated NVIDIA Jetson AGX Orin Computing Board is 100% quality-inspected before it ships and packed as protected electronic cargo, so it arrives in perfect working order. From there it is an internal computer like any other — dry hands, gentle with the connectors, and it goes back to the same internal slot it came from.
Last updated: November 2026.
Frequently Asked Questions
Can't find what you're looking for? Contact our
