ARGOS : Augmented Reality-based Dataset Generation Framework for Ground Swarm Robotics

Purpose :

Intelligent industrial systems are integral parts of the broader Artificial Intelligence of Things (AIoT) ecosys-tem in industrial settings. Bringing information from multiple agents enable to bring more advanced androbust functionalities or services provided by the AIoT. Mobile systems that navigate and gather informa-tion in these environments can have various perception and motion capabilities. This variability, along withchanges in viewpoints, detection errors, or localization inaccuracies of scene objects, presents multiple chal-lenges. These challenges underscore the need for versatile datasets, which can facilitate the development ofnew techniques to manage variability and appearances resulting from multiple viewpoints and robot swarmsexecuting collaborative tasks and functionalities. 

This dataset represents a multi-robot scene and is based on a tool designed for robotic data augmentation, whether for single-agent or multi-agent scenarios, using video streams or point clouds to generate augmented datasets.

Dataset Description :

For this dataset, two robots with different motor capabilities were selected: the "Jaguar-V4-With-Arm" robot and the "Segway" robot. An embedded computer was installed, along with RGB cameras and LiDAR sensors of types Pandar64 and Velodyne VLP16, as well as RealSense RGB-D cameras. The experimental scene is located within a Vicon reference system, serving as ground truth.

This dataset is provided in standardized ROS1 bag file format.

The proposed method relies on the standardized ROS data format and is capable of generating images with the object integrated into the scene, as well as producing semantic images based on captured images (messages of type sensor_msgs/Image).

For LiDAR point clouds, the method generates an augmented point cloud of type sensor_msgs/PointCloud2, containing the original point cloud with added virtual objects. These objects are identifiable by a unique ID in the RGBA channels of the colored point cloud. 

In the provided dataset, a model from the Unity example Assets/ExampleAsset/Paint 5G bucket low is added to the scene at the position of Aruco1. Other models in FBX format or directly from the Unity Asset Store can also be incorporated into the scene on request. This augmentation process is based on the positions provided by the /tf and /tf_static topics of the robots, in the form of tf2_msgs/TFMessage ROS messages. During this augmentation phase, the ground truth data of the objects added to the scene is returned in the generated tf topic.

This work is partially funded under the ANR-21-ASRO-0005-01 agreement attached to the SCOPES project (ASTRID ASRO 2021 scheme funded by the Agence de l’Innovation de Défense (AID))