Reproduce “Functional-SLAM: Interaction-Aware Mapping with Online Functional Scene Graphs” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xinggang Hu, Chenyangguang Zhang, Zihan Zhu, Ruida Zhang, Xiangkui Zhang, Xiangyang Ji
Functional-SLAM introduces the first framework for simultaneously localizing a camera and constructing an online functional scene graph, providing real-time, interaction-aware scene understanding for robots. The system achieves a 16.3% reduction in localization error on the FunGraph3D dataset and builds functional graphs significantly faster than prior offline methods.
Key insights
- Integrating functional scene graph construction directly into the online SLAM pipeline is feasible, enabling real-time interaction capabilities previously limited by offline processing.
- Stabilizing the representation of small, interaction-critical elements and their relationships in a dynamic SLAM environment requires a multi-faceted approach combining geometric, semantic, and temporal contextual evidence.
- Functional scene graphs can reciprocally improve SLAM performance by providing robust topological cues for loop closure detection, particularly beneficial in visually challenging or repetitive environments.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.