Vitalis Vosylius

Papers

Total Citations

H-Index

About

Vitalis Vosylius is a pioneering researcher at the intersection of computer vision, robotics, and generative AI. His work focuses on bridging the gap between web-scale machine learning models and real-world robotic manipulation, enabling robots to operate intelligently in unstructured environments. Vosylius’s most notable contribution is **DALL-E-Bot** (2022, 8 citations), the first system to integrate a web-scale diffusion model (DALL-E 2) directly into a robotic pipeline. This work allows a robot to perceive objects, infer a natural arrangement, and physically rearrange a cluttered scene—demonstrating how large language and vision models can guide physical action. In parallel, his research on **"Where To Start?"** (2022, 2 citations) addresses a critical bottleneck in robot learning: transferring simple grasping skills from clean, open spaces to complex, cluttered environments. By introducing affordance-aware initialization, Vosylius provides a principled method for avoiding collisions and improving skill transfer. Though early in his career, his work is already shaping a new paradigm where generative models serve as both planners and teachers for embodied agents.