An open-source 6-axis Force/Torque (F/T) sensor based on Hall effect sensors and magnetic field displacement detection. Features 16 Hall sensors in 4 clusters, real-time visualization, and two calibration approaches (geometric and matrix-based).

openFT-sensor/
├── hardware/
│   ├── gerber/              # PCB fabrication files
│   ├── mechanical/          # 3D models (STEP files)
│   └── *.xlsx               # Bill of Materials
├── software/
│   ├── ft_driver.py         # Geometric force/torque driver
│   ├── ft_min_driver.py     # Minimal raw sensor driver
│   ├── ft_vis.py            # Visualization dashboard
│   └── calibration/
│       ├── calc_calibration_matrix.py  # Calibration tool
│       ├── ft_calibration.json         # Example calibration
│       └── CALIBRATION_GUIDE.md        # Detailed calibration guide
└── requirements.txt

1. PCB Fabrication: Use Gerber files in hardware/gerber/ or the pre-packaged zip file
2. Assembly: Upload BOM to JLCPCB for automated assembly, or solder manually
3. Mechanical: 3D print magnet holder from STEP file in hardware/mechanical/
4. Connection: Connect via USB/serial to your computer

git clone <your-repo-url>
cd openFT-sensor
pip install -r requirements.txt

Requires: Python 3.7+, pyserial, zmq, numpy, pandas, dash, plotly, tabulate

• How: Uses sensor geometry and coordinate transformations
• Pros: No calibration needed, quick start, good for testing
• Cons: Less accurate, assumes ideal geometry
• Use for: Testing, relative measurements

• How: Learned 6x16 matrix via least-squares regression from calibration data
• Pros: Much higher accuracy, compensates for real-world effects, accounts for cross-coupling
• Cons: Requires calibration equipment (load cell or reference sensor)
• Use for: Production systems, absolute measurements

Start driver:

cd software
python ft_driver.py

• Default port: /dev/ttyACM0 at 115200 baud (edit line 163 to change)
• Keep sensor unloaded for first 3 seconds to zero-bias
• Common ports: /dev/ttyUSB0, /dev/ttyACM0 (Linux), COM3 (Windows)

Optional visualization:

python ft_vis.py  # Open browser to http://localhost:8050

For detailed calibration instructions, see CALIBRATION_GUIDE.md

Step 1: Collect calibration data

• Need reference F/T sensor or calibrated weights
• CSV format: sensor_1,...,sensor_16,force_local_x,force_local_y,force_local_z,torque_local_x,torque_local_y,torque_local_z
• Collect 100-200 samples across expected force/torque range

Step 2: Calculate calibration matrix

cd software/calibration
python calc_calibration_matrix.py --csv your_data.csv --output calibration.json

See ft_calibration.json for example output.

Step 3: Run minimal driver

cd software
python ft_min_driver.py --port /dev/ttyACM0 --zmq-port 5555 --verbose

Step 4: Apply calibration live

cd software/calibration
python calc_calibration_matrix.py --live --calibration calibration.json

Subscribes to raw data (port 5555), applies calibration, publishes calibrated F/T (port 5556).

Data flow:

ft_min_driver.py → calc_calibration_matrix.py → your_application
   (port 5555)            (port 5556)

Sensor Principle: Hall sensors measure magnetic field changes as the magnet assembly displaces under load. 16 sensors in 4 clusters (4 sensors each) detect X/Y/Z displacements, which are combined to compute 6-axis forces and torques.

Geometric approach: Uses sensor positions (Middle at 0°, Top at 0°/25mm, Bottom-Right at 240°/25mm, Bottom-Left at 120°/25mm) to calculate F/T from first principles.

Calibration approach: Maps raw sensor values directly to F/T using learned matrix that accounts for non-linearities, cross-coupling, and manufacturing variations.

Signal processing: Raw ADC → Moving average (window=3) → Zero-bias/calibration → Displacement calc → Coordinate transform → F/T output

Edit parameters in driver files:

• ft_driver.py line 11: moving_average_window, calibration_window
• ft_driver.py lines 17-19: Sensor geometry (cluster_angles, magnet_dist)
• ft_vis.py line 91: Dashboard refresh rate (interval=100 ms)
• ft_min_driver.py: Use --help for CLI options

Serial connection failed: Check cable, verify port name, add user to dialout group on Linux (sudo usermod -a -G dialout $USER)

Invalid sensor values: Should be 9000-21000. Check power supply, verify sensors soldered correctly, check magnet positioning

No visualization data: Ensure driver running first, check ZMQ ports match, verify no firewall blocking localhost

Zero-bias issues: Keep sensor unloaded during startup, wait for "Calibrated Averages" message

The calibration approach fuses data from all 16 sensors optimally:

• Learns optimal sensor fusion: Each F/T axis uses weighted combination of all 16 sensors
• Compensates non-linearities: Captures real sensor behavior vs. ideal assumptions
• Handles cross-coupling: Accounts for how forces in one direction affect sensors measuring other directions
• Corrects manufacturing variations: Each sensor is slightly different; calibration learns these differences

This is why calibration gives much higher accuracy than geometric calculations.

Contributions welcome! Report bugs, suggest improvements, submit PRs, or share your builds.

Open-source. See license file or contact maintainer for details.

Based on magnetic Hall effect sensing principles. Inspired by the open-source robotics community.