Software

The complete acquisition and analysis software, from the readout firmware to paw-labelled stance events, is available in the repository under the MIT License. This page summarizes its organization; the repository README provides the full reference. Scripts used to compute the gait metrics and statistical analyses reported in the manuscript are available from the corresponding authors upon request.

Repository structure

Folder	Contents
acquisition/	record_runway.py: synchronized recording of the four boards and the overhead and side cameras (.mat and .mp4); live_view_runway.py: live pressure map; firmware/: readout-board firmware (Teensy 4.1)
pipeline/	gait_config.m: paths and study settings; stages s1–s5 (below)
tailbase/	tailbase_cnn.py: tail-base localization (annotation, training and prediction)
demo/	run_demo.m: runs the pipeline on the demonstration recording and verifies the output
data/	pretrained tail-base model, calibration curves and the platform template
hardware/	KiCad sources, fabrication files and bill of materials (CERN-OHL-P-2.0)
docs/, site_tools/	this website and the scripts that generate its figures and videos from the recordings

Requirements

• MATLAB (tested with R2023b) with the Image Processing, Signal Processing, and Statistics and Machine Learning toolboxes.
• Python 3 for the tail-base network: pip install -r requirements.txt (NumPy, OpenCV, PyTorch). A GPU is not required; on the demonstration recording, a CPU-only run required the same time and produced identical labels.
• Acquisition only: the readout boards, a host computer with four USB ports and the two cameras (overhead and side), together with pyserial and scipy.

Demonstration workflow

The demonstration package contains one complete recording (rat 4, 24 February 2026, two weeks before surgery; an 85-s session with five runway traversals) organized in the same structure as the full data deposit (see Data).

setenv('GAIT_PYTHON', 'C:\path\to\python.exe')   % Python with PyTorch, if not on the PATH
addpath('<repo>/demo')
run_demo('C:\gait_demo\data', 'C:\gait_demo\work')

The workflow runs stages s1–s4 and writes the 5 walking windows, the 58 labelled stance events and a labelled video of each walking window. It then compares the tables with the reference outputs in demo/expected/ and reports DEMO PASSED on success. The run requires approximately 9 minutes on a 32-thread workstation.

Pipeline stages

Stage	Operation	Output
s1 Preparation	Median background image of each overhead video; for new data, video selection and annotation of the four platform corners (sensor corners and homography follow from the platform template)	background model, homography
s2 Contact detection	Background subtraction and warping of video into the 256 × 64 board space; tail-base localization; reconstruction of the pressure movie from the board packets; contact segmentation; camera-derived votes	pressure movie (~120 frames/s), contacts
s3 Walking bouts	Whole-animal centre of pressure at ~100 Hz; candidate walking bouts from its speed; review interface to retain or trim candidates	COP trace, walking windows
s4 Paw assignment	Re-segmentation of stance contacts within each window and assignment to FL/FR/HL/HR from position relative to the smoothed COP and heading, with camera votes; Viterbi decoding of left/right sequences	stance events (with labelled video)
s5 Calibration	Conversion of summed responses to mass-equivalent grams with the calibration curve of each board	loads in grams

Conventions: time in seconds on the shared recording clock; positions in taxels on the 64 × 256 grid (one taxel = 2 mm; boards 4-3-2-1 from left to right); pressure in raw 12-bit counts until calibrated.

Website and media

This website is built with MkDocs Material from docs/. The scripts in site_tools/ regenerate its pressure videos, supplementary videos and sampling displays from the raw recordings, using the same frame reconstruction as the analysis pipeline (most recent packet of each board; boards 4-3-2-1 from left to right).