2.5-D decomposition pipeline boosts LLM spatial reasoning in construction tasks

Researchers have developed a novel neuro-symbolic pipeline called "2.5-D decomposition" to improve the spatial reasoning capabilities of large language models (LLMs) in construction tasks. This method separates the LLM's planning into a 2D horizontal plane while a deterministic executor handles vertical placement, significantly reducing coordinate errors. The system achieved 94.6% structural accuracy on the Build What I Mean benchmark using GPT-4o-mini, outperforming GPT-4o and other leading systems. Notably, the pipeline also demonstrated strong performance when run locally on edge hardware with the Nemotron-3 120B model. AI

IMPACT This method could significantly improve the reliability of LLMs in real-world construction and assembly tasks by reducing spatial errors.

RANK_REASON The cluster contains a research paper detailing a new method for LLM spatial reasoning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• 2.5-D decomposition
• Build What I Mean benchmark
• GPT-4o
• GPT-4o-mini
• Nemotron-3 120B
• NVIDIA Jetson Thor AGX

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →