OpenSplatGraph: From Dense Semantic Maps to Structured Scene Graphs for Open-Vocabulary Robot Perception

Accepted at ACCV 2026

Binh Long Nguyen^1,2 , Kien Nguyen¹ , Clinton Fookes¹ , Peyman Moghadam^1,2

¹Queensland University of Technology   ²CSIRO Robotics

This repository hosts materials for OpenSplatGraph, a unified framework that constructs persistent 3D scene graphs from an online Gaussian-based open-vocabulary semantic map for robot perception, accepted at ACCV 2026.

News

• September 2026: OpenSplatGraph is accepted to ACCV 2026.
• October 2026: arXiv preprint is available.
• Code will be updated when available.

Installation

Source code and installation instructions will be released soon.

Acknowledgement

This work was supported in part by the Australian Research Council Discovery Project under Grant DP250103634, and in part by the Commonwealth Scientific and Industrial Research Organisation (CSIRO). The authors acknowledge continued support from the CSIRO's Embodied AI Cluster.

Citation

If you find this repository useful, please cite:

@inproceedings{nguyen2026opensplatgraph,
  author    = {Nguyen, Binh Long and Nguyen, Kien and Fookes, Clinton and Moghadam, Peyman},
  title     = {OpenSplatGraph: From Dense Semantic Maps to Structured Scene Graphs for Open-Vocabulary Robot Perception},
  booktitle = {Asian Conference on Computer Vision (ACCV)},
  year      = {2026}
}