3D semantic scene-graph paper demonstrates open-vocabulary object retrieval on Spot

Evidence notes

• The paper builds hierarchical 3D scene-graph forests using VLM and LLM reasoning over semantic relationships.
• The authors demonstrate downstream robotics use in an open-vocabulary object-retrieval task using a real-world indoor deployment on Boston Dynamics Spot.
• The event is external product research and is bound to Spot rather than a Boston Dynamics company research release.