Using DUSt3R-2 to reconstruct a 3D space from three 2D images (Image provided by Naver Labs)
SEOUL, Nov. 24 (Korea Bizwire) — Naver Labs Europe has introduced two advanced artificial intelligence models designed to enhance robots’ ability to understand physical spaces and human bodies, the company announced Monday. The models were unveiled at the fourth “AI for Robotics” workshop, held Nov. 20–21 in France under the theme of spatial intelligence.
The first model, DUSt3R-2, is the next-generation version of Naver’s DUSt3R system, a 3D vision model first revealed in late 2023. DUSt3R reconstructs three-dimensional space from a single photograph, enabling robots to interpret unfamiliar environments without relying on prebuilt maps. The updated DUSt3R-2 integrates numerous derivative studies and is optimized with commercial applications in mind.
Naver Labs Europe also introduced ANNY, a 3D body model capable of representing diverse human body types with high precision. Built using open-source anthropometric datasets and World Health Organization population statistics, ANNY can model a wide range of physiques—from children to older adults—without relying on personal data, addressing key privacy concerns.
The workshop featured discussions on how spatial intelligence can elevate robots’ perception, reasoning, and movement, reflecting growing academic and industry interest in applying advanced 3D reconstruction and human-modeling tools to real-world robotic systems.
A 3D reconstruction of an adult and a child using the ANNY 3D body model
(Provided by Naver Labs)
Naver Labs Europe, based in Grenoble, collaborates with researchers from 26 countries and focuses on spatial intelligence as a core component of future robotics. “We will continue sharing our research globally and deepening our focus on AI for robotics,” said Lee Dong-hwan, head of Naver Labs.
The “AI for Robotics” workshop, launched in 2019 and held biennially, brings together leading researchers and experts to explore the evolving role of artificial intelligence in robotic technologies.
Kevin Lee (kevinlee@koreabizwire.com)
