Paddling Improvement System

NYU Concrete Canoe’s machine-learning coaching tool

Live Web Application

Upload paddling session data and receive stroke-level analytics, visualizations, and model-based feedback.

Overview

The Paddling Improvement System (PIS) is a data-driven training interface developed by the NYU Concrete Canoe team. It processes motion data collected from Apple Watches and mobile devices during paddling sessions, segments individual strokes, and evaluates performance using trained machine-learning models.

The system is designed to support consistent technique refinement by making paddling feedback quantitative, repeatable, and accessible.

Full Documentation

Comprehensive technical documentation detailing data collection, preprocessing, stroke segmentation, model training, backend APIs, and frontend processing is available in the project repository.

View GitHub Repository

How to Use the PIS Web App

Download the sample dataset
Download the sample paddling data using the button above. After unzipping the archive, you will find folders named after individual paddlers. Each folder contains five CSV files corresponding to different sensor streams.
Upload the data files
Upload the CSV files into their corresponding prompts on the web application:
- Upload 3 CSVs for Phone — Accelerometer, Gyroscope, Magnetometer
  All three files must be uploaded together using multi-select
  (macOS: ⌘ + click · Windows: Ctrl + click · Mobile: not supported).
- Upload 1 CSV for Left Watch — Left wrist Apple Watch data
- Upload 1 CSV for Right Watch — Right wrist Apple Watch data
Analyze the session
Press Analyze to execute the processing pipeline. The system aligns time-series data, segments strokes, extracts features, and applies trained models to return performance analytics and visual summaries.

Contributors

Ideation: Alex Huang
Development: Shubham Parab · Anthony Lamelas · Ahmad Hassan · Pasang Bhote
Web Page & Frontend Integration: Shubham Parab · Aunirbhan Das