SENSING ROBOTIC VISION
Humanoid Robot
Reference Design Vision Solution
SENSING provides multi-view visual perception from head to hand for the Unitree H2 Plus and NVIDIA Isaac GR00T humanoid robot reference design.
Recommended Camera Configuration
SENSING provides composable GMSL2 camera modules for humanoid robot head perception, scene observation, and hand-view operation. The following recommendation describes a reusable development configuration for customers and does not represent the actual bill of materials of any specific production robot.
Recommended Vision Combination
Astra S56x
Dual 5MP · Global shutter · 10-bit RAW
Stereo vision and VLA inputView specs →Astra S36
Dual 3MP · HDR · YUV422
HDR stereo visionView specs →SHF3L
3MP · HDR 30fps · 196°
Close-range operation-area perceptionView specs →SHW3H
3MP · HDR 60fps · 118°
Hand operation-area perceptionView specs →Camera Recommendation Table
| Model | Type | Resolution | Shutter | Data format | Horizontal FOV | Recommended role | Recommended quantity |
|---|---|---|---|---|---|---|---|
| Astra S56x | Stereo | Dual 5MP | Global shutter | 10-bit RAW | 130° / 140° options | Head: Stereo vision and VLA input | 1 |
| Astra S36 (optional) | Stereo | Dual 3MP | Rolling shutter | YUV422 | 120° | Head: HDR stereo vision | 1 |
| SHF3L | Monocular | 3MP | Rolling shutter | YUV422 | 196° | Hand: HDR 30fps operation-area perception | Up to 4 |
| SHW3H (optional) | Monocular | 3MP | Rolling shutter | YUV422 | 118° | Hand: HDR 60fps operation-area perception | Optional |
- The table above is a development recommendation based on the current driver and TRD1 G2A interface capability. It does not represent the actual configuration of any specific production robot.
- Astra S56x and Astra S36 are stereo options that can be selected according to output format, shutter type, and application requirements.
- The actual camera count and port allocation should be determined by robot structure, bandwidth budget, power design, and software configuration.
Solution Positioning
FROM REFERENCE DESIGN TO DEPLOYABLE VISION
From a real reference design to a reusable robot vision foundation
This case shows how SENSING visual perception capability can be integrated into a humanoid robot reference design with Jetson Thor as the compute core. Based on this project experience, SENSING further provides a vision development foundation centered on Jetson AGX Thor, TRD1 G2A, and GMSL2 camera modules, helping robot development teams reuse the full path from visual capture and platform adaptation to edge deployment.
Unitree H2 Plus and the NVIDIA Isaac GR00T humanoid robot reference design are the real application case introduced on this page. SENSING provides the visual perception solution in this reference case. The TRD1 G2A and SENSING camera combinations listed here describe a reusable vision development foundation and integration workflow; they do not represent the complete bill of materials of the H2 Plus robot.
Case Study: H2 Plus and Isaac GR00T
The humanoid robot reference design connects the robot body, compute and software stack, and visual perception into a unified development platform. SENSING focuses on the visual perception layer, providing camera modules, platform adaptation, and vision software support to deliver stable multi-view image input for robots.
ROBOT BODY
Unitree H2 Plus
Provides the next-generation bionic humanoid robot body and the carrier for key vision nodes such as the head and hands.
View Unitree official page →COMPUTE & SOFTWARE
Jetson Thor + Isaac GR00T
Provides onboard AI compute capability and an open software stack connecting data collection, training, inference, and real-world deployment.
VISUAL PERCEPTION
SENSING
Provides camera modules, platform adaptation, and vision software support from head to hand to build multi-view perception capability.
View SENSING official case →Capability Loop from Perception to Operation
Multi-view capture
Head stereo vision, wide-angle vision, and hand ultra-wide vision jointly cover the environment and operation area.
Stable transmission and synchronization
GMSL2 coaxial links feed multiple camera streams into the compute platform and support multi-channel video hardware synchronization.
On-device understanding and inference
Jetson Thor receives visual streams and provides input for depth processing, recognition, and VLA models.
Fine manipulation
Stable visual input is connected to the robot software stack to support navigation, grasping, and autonomous decision-making.
Vision Development Foundation
SENSING VISION FOUNDATION
A reusable multi-camera vision development foundation for customers
SENSING provides a development configuration centered on Jetson AGX Thor, TRD1 G2A, and GMSL2 camera modules to help robotics teams quickly validate multi-view perception links.
- Supports up to 8 GMSL camera inputs
- Supports RAW and YUV422 visual input
- Covers head stereo vision, scene perception, and hand-operation viewpoints
- Provides driver, platform adaptation, and integration support
