Reproduce “EvoSpatial: Experience-Driven Skill and Tool Evolution for Spatial Reasoning Agents” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Bin Yu, Shijie Lian, Zhaolong Shen, Xiaopeng Lin, Yao Zhang, Haishan Liu, Haotian Li, Yuliang Wei and 5 more

EvoSpatial gives agents reusable spatial procedures linked to tools, specifying when they apply and what evidence to check. Agents inspect and repair analyses, then use checked experience to update a library without correctness scores. Across 20 benchmarks, its equal-weight average is 63.9, versus SpatialClaw’s reported 59.9. With a frozen library, joint evolution scores 65.4 versus 62.0 initially on 10 held-out non-video benchmarks.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.