University of Bradford Motion Lab Computer station

Step 1

Connect a phone

Preparing a session
Creating
QR code...

Scan with your phone's camera.

No app or account is needed.

The phone will open a sensor page and ask permission before sharing readings.

No phone? Use the signal simulator

The simulator lets you test the entire activity on this computer.

Live signal

Sensor readings

0 samples/s
Accel X0.00m/s²
Accel Y0.00m/s²
Accel Z0.00m/s²
Gyro X0.00°/s
Gyro Y0.00°/s
Gyro Z0.00°/s
Acceleration X Y Z
Gyroscope rotation rate X Y Z

Step 2

Record movement examples

Each recording is an example that teaches the model what a movement looks like.

Choose a movement

Follow the picture and description for the full three seconds.

Get ready...
•••

Your recordings will appear here.

Recorded series

Acceleration X Y Z
Gyroscope rotation rate X Y Z

Step 3

Train the model

Loading Python

Once you have recorded three examples of at least two movements, train the model and check how well it recognises them.

Your movement names tell the model the correct answer for each training recording. Python gives every recording 90 equally spaced time points, then replaces each signal with summary values such as its average, variation, range and rate of change. The three closest examples vote, with closer examples having more influence. This is the faster option.

Collect three recordings for each of at least two movements.

Accuracy on hidden recordings One complete three-second recording is hidden while the model learns from the others. It compares the hidden recording with the three closest remaining recordings, giving closer matches more influence. This repeats for every recording, and the percentage shows how many were identified correctly. This is called leave-one-recording-out cross-validation.
—

Step 4

Test the model

Repeat a trained movement and watch the live prediction, or record a separate three-second test.

Train the model in Step 3 to unlock prediction.
STILL ? ? •••
Live prediction Waiting for movement...
Collecting a three-second window
“Unknown” is a confidence rule, not a fifth movement learned by the model. When enabled, the app shows “Unknown” if no movement receives at least 60% of the distance-weighted neighbour vote. Turning it off always shows the trained movement with the highest vote.
Recorded prediction

Repeat one of your movements for three seconds. This test will not be added to the training examples.

Get ready...
STILL ? ? •••
Predicted movement —
Acceleration X Y Z
Gyroscope rotation rate X Y Z

Your data is not saved.

Sensor readings, recordings and the trained model are held only in this computer tab's temporary memory. They are deleted when this tab is closed or refreshed.

Connect your phone

Enable the sensors, then keep this page open while using the computer.

Connecting to computer

Your browser may ask for permission to access its motion sensors.

Live readings 0 samples/s
Accel X0.00m/s²
Accel Y0.00m/s²
Accel Z0.00m/s²
Gyro X0.00°/s
Gyro Y0.00°/s
Gyro Z0.00°/s
Hold on securely.

Use small, controlled movements and stay clear of other people and equipment.

Your data is not saved on this phone.

Live readings are held only while this tab is open. Readings already sent remain temporarily in the computer tab until that tab is closed or refreshed.