Experimental Data

Access the datasets collected during our Fusion Robotics project, understand their format, and how they are processed.

Open Data for Reproducibility and Research

In the spirit of open source and collaboration, we are sharing the experimental data collected during the development and testing of our Quadruped-UR5 testbed. This data primarily consists of ROS bag files containing timestamped sensor readings and system states recorded during autonomous runs.

Potential uses for this data include:

  • Reproducing our pose estimation comparisons.
  • Testing alternative sensor fusion algorithms.
  • Analyzing quadruped gait performance under different parameters.
  • Developing simulation models based on real-world sensor characteristics.

Accessing the Data

The datasets are hosted on GitHub within the project repository. Please follow the link below to access the data directory:

Access Experimental Data on GitHub

Data Format and Content

The data is primarily stored as ROS 2 bag files (`.db3` or directory format). You will need a ROS 2 Humble installation to easily inspect and replay these bags using tools like ros2 bag info and ros2 bag play.

Key topics typically included in the bags are:

  • /imu/data (sensor_msgs/Imu)
  • /imu/orientation (geometry_msgs/Quaternion)
  • /imu/pose_estimate (geometry_msgs/PoseStamped)
  • /orb/pose (geometry_msgs/PoseStamped)
  • /servo/state (std_msgs/Float64MultiArray)
  • /servo/command (std_msgs/Float64MultiArray)
  • /tf and /tf_static

Data Filtering and Usage

The raw sensor data undergoes processing within our nodes:

  • IMU Bias Removal: Initial calibration calculates and subtracts biases.
  • IMU Filtering: Low-pass filtering applied; velocity decayed.
  • Visual Pose Smoothing: Savitzky-Golay filter applied to translation estimates.
  • UI Plotting: Key topics buffered for real-time web UI visualization.
  • Pose Estimation Comparison: IMU-based and visual-based poses published for analysis.

Data Visualization Examples

Examples of typical data plots generated from our experiments.

Please refer to any accompanying README.md files within the data repository for specific details about individual datasets or recording conditions.