Who's a Good Boy? Reinforcing Canine Behavior using Machine Learning in Real-Time

  • 2021-01-07 05:38:09
  • Jason Stock, Tom Cavey
  • 25

Abstract

In this paper we outline the development methodology for an automatic dogtreat dispenser which combines machine learning and embedded hardware toidentify and reward dog behaviors in real-time. Using machine learningtechniques for training an image classification model we identify threebehaviors of our canine companions: "sit", "stand", and "lie down" with up to92% test accuracy and 39 frames per second. We evaluate a variety of neuralnetwork architectures, interpretability methods, model quantization andoptimization techniques to develop a model specifically for an NVIDIA JetsonNano. We detect the aforementioned behaviors in real-time and reinforcepositive actions by making inference on the Jetson Nano and transmitting asignal to a servo motor to release rewards from a treat delivery apparatus.

 

Quick Read (beta)

loading the full paper ...