A new training-device project runs production-shaped Zephyr firmware on a Renode digital twin of its target microcontroller. The model feeds trigger samples at 1,000 times per second, replays inertial measurement unit (IMU) recordings, and connects two virtual devices over a Controller Area Network (CAN). One automated scenario fires 10,000 shots without reserving two boards or an instrumented range.
The repository connects 76 native unit tests to 29 full-system Robot Framework tests. Those system tests cover stoppages, timing drift, sensor profiles, flash settings, watchdogs, and the messages exchanged between devices. The CAN message definitions also generate test vectors checked against the firmware encoder, so the desktop console and embedded code share a machine-readable contract.
The GitHub Actions workflow had not run when this release landed, so the published counts still need an independent green run. Even with that caveat, the structure is worth copying: replay real sensor traces in emulation, stress the multi-device protocol there, and save physical hardware for electrical behavior the model cannot represent. The first team to adopt it should keep one board test that proves the emulator has not become a more convenient fiction.