Pressure Myography
Meets Machine Learning

Hall-effect sensors measure forearm pressure patterns while machine learning predicts individual finger movements — no EMG, no electrodes, no patents.

How It Works

1

Sense

Hall-effect sensors in a flexible grid detect pressure changes on the forearm as muscles contract. The FlexGrid-64 uses a 15×4 array — 60 sensors capturing the full pressure distribution.

2

Predict

An ExtraTreesRegressor trained on paired sensor/actuator data predicts individual finger positions from raw sensor readings alone. No feature engineering — just 60 raw values in, 4 finger positions out.

3

Move

Predicted finger positions drive a prosthetic hand in real time. The LASK5 linear actuator system provides ground truth during training, then the model runs independently on new sensor data.

Proof of Concept: The results below demonstrate that finger movement data can be extracted from pressure myography sensors using machine learning. These models were trained on very small datasets (under 2 minutes of recording) in controlled conditions. Prediction accuracy is limited — we believe that hardware design improvements and substantially larger datasets (10+ minutes of continuous capture) will yield significantly better results. This is early-stage research, not a finished product.

Real Sensor Data

683 holdout samples from the FlexGrid-64 — data the model never saw during training. Press Play to watch the pressure patterns change in real time.

FlexGrid-64 Sensor Heatmap

0 / 0
Low
High

Finger Prediction Accuracy

Blue is ground truth from the LASK5 actuators. Orange is what the ML model predicted from sensor data alone. Select a finger to compare.

FlexGrid-64 60 sensors — ExtraTrees — R² = 0.660

FlexGrid-64 — Actual vs Predicted

Actual (LASK5) Predicted (ML)
OM-12 SensorBand 12 sensors — RandomForest — R² = 0.264

OM-12 SensorBand — Actual vs Predicted

Actual (LASK5) Predicted (ML)
Why the difference? The FlexGrid-64 has 5× more sensors (60 vs 12) and captures full hand pressure distribution. The OM-12 uses 3 separate SensorBand banks with only 4 sensors each — less spatial resolution means less signal for the ML model to work with.

Model Selection

Six algorithms evaluated on the same train/test split. ExtraTrees won with the best balance across all four fingers, especially the harder-to-predict pinky (label_3).

Model Comparison (FlexGrid-64)

R² score on 683 holdout samples. Higher is better. ExtraTrees selected as best overall performer.

Devices

FlexGrid-64

4×16 hall-effect pressure sensor grid. The current-generation device with full hand pressure distribution capture.

Sensors60 (15×4)
Best R²0.660
ModelExtraTrees (300)
Training Data1,669 samples

OM-12 SensorBand

3 SensorBand banks with 4 sensors each. Legacy wearable form factor — the original proof-of-concept device.

Sensors12 (3×4)
Best R²0.264
ModelRandomForest (100)
Training Data~6,400 samples

Methodology

1

Data Capture

Sensor packets and LASK5 labels received over UDP, paired by timestamp

2

CSV Export

Paired data saved with sensor columns and label columns (label_0 through label_3)

3

Train/Test Split

Holdout data kept completely separate — never seen during training

4

Model Training

Tree-based regressors trained on sensor-to-label mapping (raw values, no feature engineering)

5

Inference & Export

Trained model predicts labels from holdout sensor data; results exported as JSON for visualization

Key finding: Sensor baselines shift between sessions (mean values drift by ~300), making cross-session prediction unreliable. All training and test data must come from the same session until calibration/normalization is solved.