
The 275 TOPS Unitree H1 AGX module pairs one Jetson AGX Orin assembly with a rear external mount, adding 275 TOPS of CUDA compute to the H1 and H1-2 for AI, vision and autonomy workloads.
The Unitree H1 AGX module is the heavy-duty external AI compute upgrade for Unitree's full-size humanoids: one or two NVIDIA Jetson AGX Orin computing board assemblies in a single enclosure that mounts on the robot's back and turns a CPU-only machine into a CUDA-equipped autonomous platform. It comes in two versions — the 275 TOPS module built around a single Jetson AGX Orin, and the 550 TOPS module that pairs two of them for the heaviest perception stacks — and both versions fit the H1 and the H1-2 alike.
Inside every H1, an Intel Core i5 runs the platform and an Intel Core i7 is reserved for your code — but neither is a GPU. When a stack outgrows CPU compute — multi-camera perception, learned navigation, VLA-class models, real-time sensor fusion — the Unitree H1 AGX module adds the missing layer without opening the body: it mounts on the rear exterior, the internal computers keep their roles, and it comes back off as easily as it went on when a robot changes jobs.
Choose the 275 TOPS version for demanding-but-standard autonomy and machine-vision workloads, or the 550 TOPS dual version when several large models must run at once. The Unitree H1 AGX module arrives fully assembled — one enclosure, one mount, no integration project — and it is the same external module for the H1 and the H1-2, so a mixed fleet of both robots can share one module pool.
Unitree's H1 specification lists the standard compute as an Intel Core i5 for the platform and a Core i7 reserved for user development, with GPU options named as extras. That pair covers locomotion, control and housekeeping well, but it is CPU compute — there is no CUDA GPU in the standard machine. The Unitree H1 AGX module closes exactly that gap: NVIDIA Jetson AGX Orin compute mounted on the robot's back, adding 275 or 550 TOPS of GPU-accelerated AI performance for camera-heavy perception, deep-learning inference and multi-sensor fusion.
Because the module lives outside the body, nothing inside is disturbed: the internal computers keep their roles, and the GPU layer rides on top. When an H1 is reassigned, or the module needs to move to another unit in the fleet, an external mount makes that a hardware swap rather than a teardown.
Both versions of the Unitree H1 AGX module mount the same way and fit the same robots — the H1 and the H1-2 — and the difference is how much CUDA compute rides on the back. The 275 TOPS module carries a single Jetson AGX Orin assembly; the 550 TOPS module carries two, doubling the GPU budget for stacks that run several large models at once.
The H1 offers two external GPU paths, and they are not interchangeable. The Unitree H1 dedicated computing module is built on NVIDIA Jetson Orin NX — 100 TOPS per board, up to two boards, on its own page. The Unitree H1 AGX module is built on the larger Jetson AGX Orin — 275 or 550 TOPS in one rear-mounted enclosure. The two are complements, not competitors: a fully loaded H1 pairs two Orin NX boards with a 550 TOPS AGX module for roughly 750 TOPS of combined AI compute. Match the module to the workload — Orin NX for standard perception stacks, AGX for the heaviest multi-model rigs.
The Unitree H1 AGX module is one shared external part for both humanoids: the H1 and the H1-2 take the same rear-mounted module under the same SKU, so a mixed fleet of the two robots can share a single module pool, and a module bought for one robot moves to the other when the workload does. The module mounts on the back exterior of either robot — no internal disassembly, and the body is not modified.
It is not a cross-model part beyond that pair, though. The B2 family takes its own AGX module — the Unitree AGX module for B2/B2-W — and the G1 takes AGX Orin and AGX Thor backplane docks. Note that the smaller Orin NX computing module is an H1-only part and does not fit the H1-2; this AGX module is the external GPU path for both humanoids.
| 275 TOPS | 550 TOPS | |
|---|---|---|
| Module | 1 × NVIDIA Jetson AGX Orin | 2 × NVIDIA Jetson AGX Orin (dual assembly) |
| AI performance | 275 TOPS | 550 TOPS (2 × 275 TOPS) |
| CPU | 12-core Arm Cortex-A78AE (NVIDIA module spec) | 2 × 12-core Arm Cortex-A78AE (NVIDIA module spec) |
| GPU | 2048-core NVIDIA Ampere with 64 Tensor Cores (NVIDIA module spec) | 2 × 2048-core NVIDIA Ampere with 64 Tensor Cores (NVIDIA module spec) |
| Memory | 64GB 256-bit LPDDR5 (NVIDIA module spec) | 2 × 64GB 256-bit LPDDR5 — 128GB total (NVIDIA module spec) |
| Mounting | External — mounts on the robot's back | External — mounts on the robot's back |
| Per robot | 1 module | 1 module |
| Fits | Unitree H1 · Unitree H1-2 | Unitree H1 · Unitree H1-2 |
| In the box | 1 × Unitree H1/H1-2 AGX Module — 275 TOPS (single Jetson AGX Orin computing board assembly) | 1 × Unitree H1/H1-2 AGX Module — 550 TOPS (dual: 2 × Jetson AGX Orin computing board assemblies) |
The module 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 H1 AGX module is 100% quality-inspected before it ships and packed as protected electronic cargo, so it arrives in perfect working order. From there it is sensitive electronics like any other: dry hands, gentle with the connectors, and keep the module dry once it is on the robot.
Last updated: October 2026.
Can't find what you're looking for? Contact our