Project Abstract

PUPPy is a deep learning model designed to process accelerometer and gyroscope data from a smartphone to predict a user’s motion. We compare PUPPy to existing deep learning models with a new measure of cumulative error to evaluate model performance. We introduce a novel concept of sequence prediction and include concepts from existing solutions to create a better performing model.

Team Members

Presentation Video

Poster Download

Interactive Demo

Link to demo using Google Colab.

Video Presentation Slides

View the video presentation slides

Example of Path Tracking

Path sample for PUPPy, IONet, and Ground Truth