2606.16190v1 Jun 15, 2026 cs.AR

Embedded Arena: Iterative Optimization via Hardware Feedback

Yujia Liu
Yujia Liu
Citations: 0
h-index: 0
Jiuyang Lyu
Jiuyang Lyu
Citations: 4
h-index: 1
Natasha Jaques
Natasha Jaques
Citations: 5,681
h-index: 33
Zhihan Zhang
Zhihan Zhang
Citations: 69
h-index: 4
A. Metzger
A. Metzger
Citations: 23
h-index: 3
Chun-Cheng Chang
Chun-Cheng Chang
Citations: 5
h-index: 1
Jiayi Shao
Jiayi Shao
Citations: 140
h-index: 4
Emmanuel Azuh Mensah
Emmanuel Azuh Mensah
Citations: 3
h-index: 1
E. Wang
E. Wang
Citations: 5
h-index: 2
Kurtis Heimerl
Kurtis Heimerl
Citations: 1,187
h-index: 18
Gregory D. Abowd
Gregory D. Abowd
Citations: 31
h-index: 3
Shwetak Patel
Shwetak Patel
Citations: 72
h-index: 4
Vikram Iyer
Vikram Iyer
Citations: 163
h-index: 7

Embedded devices from wildlife monitoring stations to clinical wearables require local AI inference due to latency, communication, or privacy constraints. Optimizing models for heterogeneous microcontrollers (MCUs) requires simultaneously satisfying hard physical constraints on memory, power, and temperature while preserving accuracy, a multidimensional optimization that is today performed manually by experts. We ask whether an LLM agent can autonomously navigate this complex, multi-turn pipeline guided by real hardware feedback, and introduce a hardware-in-the-loop agent arena in which the agent iteratively refines both model and firmware -- compiling, flashing, and measuring on real hardware -- to enable closed-loop optimization. Frontier models, including Claude Opus 4.7 and Gemini 3.1 Pro, fail entirely without hardware feedback (0% deployment success), whereas our hardware-in-the-loop formulation achieves the first successful deployment within three iterations and can surpass human expert results within seven. This agentic co-optimization achieves 250x compression for vision models with <3.3% accuracy loss and 400x for audio with <6% Feature Error Rate loss, enabling battery-free operation on a commercial MCU via solar harvesting. We demonstrate practical impact in two real-world systems: an elk-detection camera trap (96.7% accuracy) and a phonetic-transcription wearable (8.44% FER) for child development research.

0 Citations
0 Influential
16.5 Altmetric
82.5 Score
Original PDF

No Analysis Report Yet

This paper hasn't been analyzed by Gemini yet.

Log in to request an AI analysis.

댓글

댓글을 작성하려면 로그인하세요.

아직 댓글이 없습니다. 첫 번째 댓글을 남겨보세요!