// the find
amishimmathur-ux/Digit-Recognition--Sentis-AI-with-IMU-Calibration-3D-Game-
A 3D game developed in Unity which integrates Sentis AI library to execute AI functions and actions inside with game paired with an industrial-grade calibration framework built on STM32 platforms with the ISM330DHCX MEMS hardware layer which controls the functions inside the game.
A hackathon project (TechBlazers 2.0) pairing an STM32 board with an ISM330DHCX IMU to drive mouse/keyboard input for a Unity game, with a Unity Sentis model meant to turn the motion data into gesture/digit predictions. It's aimed at people curious about embedded-sensor-to-game pipelines, not at anyone looking for a reusable library.
The gesture controller firmware is the one part that's actually detailed and credible: EMA filter at α=0.4 with a 500 mdps dead-zone for the gyro, 50ms button debounce, 400ms double-tap window, and an explicit, reasoned tradeoff (letting the first tap of a double-tap fire immediately to avoid adding latency to normal single presses). The Python host also uses the Windows SendInput API for relative mouse movement instead of SetCursorPos, which is the correct call since most game engines ignore absolute cursor repositioning — and it bothers to release held keys/buttons on disconnect or Ctrl+C so you don't get stuck inputs.
There's no Unity project in the repo at all — the directory tree is STM32 firmware, HAL drivers, and ST motion middleware .a blobs, plus a loose MNIST Sample.exe and UnityCrashHandler64.exe committed straight into the repo root. The actual Sentis/AI/Unity side that the README leads with isn't here to inspect or build. The README itself repeats the same pipeline diagram four separate times and pads out with generic 'Learning Outcomes' and 'Applications' sections that say nothing project-specific. No license, no build/flash instructions for the STM32CubeIDE project, and no tests — and shipping compiled .exe files in source control is something you'd flag in any PR review.