Gaitr

What if I could analyze Eliud Kipchoge's running gait from video, model him in Blender to scale, export that motion data, and then compare my own gait to his in real-time using wearable sensors? And what if I could get audio feedback through headphones telling me to adjust my form as I ran? Yeah... I don't know. I don't get it either.

Why did I think this was a reasonable first hardware project? I'd never touched a microcontroller. Never soldered anything. Never written a line of code that talked to a physical sensor. And yet, my brain said: "Yeah, let's build a full-body motion capture system and compare it to an Olympic athlete."

I had no idea if this was even possible — at least not for someone like me. But I wanted to learn. Microcontrollers. Accelerometers. Wiring. Soldering. This project touched all of it. Turns out, it's a lot harder than it sounds. Which is funny because it sounds hard.

The plan was ambitious: strap 7 ESP32 devices with accelerometers to different body parts (feet, shins, thighs, hips, etc.), stream all that motion data to a phone app via Bluetooth, and compare it frame-by-frame to Kipchoge's reference gait. The app would calculate a "Kipchoge Similarity Index" (KSI) and give you real-time feedback.

Did I finish it? No. Did I learn a ton about hardware, Bluetooth protocols, and motion analysis? Absolutely. Did I have minor success getting sensor data streaming to the app? Yes! The foundation works — it just needs someone with more patience (or obsession) to take it further.

I bought velcro straps and a hot glue gun. That was my sensor mounting solution. Classy, I know. Each ESP32 got wrapped in velcro and strapped to a body part — two on my feet, two on my shins, two on my hips, one on my pelvis. Seven little computers all trying to talk to my phone at once.

Getting data on the screen was incredibly satisfying. Watching numbers stream in from sensors strapped to my body — that felt like magic. Real motion data, in real time, from hardware I'd wired together myself. For a moment, I felt like a genius.

Then I tried to actually use the data, and reality hit. Calibration was a nightmare. Every time I put the sensors on, they'd be in slightly different positions. The accelerometers had no idea where they were in 3D space relative to each other. I needed some way to establish a reference frame — to tell the system "okay, this is what standing still looks like" — and then maintain that calibration as I moved.

The other problem? Gait phase detection. Even if I could perfectly capture my movement, how do I know which frame of Kipchoge's gait I should be comparing to? Am I at foot strike? Mid-stance? Toe-off? The reference data is a loop, but my running is continuous. Syncing them up in real-time turned out to be a much harder problem than I anticipated.

This is where the project stalled. Not because I gave up, but because I realized I'd need to go way deeper into biomechanics and signal processing than I had time for. The foundation worked. The vision was sound. I just ran out of runway.