Context
Smart Mat started as a concept for a smart Pilates mat and evolved, through real user research, into a broader fitness-tracking product. This is my current capstone. It's unfinished, and I'm presenting it here exactly as it stands.
The Need
We ran a 62-response survey and found that 66% of respondents already use an exercise mat, but reported only 6.8/10 average confidence in their own form when exercising without an instructor. When asked to rank tracking priorities, "workout progression" came out on top, above metrics most wearables already track, like heart rate or calories.
Competitive Landscape
We benchmarked seven existing products (Tonal, YogiFi, Tempo, Whoop, Frame Fitness, lululemon Mirror, Forme) across pricing and user complaints. The pattern we found: every competitor nails exactly one layer, pressure sensing, computer vision, or wearable biometric data, but none combine all three. That gap became our product thesis: pressure sensing and computer vision each cover what the other misses, pressure data calibrates stability and balance, while an iPhone camera tracks form, together generating a real-time Form Score, Stability Score, and Overall Performance Score. Our V1 concept narrows this to a single exercise (a bodyweight squat) as our MVP.
A Real Pivot: From Home User to Studio Owner
Two additional interviews changed our primary persona. We'd initially designed around Olivia, a home exerciser. But a conversation with a Pilates studio owner surfaced real tension we hadn't anticipated: concerns about cameras in a studio setting, and fatigue around needing constant real-time feedback during a class. That conversation led us to pivot our primary persona to Lauren, a boutique studio owner, reframing the product from a B2C home-fitness tool into a B2B offering for studios, with instructors and members as the actual day-to-day users.
Current Proof-of-Concept Work
Right now we're building and testing the two halves of the system separately, before attempting to fuse them into one calibrated product. On the pressure side, we're building an 8×8 sensor matrix using velostat, conductive material, and an Arduino microcontroller. Our first milestone, what we call our initial "heartbeat," is simply detecting a pressure change, transmitting that reading, and displaying it on a receiving device. On the computer vision side, we're using Google's MediaPipe Pose to identify body landmarks and track squat movement through an iPhone camera, testing whether a smartphone camera can estimate movement depth accurately enough to be useful on its own.
What we've found so far on the camera side is genuinely unresolved, and that's the honest update: testing with a second participant roughly doubled our measurement error and dropped accuracy from 88% to 70%.
That told us something more useful than a clean success would have. We're now redesigning our camera validation around testing more participants, rather than more reps per participant, to actually understand that variation. Once the camera and pressure sensing each hold up on their own, formal validation studies, measuring accuracy once the two are fused, come next.
What's Next
Proving out the camera and pressure sensing separately, then fusing them into one calibrated system. Our final proof-of-concept milestone for this phase is achieving a complete system "heartbeat": Smart Mat successfully collecting pressure-distribution data, tracking a squat through computer vision, processing both signals together, and displaying integrated feedback in one application, tested across repeated trials for consistency. We're targeting this by the end of our current capstone phase, with the next phase focused on refining the prototype and improving reliability through formal validation studies before any commercial direction is decided.
Reflection
The most useful mindset shift on this project came directly from how we approached testing: a result that disproves your assumption is real data, not a failure. Measurement error doubling wasn't a setback, it told us exactly what to investigate next. That's a better habit than I had going into this project, and I think it's the single most transferable thing from Smart Mat to how I'd approach any ambiguous product problem.