Briefing
Each benchmark scores what the board reports: current and energy for power minimization on the MAX78000, temperature for thermal management on the ESP32-S3, and model compression on the MAX78000 and STM32N6. The repo ships seed firmware, host-side drivers, deterministic checks and setup scripts for each toolchain, plus a no-hardware smoke test that runs the agent loop in Docker.
The paper reports about 250 times compression for vision models with under 3.3% accuracy loss. Its case studies include an elk-detection camera trap and a speech wearable for child development research. The hardware runs need the same boards and a measurement setup, and the README shows no raw run logs.
Written by the Boards & Bots curator
From the GitHub API
Repository facts
| Repository | ubicomplab/ |
|---|---|
| Stars | 13 |
| Language | C |
| Last push | 8 Jul 2026 3 mo ago |