Dataset Viewer
Auto-converted to Parquet
Search is not available for this dataset
video
video
label
class label
176 classes
020250828132109
020250828132109
020250828132109
020250828132109
120250829112715
120250829112715
120250829112715
120250829112715
220250829114053
220250829114053
220250829114053
220250829114053
320250829124537
320250829124537
320250829124537
320250829124537
420250829124707
420250829124707
420250829124707
420250829124707
520250829130046
520250829130046
520250829130046
520250829130046
620250829132431
620250829132431
620250829132431
620250829132431
720250829132944
720250829132944
720250829132944
720250829132944
820250829140325
820250829140325
820250829140325
820250829140325
920250829142148
920250829142148
920250829142148
920250829142148
1020250829144220
1020250829144220
1020250829144220
1020250829144220
1120250829151050
1120250829151050
1120250829151050
1120250829151050
1220250829153125
1220250829153125
1220250829153125
1220250829153125
1320250829153901
1320250829153901
1320250829153901
1320250829153901
1420250829154056
1420250829154056
1420250829154056
1420250829154056
1520250829154458
1520250829154458
1520250829154458
1520250829154458
1620250829160009
1620250829160009
1620250829160009
1620250829160009
1720250829163902
1720250829163902
1720250829163902
1720250829163902
1820250829165457
1820250829165457
1820250829165457
1820250829165457
1920250829165805
1920250829165805
1920250829165805
1920250829165805
2020250829170007
2020250829170007
2020250829170007
2020250829170007
2120250829172221
2120250829172221
2120250829172221
2120250829172221
2220250829183629
2220250829183629
2220250829183629
2220250829183629
2320250829200206
2320250829200206
2320250829200206
2420250829200212
2420250829200212
2420250829200212
2420250829200212
2520250829200558
End of preview. Expand in Data Studio
YAML Metadata Warning: empty or missing yaml metadata in repo card (https://huggingface.co/docs/hub/datasets-cards)

Flywheel - Frodobots : Excavator 100 Dataset

A comprehensive multi-modal dataset for excavator operation, featuring synchronized video recordings from multiple camera angles (front, left, back, and right angle) and precise joystick control data for machine learning applications in construction automation and operator behavior analysis. This dataset is made with collaboration with frodobots.ai and bitrobot.ai.

Dataset Description

This dataset captures real excavator operations through synchronized multi-camera video recordings and corresponding joystick control inputs. Common usecases of the dataset:

  • Machine Learning Applications: Training models for excavator operation prediction, automation, and operator assistance systems
  • Computer Vision Research: Multi-view video analysis, object tracking, and scene understanding in construction environments

Dataset Statistics

  • Total Sessions: 176
  • Total Unique Duration: 24h 38m 28.4s
  • Total Raw Video Duration: 96h 26m 6.5s
  • Total File Size: 84.30 GB

Video Description

Each recording session contains synchronized video footage from 4 different camera angles, providing comprehensive coverage of the excavator operation from multiple perspectives.

Video Metadata

  • Resolution: 1920 x 1080
  • FPS: 25 frames per second
  • Compression: HEVC(H.265)

Camera Angles & Naming Convention

Each folder contains 4 synchronized video files with the naming pattern D0X_YYYYMMDDHHMMSS.mp4

Prefix Camera Angle
D01 Front Angle
D02 Left Angle
D03 Back Angle
D04 Right Angle

Filename Structure

D0X_YYYYMMDDHHMMSS.mp4
β””β”€β”¬β”€β”˜ β””β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”˜
  β”‚          └── Timestamp: Year-Month-Day-Hour-Minute-Second
  └── Camera ID: 01=Front, 02=Left, 03=Right, 04=Rear

Example: D01_20250828132109.mp4 = Front camera recording from August 28, 2025 at 13:21:09 (timestamp start time)

Data Structure

excavator_dataset/
β”œβ”€β”€ 20250828132109/
β”‚   β”œβ”€β”€ D01_20250828132109.mp4    # Front camera
β”‚   β”œβ”€β”€ D02_20250828132109.mp4    # Left side camera  
β”‚   β”œβ”€β”€ D03_20250828132109.mp4    # Back side camera
β”‚   β”œβ”€β”€ D04_20250828132109.mp4    # Right camera
β”‚   └── D05_20250828132109.csv    # Synchronized control data
β”œβ”€β”€ 20250829112715/
β”‚   β”œβ”€β”€ D01_20250829112715.mp4
β”‚   β”œβ”€β”€ D02_20250829112715.mp4
β”‚   β”œβ”€β”€ D03_20250829112715.mp4  
β”‚   β”œβ”€β”€ D04_20250829112715.mp4
β”‚   └── D05_20250829112715.csv
└── ...

Parquet Dataset File

Overview

For efficient data access and analysis, the entire dataset has been consolidated into a single Parquet file: excavator_dataset.parquet. This file provides a session-level index of all recordings with direct file path references and dual timestamp formats.

Parquet File Structure

Each row in the parquet file represents one complete recording session with all associated files and metadata.

Column Description

Column Data Type Description Example
session_folder string Session directory name (timestamp format YYYYMMDDHHMMSS) "20250828132109"
timestamp_start int64 Session start time (Unix timestamp in milliseconds) 1756387269080
timestamp_stop int64 Session end time (Unix timestamp in milliseconds) 1756387478920
timestamp_start_str string Human-readable start time "2025-08-28 13:21:09.080"
timestamp_stop_str string Human-readable stop time "2025-08-28 13:24:38.920"
duration_seconds float Total session duration in seconds 209.83999633789062
front_angle_file string Relative path to front camera video (D01) "20250828132109\D01_20250828132109.mp4"
left_angle_file string Relative path to left camera video (D02) "20250828132109\D02_20250828132109.mp4"
back_angle_file string Relative path to back camera video (D04) "20250828132109\D04_20250828132109.mp4"
right_angle_file string Relative path to right camera video (D03) "20250828132109\D03_20250828132109.mp4"
action_data_file string Relative path to joystick control data CSV "20250828132109\D05_20250828132109.csv"

Action Data Format

CSV Structure

Each joystick_data.csv file contains timestamped joystick positions with the following columns:

timestamp,filename,marker_id,normalized_x,normalized_y
2025-08-28 13:21:09.080,D05_20250828132109.mp4,0.0,-0.0042733113763212,-0.0030821448966519
2025-08-28 13:21:09.080,D05_20250828132109.mp4,1.0,0.0250226657342554,-0.00076719496794
2025-08-28 13:21:09.120,D05_20250828132109.mp4,0.0,-0.0041104821925982,-0.001967974023523
2025-08-28 13:21:09.120,D05_20250828132109.mp4,1.0,0.0210680148333473,0.0003680537962328
...
  • timestamp: Time in seconds from recording start (matches video timestamps)
  • filename: The filename of the CSV
  • marker_id: Which side of the joystick (0 = Right Joystick, 1 = Left Joystick)
  • normalized_x: Joystick X-axis position
  • normalized_y: Joystick Y-axis position

Sampling Rate

  • Video: 25 FPS (40ms intervals)
  • Joystick Data: 25 Hz (40ms intervals, synchronized with video frames)

Action Data Coordinate System

Marker Assignment & Excavator Functions

  • Marker ID 0: Right joystick (operator's right hand)
    • Controls: Bucket (X-axis) and Boom (Y-axis)
  • Marker ID 1: Left joystick (operator's left hand)
    • Controls: Cab (X-axis) and Arm (Y-axis)

Axis Definition & Excavator Operations

Marker ID 0 - Right Joystick

X-axis: Bucket Control (lateral movement)

  • Negative values (-1.0): Bucket close/curl in (scoop material)
  • Positive values (+1.0): Bucket open/curl out (dump material)

Y-axis: Boom Control (longitudinal movement)

  • Negative values (-1.0): Boom down (lower boom arm)
  • Positive values (+1.0): Boom up (raise boom arm)

Marker ID 1 - Left Joystick

X-axis: Cab Control (lateral movement)

  • Negative values (-1.0): Cab rotate left (counterclockwise)
  • Positive values (+1.0): Cab rotate right (clockwise)

Y-axis: Arm Control (longitudinal movement)

  • Negative values (-1.0): Arm down/retract (pull arm toward cab)
  • Positive values (+1.0): Arm up/extend (push arm away from cab)

Excavator Function Summary

Marker Axis Function -1.0 0.0 +1.0
0 (Right) X Bucket Close/Curl In Neutral Open/Curl Out
0 (Right) Y Boom Down Neutral Up
1 (Left) X Cab Rotate Left Neutral Rotate Right
1 (Left) Y Arm Down/Retract Neutral Up/Extend

Coordinate Range

All coordinates are normalized to the range [-1, 1] where:

  • 0, 0: Neutral/center position (joystick at rest, no excavator movement)
  • Β±1: Maximum deflection in respective direction (full excavator function activation)
  • Values are clipped/normalized to ensure the ranges of joystick motion remain within bounds

Excavator Operation Context

This coordinate system directly maps to standard excavator controls:

  • Digging sequence: Boom down (-Yβ‚€) β†’ Arm extend (+Y₁) β†’ Bucket close (-Xβ‚€)
  • Dumping sequence: Boom up (+Yβ‚€) β†’ Bucket open (+Xβ‚€)
  • Positioning: Cab rotation (Β±X₁) for directional changes
  • Reach adjustment: Arm retract/extend (Β±Y₁) for distance control

How to Use the Dataset

This dataset provides synchronized multi-angle video recordings (.mp4 at 25 FPS) and joystick control data (.csv at 25 Hz, 40 ms intervals). Each video frame directly corresponds to a row in the joystick CSV, so all sensor and visual information is time-aligned for frame-by-frame analysis.

Downloads last month
2,197