Dog Orthopedic Sensor Abstract

Functional Prototype Completed - 8/1/2025
*Development Continues
 

Team Members

• Tiffany Lin – Project Lead, Prototype Board Design/Firmware/Schematics, ML collaborator

• Cameron Potvin – Project and ML Supervisor, Board Design/Case

• David Tanioka – Machine Learning collaborator

​

Goal / Introduction

The Dog Orthopedic Sensor project presents a portable, low-cost gait analysis device for canine patients, developed as a proof of concept to transform veterinary diagnostics. The system is designed to enable faster, more cost-efficient assessment of orthopedic disorders while providing veterinarians with machine learning–driven decision support. By replacing expensive laboratory-grade motion analysis systems with a field-ready device, the project aims to reduce diagnostic costs, accelerate decision-making, and support earlier intervention for improved treatment outcomes.

​

Methodology

The project began in January 2025, building on prior work by Cole Schreiner, whose prototype used an Arduino Nano and a single MPU9250 sensor in a bulky breadboard setup. Under the mentorship of Don Wahlquist, the design was refined to reduce weight and increase accuracy. Two MPU6050 sensors replaced the original MPU9250, positioned at the spine above the shoulders (MPUA) and hips (MPUB) to capture more comprehensive gait data.

​

The Arduino Nano was replaced with an ESP32-Wroom32 for its webserver hosting capabilities, later downsized to an ESP32C3 to reduce device mass while maintaining performance. Data is stored on an SD card in CSV format, transferred to a SQL database, and fed into a Temporal Convolutional Network (TCN) model for classification of healthy versus unhealthy gait patterns. The device is powered by a 7.4v Li-Po battery, with components mounted on TPU platforms and secured with a stretchable harness.

​

Firmware was optimized to collect 10 sensor readings per second, apply angle adjustments for accurate orientation, and manage data logging. The ESP32 webserver enables live streaming of gait data, sensor readings, and a simple skeletal rendering of hip and shoulder movement.

​

Results

• Prototype Functionality: Final design reduced weight and bulk, improving comfort for dogs of various sizes.

• Data Capture: Accurate accelerometer and gyroscope readings collected from two spinal locations.

• Machine Learning: TCN model accuracy improves with each new sample recorded in the SQL database.

• Visualization: Live web-based graphs and animations of gait movement are available during testing.

​

Supporting materials include:

• Scrolling graphs for multiple dogs

• Images of the device hardware

• A dog skeleton render animation

• Video demonstration of the device in use

​

Conclusion

The Dog Orthopedic Sensor demonstrates the feasibility of a lightweight, portable system for canine gait analysis. By combining accessible hardware with machine learning, the device offers veterinarians a cost-effective alternative to lab-grade systems. Future work will focus on refining data collection across breeds, expanding the ML dataset, and moving toward clinical validation to establish the device as a diagnostic support tool in veterinary care.