2026.04.15- #67 -

[changh95]

ZED X Nano

• StereoLabs의 소형 카메라
• Manipulation을 위한 wrist-mount 카메라
• 1920 * 1200 글로벌 셔터 센서 + RGB + Depth 이미지 (120 FPS)
• GMSL2 카메라 인터페이스 -> 15m + locking 커넥터 사용 가능
• Vibration-proof IMU
• 센서부터 GPU 까지 Zero-copy 가능
• Neural Depth Engine을 이용해서 sub-millimeter 정확도 취득 가능
• https://www.businesswire.com/news/home/20260413750144/en/Ouster-Launches-Stereolabs-ZED-X-Nano-A-Wrist-Mount-Stereo-Camera-Built-for-Robotic-Manipulation-and-Physical-AI

Boxer: Robust Lifting of Open-World 2D Bounding Boxes to 3D

• Open-vocabulary 2D detector (OWLv2) -> BoxerNet lift 2D detection to 3D oriented bounding box -> Temporal fusion
• Camera intrinsic, Gravity direction 필요
• Optional input: Semi-dense depth.
• Temporal fusion -> Hungarian algorithm
• https://facebookresearch.github.io/boxer/

SLAM and VIO in Egocentric Data: Where Long-Horizon Tracking Breaks

• Egocentric VLA 모델 개발 시 long-horizon 데이터를 만들기 어려운 이유에 대해 SLAM/VIO 관점에서 설명
• 1. Long horizon task는 정확도가 굉장히 중요하다
• 2. Degenerate case 풀기 상당히 어렵다
• 3. Head motion이 굉장히 빨라서 어렵다
• 4. IMU 매우 중요하다
• https://www.fpvlabs.ai/essays/slam-and-vio-in-egocentric-data-where-long-horizon-tracking-breaks

Rust robotics

• Python robotics를 Rust로 포팅!
• 'Rust 한입 잡사봐'
• https://github.com/rsasaki0109/rust_robotics

EUPE - Efficient Universal Perception Encoder

• Meta의 새로운 Foundation vision encoder
• https://github.com/facebookresearch/EUPE

Google Gemini ER1.6

• https://deepmind.google/blog/gemini-robotics-er-1-6/
• Agentic Physical AI를 위한 한걸음
• Gemini 3.0 Flash, Gemini ER1.6 보다 발전
• This model specializes in reasoning capabilities critical for robotics, including visual and spatial understanding, task planning and success detection. It acts as the high-level reasoning model for a robot, capable of executing tasks by natively calling tools like Google Search to find information, vision-language-action models (VLAs) or any other third-party user-defined functions.

[U-AMC]

생체 모방형 4절 링크: 테슬라의 무릎 특허(US20260097493A1) 공개!!

SLNR

LINK
A super lightweight neural representation for large-scale 3D mapping