NVIDIA AGX Thor Backplane Expansion Dock for Unitree G1
NVIDIA AGX Thor Backplane Expansion Dock for Unitree G1 — the rear-mounted training-grade compute upgrade: Jetson AGX Thor module, up to 2,070 TOPS, 128GB LPDDR5X, backplane dock and cables included. Fits G1 and G1-D.
- NVIDIA Jetson AGX Thor module included — up to 2,070 TOPS
- 14-core Arm Neoverse-V3AE CPU, Blackwell GPU with 2560 CUDA + 96 Tensor cores
- 128 GB 256-bit LPDDR5X (273 GB/s)
- Rear backplane dock purpose-built for the G1's back expansion bay
- Fits Unitree G1 and G1-D
About the NVIDIA AGX Thor Backplane Expansion Dock for Unitree G1
The NVIDIA AGX Thor Backplane Expansion Dock for Unitree G1 is the rear-mounted compute upgrade that puts training-grade edge AI inside the G1 humanoid instead of on a cart beside it: a full NVIDIA Jetson AGX Thor module seated on a purpose-built backplane that bolts into the G1's back expansion bay, delivering up to 2,070 TOPS of on-robot computing for large vision-language-action models, multi-camera fusion and reinforcement-learning research. Sold under Unitree part number G1-AGX-THOR-DOCK, the nvidia agx thor backplane expansion dock for unitree g1 arrives as a complete kit — the AGX Thor module itself, the dedicated backplane dock and the cable set.
Out of the box, a stock G1 relies on an external workstation or laptop for the heavy compute: policy inference, camera processing and SLAM all run over a tether. The NVIDIA AGX Thor Backplane Expansion Dock for Unitree G1 removes that tether and goes far past it. The Jetson AGX Thor module pairs a 14-core Arm Neoverse-V3AE CPU with a Blackwell GPU of 2560 CUDA cores and 96 fifth-generation Tensor cores, backed by 128 GB of 256-bit LPDDR5X memory at 273 GB/s. At up to 2,070 TOPS, the module runs large vision-language-action models, multi-camera perception pipelines and reinforcement-learning training loops on the robot itself.
The NVIDIA AGX Thor Backplane Expansion Dock for Unitree G1 fits the standard G1 and the G1-D. Installation is a rear-panel job: the backplane dock seats in the G1's back expansion bay and the module connects to the robot over the dock's expansion interfaces. If you run a different G1 configuration or an older revision, ask us to confirm fit for your robot before ordering — and if your workloads do not need this much headroom, the same dock family is available with NVIDIA AGX Orin for up to 275 TOPS.
What the NVIDIA AGX Thor Backplane Expansion Dock for Unitree G1 Adds
A G1 straight from the box is a body waiting for a brain: the robot's own controllers handle motion, but perception, policy inference and mapping are expected to run on an external workstation, tethered over Ethernet. That works on a lab bench and fails in the field — the cable is the first thing that breaks, and the PC is the first thing that does not fit through the door. The NVIDIA AGX Thor Backplane Expansion Dock for Unitree G1 closes that gap by moving a training-grade brain onto the robot.
The Jetson AGX Thor module inside delivers up to 2,070 TOPS from a 14-core Arm Neoverse-V3AE CPU and a Blackwell GPU with 2560 CUDA cores and 96 fifth-generation Tensor cores. The 128 GB of 256-bit LPDDR5X is the number that changes what the G1 can do: it is enough unified memory to keep a large vision-language-action model and the full multi-camera perception stack resident at the same time, so end-to-end policies run on board instead of round-tripping to a server. Where an Orin-class dock covers perception and moderate VLM inference, Thor is the tier for GR00T-class humanoid foundation models, on-robot policy training and heavy multi-camera fusion.
Because the dock lives in the G1's back expansion bay rather than on an external mount, the robot keeps its full mobility: no tether, no trailing cable set, nothing to snag on furniture, door frames or lab equipment. The 40–130 W module power envelope is supplied through the dock, so there is no separate supply to mount or waterproof.
AGX Thor or AGX Orin — Which Dock for Your G1?
Unitree's backplane expansion dock family for the G1 takes two NVIDIA modules. The NVIDIA AGX Thor Backplane Expansion Dock for Unitree G1 delivers up to 2,070 TOPS — the right class for large vision-language-action models, training-grade reinforcement learning and heavy multi-camera fusion on a mobile humanoid. The AGX Orin variant delivers up to 275 TOPS for multi-camera perception, lighter VLM inference and RL policies. The mounting is the same; only the module changes. Tell us what you plan to run and we will confirm the right dock for your workload.
Which Robots Does the NVIDIA AGX Thor Backplane Expansion Dock for Unitree G1 Fit?
The NVIDIA AGX Thor Backplane Expansion Dock for Unitree G1 fits the standard G1 and the G1-D, seated in the robot's back expansion bay — the dock is purpose-built for that bay's connector and mounting pattern, and the cable set in the box links the dock to the robot's internal interfaces.
If you run a different G1 revision or a customised configuration, ask us to confirm fit for your robot before ordering — a photo of your robot's back panel is enough for us to check.
- Unitree G1 — supported, back expansion bay
- Unitree G1-D — supported, back expansion bay
- Other G1 revisions or custom configurations — ask us to confirm fit before ordering
Specifications
| Compute module | NVIDIA Jetson AGX Thor ×1 (sold under Unitree part no. G1-AGX-THOR-DOCK) |
| AI performance | Up to 2,070 TOPS |
| CPU | 14-core Arm Neoverse-V3AE 64-bit |
| GPU | Blackwell architecture — 2560 CUDA cores + 96 fifth-gen Tensor cores |
| Memory | 128 GB 256-bit LPDDR5X, 273 GB/s |
| Module power | 40–130 W configurable |
| Mounting | Rear backplane expansion dock, purpose-built for the G1 back expansion bay |
| Compatibility | Unitree G1, Unitree G1-D |
| In the box | 1 × NVIDIA Jetson AGX Thor module · 1 × G1/G1-D dedicated backplane expansion dock · 1 × Cable set |
Shipping & Handling
Every NVIDIA AGX Thor Backplane Expansion Dock for Unitree G1 is 100% quality-inspected before it ships, so it arrives in perfect working order. The kit ships in rigid, shock-absorbing packaging with the AGX Thor module seated and protected in the dock.
The module and dock contain electrostatic-sensitive components — handle the backplane by its edges after unpacking, and ground yourself before seating or reseating the module. A spare G1 battery or charger can travel in the same parcel if you are stocking up for a deployment.
Last updated: September 2026.
Frequently Asked Questions
Can't find what you're looking for? Contact our
