DovSG: Dynamic Open-Vocabulary 3D Scene Graphs for Long-term Language-Guided Mobile Manipulation

摘要

Constructs dynamic 3D scene graphs with LLM-based task decomposition for long-term language-guided manipulation.