It’s been a busy few months since we posted that FlexGrid live prediction demo back in July. Here’s where everything stands.
Writing Our First Research Paper
The biggest update is that we’ve been writing a formal research paper for HardwareX — an open-access journal dedicated to open-source scientific hardware. It’s a great fit for this project.
The paper covers the full Open Muscle platform: the OM-12 sensor band with 12 Hall effect pressure sensors, the LASK5 finger force labeling device, and the Random Forest machine learning pipeline that predicts finger movements from forearm muscle contractions. The dataset includes over 87,000 data points from six capture sessions collected over two days.
For those who haven’t followed the technical side — the ESP32-S2 reads sensor values and transmits them as UDP packets over Wi-Fi at about 100 samples per second. A Python receiver on a host PC captures and parses the data, and we train models that can then predict finger movement in real time.
As part of the paper process, we did a thorough data validation — re-running the analysis from raw data to make sure every result is reproducible. The temporal split analysis verified at R² ≈ 0.46, matching our reported findings. We also found that pooled cross-validation across all sessions gives different results than per-session models, and we’re being upfront about that in the paper. Reproducibility is a core value here — we’re not going to hide the parts that are messy.
The paper is currently in final edits. We’ll share the link once it’s submitted.
Video: If you want to see the system working, check out the FlexGrid demo from July 2025 or the OM-12 build tutorial from October 2024.
Hardware Development
While the paper documents the OM-12, development continues on the next generation:
FlexGrid — 60 sensors in a 15×4 array running on the ESP32-S3. Significantly higher spatial resolution than the 12-sensor design, which should translate to better prediction accuracy.
Sensor redesign — We’ve moved away from keyboard switches and are targeting 3mm sensor thickness for all-day comfort. A big evolution from the early Cello module and clevis pin days.
LASK5-V3 — Nearly production-ready with better build quality and more consistent data capture.
All hardware remains under the CERN Open Hardware License, software under MIT.
Get Involved
The project is growing and there are several ways to contribute:
- GitHub: github.com/Open-Muscle — OpenMuscle-Hub is the central repository
- Discord: Join the server for development discussion and community updates
- YouTube: @turfptax — video updates on builds, data, and project progress
We’re also working toward OSHWA certification and continuing to improve documentation across all repos.
What’s Next
Immediate priorities: finishing the paper submission, advancing FlexGrid development, updating repo documentation, and exploring VR integration for data collection and applications.
New YouTube video coming soon covering all of this in detail. Stay tuned.