Status: Living document, updated as milestones complete or strategy shifts. Last updated: 2026-06-07 Maintained by: Tory (TURFPTAx), founder Scope: What OpenMuscle is building, in what order, and why.
This roadmap is intentionally honest about three things: where we are today, what's coming soon (with rough dates), and what the long-term vision looks like (without overselling). If you want to contribute, the near-term section is where to start. If you want to understand why we exist, read the mission and the ten-year vision.
For the underlying repo and firmware structure that supports this roadmap, see docs/architecture-and-repo-layout.md.
Democratize prosthetic sensor technology. Build affordable, open-source, open-hardware pressure myography sensors that anyone can build, modify, and improve. Every design file, every line of code, every dataset is published openly under MIT (software), CERN-OHL-S v2.0 (hardware), and Creative Commons (documentation) licenses.
The first beneficiaries are people with intact forearms but missing hands. The broader benefit is a shared, open dataset of pressure myography readings that the research community can use across prosthetics, human-computer interaction, and brain-machine interface research.
- FlexGrid V3 is working. The 60-sensor (15 by 4) Velostat band, about $45 in components, runs end-to-end: live sensor capture, model training, inference, and integration with a robot hand. First clean inference session hit R² = 0.854.
- FlexGrid V4 is in production. Ten dev kits ordered, arriving mid-June 2026, shipping to two outside collaborators plus the founder.
- LASK5 (the labeling wand for ground-truth capture) is functional and lives at Open-Muscle/OpenMuscle-LASK5; the migration from the legacy
turfptax/lask4repo is complete. - Software stack includes the PC application (Python, FastAPI web UI, ML training and hot-swap inference, robot hand forwarding) and the firmware in MicroPython on ESP32-S3.
- First academic use: a master's thesis at an Oregon university is testing FlexGrid on an amputee participant.
- Legal structure: not yet incorporated. Nonprofit incorporation is in preparation; details will be published as that work progresses.
The focus is on shipping V4 to outside hands, hardening the contribution flow so collaborators can start adding value, and standing up the public data pipeline.
- FlexGrid V4 dev kit rollout. Ten units ordered; first three (founder plus two collaborators) deployed by late June 2026. The V4 keeps the rigid PCB on the left side of the band, matching the V3 layout that proved out.
- Bridge PCB research. A separate interposer PCB between the flexible sensor PCB and the rigid compute board. It carries the multiplexer so the harness drops from the current 19 wires down to 8, and it takes mechanical stress off the flexible PCB to reduce wear and make repairs easier. Intended for a future hardware revision (not V4 itself), so the option exists when V5 design starts. Idea contributed by @HeatRIderHR (#3).
- LASK5 repo cleanup. The hardware repo migration to
Open-Muscle/OpenMuscle-LASK5is done. Remaining: flatten the duplicated KiCad trees from the migration, prune committed backup archives, and promote firmware (MIT) to its own repo when the hardware stabilizes, per the repo layout doc.
- Multi-input training capture. Add recording paths for keyboard input, game controller input, and musical instrument input (MIDI first). Every input becomes a labeled dataset where the bracelet sees the finger motion and the input device provides ground truth.
- Public training data repository. Stand up a hosted, downloadable corpus of labeled pressure myography sessions, free to anyone for ML research. The first application is prosthetic finger-movement prediction; the dataset enables much more. This is the long-game asset.
- VR/XR client extraction (not yet started). Extract the WebXR client out of the PC web app and into the
OpenMuscle-ARrepo. The client currently still lives inOpenMuscle-Softwareunderpc/src/openmuscle/web/static/vr/. PC app keeps the FastAPI server and the VR bridge endpoints. - CONTRIBUTING.md, Code of Conduct, and issue templates finalized in the Hub repo so external contributors can open issues, route them to the right sub-repo, and get credit for ideas as well as code.
- Nonprofit incorporation (501(c)(3) most likely). The nonprofit will own the OpenMuscle brand, the core open-source hardware designs, and the public datasets. Revenue comes from brand licensing fees, donations, and grants.
- Brand-licensing program. Stand up the process by which any company, including future commercial OpenMuscle ventures, can license the OpenMuscle trademark for products built on the open designs. Other companies are explicitly welcome to build commercial products; the nonprofit cannot, and will not, prevent commercial forks.
The near-term work earns OpenMuscle the right to talk about scale. This section is what scale looks like.
- FlexGrid V5 with the bridge PCB integrated, targeting a smaller, lighter, more wearable form factor. Goal is "all-day-wearable" comfort, not just "good for a session."
- Per-device firmware repos promoted as hardware stabilizes (FlexGridV4-Firmware, LASK5-Firmware, etc.), each pinned to a single hardware major.
- First commercially-licensed third-party hardware. Another company ships a product using OpenMuscle designs under the nonprofit's brand license. We help, they pay licensing fees, the nonprofit reinvests.
- Mature PC application with polished UI, packaged installer, multi-platform support.
- Native VR headset application. WebXR is the bridge; native Quest (or equivalent) APK is the target. Lives in
OpenMuscle-AR. - Mobile / Android application with Bluetooth sync to a paired bracelet. Phone becomes a portable training and capture device.
- Inference at the edge. Models small enough to run on the device or on the headset, not just on a tethered PC.
- Public dataset at meaningful scale. Target: tens of thousands of labeled sessions across diverse users, hand sizes, and tasks (typing, gaming, instruments, daily activities).
- First peer-reviewed publications beyond the Oregon thesis. Co-authored work with collaborating universities.
- Open dataset citations in third-party prosthetics, HCI, and BMI papers.
- Nonprofit board filled with independent directors (prosthetics expertise, academic research, disability advocacy).
- Self-sustaining revenue. Licensing fees and donations cover operating costs without founder cash injection.
- First nonprofit-funded device giveaways to people who need a prosthetic-control sensor but can't afford one.
The further out a roadmap goes, the more it should describe outcomes rather than predictions. These are the outcomes we are building toward. They are conditional on a lot of things going right, and we say so openly.
If OpenMuscle becomes the de facto standard for pressure myography sensing, the following becomes possible:
- Large-scale consumer applications. A hardware standard cheap enough and reliable enough to be used as a human interface device in games, productivity tools, creative software, accessibility tools, and XR experiences. Every use generates additional labeled training data, with explicit consent, that flows back into the public dataset.
- Anti-cheat as a side effect. Real hand motion is extremely hard to fake. A platform that reads what your hand is actually doing is, incidentally, a strong anti-cheat signal. Game publishers integrating this gain both an input modality and an integrity check. Players gain the option to contribute their gameplay motion data to prosthetic research.
- Prosthetic AI training at unprecedented scale. With orders of magnitude more diverse hand data than currently exists, machine learning models for prosthetic control improve dramatically. Models trained on this data improve quality of life for amputees, including veterans.
- Brain-computer interface research enabled. Companies and labs working on neural implants (Neuralink and the broader BMI field) need ground-truth movement data to correlate against neural signals. An open dataset of correlated pressure myography readings is one of the inputs that makes passive BCI viable. We will contribute data, not depend on partnership.
- Free or subsidized prosthetic devices distributed globally through the nonprofit, prioritizing regions where commercial prosthetics are unaffordable. Commercial licensees subsidize the giveaway program through brand-license fees, alongside donations and grants.
Many people, around the world, using free open-source prosthetic hands made possible by OpenMuscle, regardless of their ability to pay.
This is the ten-year north star. Every milestone above is a step toward it. We may not get there in ten years. We are still building toward it.
OpenMuscle is preparing to incorporate as a nonprofit. Full structural details will be published as incorporation work progresses. The core commitment is unchanged: hardware designs, software, and datasets remain open under MIT, CERN-OHL-S-2.0, and Creative Commons licenses. The only thing the nonprofit will gate is the OpenMuscle trademark, via a brand-licensing program. Other companies are explicitly welcome to build commercial products on top of the open designs; the nonprofit will not, and structurally cannot, prevent commercial forks.
Project costs are currently funded by the founder personally. This will change as the nonprofit comes online and a sustainable mix of donations, grants, and brand-licensing fees takes over.
- Quarterly review. Once a quarter, this document gets updated to reflect what shipped, what slipped, and what changed.
- Major decisions (incorporation, repo restructuring, hardware revisions) are recorded as their own docs under
OpenMuscle-Hub/docs/and linked from here. - Open discussion happens in GitHub Discussions on this repo, and (when stood up) on the OpenMuscle Discord.
- Contributors get credit. Ideas surfaced via issues, even ones that don't ship, are tracked back to the contributor in commit co-author trailers, the CONTRIBUTORS file, or release notes.
If you have a use case, a critique, a proposed change to a milestone, or a question about why something is or isn't on this roadmap, open an issue. The Hub is the routing point.
- FlexGrid: the 60-sensor (15 by 4) Velostat-based pressure-sensing forearm band. Currently shipping V3 hardware, V4 in production.
- LASK5: the handheld labeling wand used to capture ground-truth finger positions while a user wears FlexGrid, for supervised ML training.
- OpenMuscle-AR: new repo for the WebXR client and future native VR/XR headset application.
- PMG: Pressure Myography. Reading muscle activity by sensing the pressure changes muscles cause under the skin.
- TDMG: Tissue Deformation Myography. Reading muscle activity by sensing how the muscle shape changes.
- CERN-OHL-S-2.0: Open hardware license used for OpenMuscle hardware. Permissive and reciprocal.
- BHAG: Big Hairy Audacious Goal. From Jim Collins's Built to Last. The ten-year north star.