Source record

Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2610.10044.

Pipeline notes

The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.

Introduces ALPS-Bench, a 1,000-layout benchmark for real-world facility design with instance-specific scoring, and CANDO, a training-free multi-agent framework where specialized agents iteratively refine layouts via verification-grounded loops, outperforming trained and LLM baselines on PubLayNet, RICO, and PKU-PosterLayout.

Concrete example of a training-free cooperative-agent recipe with a benchmark that replaces distributional metrics (FID/IoU) with structured manual-grounded scoring, relevant to anyone studying multi-agent design workflows.

Cited by

No citations on record.