title: EXOKERN
emoji: ⚡
colorFrom: blue
colorTo: red
sdk: static
pinned: false
short_description: The Data Engine for Physical AI
EXOKERN — The Data Engine for Physical AI
Contact-rich. Sensor-annotated. Industrially validated. EU AI Act ready.
We produce high-fidelity force/torque manipulation datasets for enterprise robotics, humanoid robot manufacturers, and research institutions — with full data provenance and compliance documentation.
🎯 The Problem
Over 95% of existing robotics datasets lack force/torque sensor data. Vision-only approaches fail at contact-rich tasks like insertion, assembly, and manipulation — where precise haptic feedback is critical.
Starting August 2026, the EU AI Act requires documented data provenance for AI training data. Most existing datasets cannot meet this standard.
💡 Our Solution
EXOKERN provides industrially calibrated 6-axis force/torque (wrench) data alongside standard observations, enabling robots to learn manipulation skills that require physical contact understanding.
Capture → Anchor → Ship
| Phase | What We Do | Output |
|---|---|---|
| 01 Capture | Operators generate specific contact forces incl. systematic failures in calibrated environments | Raw data + Failure Taxonomy + QC Report |
| 02 Anchor | Real friction logs calibrate simulation. Physics parameters validated against F/T measurements | Calibrated Sim Assets (MJCF/USD/JSON) |
| 03 Ship | ML-ready exports in all major formats — clean, tagged, quality-certified | RLDS, HDF5, Zarr, MCAP, LeRobot v3.0 |
📦 Datasets
| Dataset | Task | F/T Data | Status |
|---|---|---|---|
| contactbench-forge-peginsert-v0 | Peg-in-Hole Insertion | ✅ 6-axis Wrench | Available |
🗺️ More contact-rich datasets in development — including assembly, grasping, and deformable object manipulation. Commercial Contact Skill Packs available for enterprise customers.
🔧 What Makes Our Data Different
- Force/Torque annotations on every frame (Fx, Fy, Fz, Mx, My, Mz)
- LeRobot v3.0 compatible — plug directly into policy training pipelines
- Industrially relevant tasks — insertion, assembly, contact-rich manipulation
- Calibrated sensors — not estimated, not vision-derived, directly measured
- Full data provenance — capture methodology, operator ID, QC reports, EU AI Act compliant
- Sim-anchoring — calibrated simulation assets bridge the Sim2Real gap
🏗️ Built With
- NVIDIA Isaac Lab for high-fidelity physics simulation
- LeRobot v3.0 format for interoperability
- Franka FR3 (7-DOF) robot platform
- Bota Systems SensONE 6-axis F/T sensor
📬 Contact
- 🌐 Website: exokern.com
- 🤗 HuggingFace: huggingface.co/EXOKERN
- 📧 Enterprise inquiries: info@exokern.com
EXOKERN — Bridging the Haptic Data Gap in Physical AI