RollingWheels: Wheelchair IMU & GPS Dataset

The RollingWheels Dataset contains multimodal time-series data collected from smartphones attached to real manual wheelchairs. This dataset is designed to support research in surface type classification, wheelchair navigation, and activity recognition. It includes high-frequency inertial motion data (Accelerometer, Gyroscope), atmospheric pressure, and GPS trajectories.

Index

• Overview
• Directory Structure
• Sensor Specifications
• File Naming Convention
• Data Format
• License

---

Overview

The data was collected by mounting smartphones (e.g., Samsung Galaxy S7, Motorola moto G 7/8/9) to wheelchairs. The dataset is split into two primary categories to facilitate different machine learning tasks:

1. Labeled Data (Without GPS): Focused on surface type identification across different countries (USA, Vietnam). These sessions were conducted in controlled environments to ensure ground truth for surface types.
2. Unlabeled Data (With GPS): Focused on real-world navigation and route mapping in European cities, capturing naturalistic wheelchair usage.

---

Directory Structure

The dataset is organized by labeling status, geography, and surface/device type as shown in the file hierarchy:

RollingWheels/

• Datasets/
  • Labeled_Data_Without_GPS/
    • USA/
      • SurfaceTypeID_1/ ... SurfaceTypeID_10/
    • Vietnam/
      • SurfaceTypeID_1/ ... SurfaceTypeID_12/
  • Unlabeled_Data_With_GPS/
    • Europe/
      • Austria/
        • Phone 7/
        • Phone 8/ (e.g., Mirabellplatz_GPSData.csv)
        • Phone 9/
      • France/
      • Germany/

---

Sensor Specifications

Data captured via smartphone onboard sensors include:

Sensor	Measurement	Typical Units
Accelerometer	Linear Acceleration (3-axis)	m/s²
Gyroscope	Angular Velocity (3-axis)	rad/s or deg/s
Pressure	Atmospheric Pressure/Altitude	hPa
GPS	Geospatial Positioning	Latitude/Longitude

---

File Naming Convention

Labeled Data (USA & Vietnam)

Files in the labeled directories follow a structured naming convention to encode the date, surface, hardware, and subject:

YYYY-MM-DD_SurfaceTypeID_X_DeviceModel_exp#_subject#.csv

• Example: 2019-08-30_SurfaceTypeID_1_SamsungGalaxyS7_exp3_subject1.csv

Unlabeled Data (Europe)

Files are typically named based on the specific location or landmark where the trace was recorded:

<LocationName>_GPSData.csv

• Example: Mirabellplatz_GPSData.csv

---

Data Format

All data is provided in .csv format. While column headers may vary slightly between Android and iOS collection tools, the standard structure includes:

• timestamp: Time of measurement (Unix epoch or relative elapsed time).
• accel_x, accel_y, accel_z: 3-axis acceleration.
• gyro_x, gyro_y, gyro_z: 3-axis rotation.
• pressure: Barometer data (useful for detecting ramps or elevators).
• latitude, longitude: GPS coordinates (available in "Unlabeled" subset only).

---

Usage Example (Python)

To load and inspect a surface-specific file:

import pandas as pd
import os

# Define path to a specific surface experiment
file_path = "Datasets/Labeled_Data_Without_GPS/USA/SurfaceTypeID_1/2019-08-30_SurfaceTypeID_1_SamsungGalaxyS7_exp3_subject1.csv"

# Load Dataset
df = pd.read_csv(file_path)

# Preview sensor data
print(df[['accel_x', 'accel_y', 'accel_z']].head())