← All products
Vision Perception Module
Embodied RGB-D module with integrated vision-brain controller
The RGB-D module combining 0.1mm accuracy, 157 TOPS on-device compute, a native VLA interface and auto-calibration — purpose-built for embodied AI with an exceptionally high moat.
Contact
Key Specifications
Range
0.25m – 4m
Accuracy
0.1mm @ 0.25m
FOV
68° × 54°
RGB
3.2MP global shutter
AI Compute
157–2070 TOPS (optional)
Eye Safety
Eye Safety Class 1
Key Advantages
1
Sub-mm accuracy
0.5mm depth; one module covers close manipulation to far-field sensing across 0.25–4m.
2
Robust spatial decoding
A proprietary spatial-decoding algorithm tracks reliably even under fast motion.
3
High on-device AI
Onboard NVIDIA Ampere GPU + 32 Tensor Cores run VLA inference at the edge.
4
Auto-calibration
Optical design minimizes thermal drift; stable from -20 to 60℃ without manual calibration.
5
Native VLA interface
A standard interface for vision-language-action models makes robots plug-and-play.
6
Patented IR projection
Point-cloud density is 5–7× that of conventional speckle techniques.
Competitive Comparison
| 桐识 ArgusPro | 竞品A | 竞品B | 竞品C | |
|---|---|---|---|---|
| Accuracy | 0.1mm@0.25m | ~2mm@0.5m | ~1mm@0.5m | ~2mm |
| On-device AI | 157 TOPS | ❌ 无 | ❌ 无 | ⚠️ 需外置 GPU |
| Native VLA | ✅ 专为具身设计 | ❌ | ❌ | ❌ |
| Auto-calib | ✅ 免人工校准 | ❌ 需校准 | ⚠️ 部分 | ❌ 需校准 |
| Point density | 5–7× 专利优势 | 基准 | 基准 | 高 |
In Action
