Reproduce “Embedded Arena: Iterative Optimization via Hardware Feedback” 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

Zhihan Zhang, Alexander Le Metzger, Jiuyang Lyu, Chun-Cheng Chang, Jiayi Shao, Yujia Liu, Emmanuel Azuh Mensah, Edward Wang and 5 more

Embedded Arena tests whether agents can deploy and tune models on microcontrollers using feedback from physical hardware. Agents edit a model or firmware, compile and flash it, then use valid-run measurements to guide further changes; failed runs return diagnostics. In a YOLO object-detection experiment on MAX78000, Gemini 3.1 Pro reaches 18.03% mAP50-95 on held-out COCO, above the 14.15% expert baseline.

Summary by alphaXiv.

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