Reproduce “BrickBench: Evaluating Agentic Brick Design” 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
Peter Kulits, Yiqing Xu, R. Kenny Jones, Cordelia Schmid, Jiajun Wu
BrickBench evaluates whether coding agents can turn written prompts into LEGO assemblies that meet construction checks and depict the requested scene. In BrickAgent, agents select parts, place them through compatible connectors, render builds, and revise them using validation feedback. GPT-6 Astra scores 1.00 on overall validity and 0.954 on VQA, but stability is simulated and does not establish real-world strength.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.