Automatically Annotated Dataset of a Ground Mobile Robot in Natural Environments via Gazebo Simulations

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Unmanned Ground Vehicles (UGVs) require robust navigation systems for diverse natural terrains.
• Developing and testing these systems often relies on extensive real-world data collection, which can be time-consuming and costly.
• Synthetic datasets offer a controlled and scalable alternative for training and evaluation.

Purpose of the Study:

• To introduce a novel synthetic dataset for Unmanned Ground Vehicle (UGV) navigation research.
• To provide researchers with a valuable resource for developing and benchmarking autonomous navigation algorithms in natural environments.
• To release the simulation code, enabling further customization and expansion of the dataset.

Main Methods:

• Gazebo simulations were used to generate data for a Husky mobile robot.
• The robot was equipped with a tridimensional (3D) Light Detection and Ranging (LiDAR) sensor, stereo camera, Global Navigation Satellite System (GNSS), Inertial Measurement Unit (IMU), and wheel tachometers.
• LiDAR scan points and camera image pixels were automatically labeled with object classes using unique reflectivity values and flat colors.

Main Results:

• A comprehensive public dataset comprising simulated UGV movement in natural environments was created.
• The dataset includes 3D pose ground-truth information.
• Both ROS bag files and human-readable data formats are provided for accessibility.

Conclusions:

• The released synthetic dataset and simulation code facilitate research in UGV navigation.
• This resource supports supervised learning and benchmarking for autonomous navigation in challenging natural terrains.
• The availability of the dataset and code accelerates the development and validation of UGV technologies.