arXiv:2606.09134v1 Announce Type: cross Abstract: Constructing knowledge graphs from 3D simulation scenes is essential for robot task reasoning, but the key bottleneck, grounding scene objects to formal ontology classes, still relies on manually curated dictionaries that are brittle and do not generalize across assets. We investigate whether large language models (LLMs) can automate this grounding step for Universal Scene Description (USD) scenes as a zero-shot, training-free alternative. On a kitchen scene (125 objects) with SOMA-HOME Ontology, LLMs achieve 90-96% exact-match accuracy with de
The increasing sophistication of LLMs and the growing need for efficient 3D scene understanding in robotics make this a timely application of AI to a key bottleneck.
This development significantly lowers the barrier for constructing knowledge graphs from 3D simulation scenes, accelerating robot task reasoning and potentially commercial robot deployment.
The reliance on manually curated dictionaries for grounding scene objects to formal ontologies can be largely replaced by zero-shot LLM techniques, making the process faster and more generalizable.
- · Robotics companies
- · AI software developers
- · Simulation platforms
- · Defense contractors
- · Manual data annotators
- · Companies reliant on proprietary 3D scene knowledgebases
Automated knowledge graph generation from 3D scenes becomes more scalable and accurate.
This leads to more sophisticated and adaptable robotic systems capable of understanding and interacting with complex environments.
The acceleration of practical robotics applications could spur demand for related AI agentic systems in physical domains.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.AI