Go back to publications View external

F3DGS: Federated 3D Gaussian Splatting for Decentralized Multi-Agent World Modeling

April 2026 Morui Zhu, Mohammad Dehghani Tezerjani, Mátyás Szántó, Márton Vaitkus, Song Fu, Qing Yang IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2026
#Gaussian Splatting #World Modeling #Federated Learning

F3DGS is a federated framework enabling multiple robotic agents to collaboratively reconstruct 3D scenes without centralizing all data. It preserves geometric consistency by fixing Gaussian positions while allowing distributed updates to appearance attributes, with visibility-aware aggregation resolving partial observability across agents.