On the opening day of ICCV 2025 – one of the world’s leading conferences in artificial intelligence and computer vision – Dr. Jan-Nico Zaech, Research Scientist at INSAIT, delivered a keynote lecture at the workshop “Foundation Data for Industrial Tech Transfer.”
In his talk, Dr. Zaech presented a new approach to developing robotic artificial intelligence by leveraging human videos and other non-robotic data sources rather than relying solely on robot demonstrations. This approach enables robots to acquire new skills and operate more reliably in real-world industrial environments.
ICCV is the premier global venue where advances in computer vision intersect with robotics, turning visual perception and 3D understanding into action. The workshop highlights the growing importance of foundation data and technology transfer in bridging research innovations with industrial applications.
INSAIT’s research contributions presented at the workshop include MotoVLA, a system that learns new robotic skills from human videos; ReVLA, improving model robustness to change and new environments; GaussianVLM, connecting visual-language models with 3D “Gaussian Splatting” scenes for embodied open-world understanding; and Articulate3D, a dataset modeling objects, parts, and motion to help robots reason about interactions.
By incorporating diverse, non-robotic data sources, these approaches enable the creation of more adaptable and reliable AI systems, marking an important step toward practical industrial deployment and faster technology transfer from research to real-world applications.