Skip to Content

A System for the Design and Development of Vision-based Multi-robot Quadrotor Swarms

  

Fig 1. (left) Experimental flight of the basic instance of the framework, which is characterized by swarm behavior, using 5 AR Drones. The row of columns represents a virtual wall with a single 1.5~m opening in the middle. (right) Visualization of a simulated 5 drone flight in the Pinball Mission using the Rviz Interface. Drone axes: [front] red x axis, [leftwards] green y axis and [upwards] blue z axis.

We present a cost-effective framework for the prototyping of vision-based quadrotor multi-robot systems, which core characteristics are: modularity, compatibility with different platforms and being flight-proven. The framework is fully operative, which works in simulation and in real flight tests of up to 5 drones, and was demonstrated with the participation in the 2013 International Micro Air Vehicle Indoor Flight Competition (Toulouse, France) where it was awarded with the First Prize in the Indoors Autonomy Challenge.

The motivation of this framework is to allow the developers to focus on their own research by decoupling the development of dependent modules, leading to a more cost-effective progress in the project. This goal is achieved by enabling separate modules to be included along with legacy modules of the framework to be tested in simulation or experimental flights, thus accelerating the development at early stages of the project. Leveraging the modularity capabilities of the Robot Operating System (ROS) middleware framework, each module's behavior can be simulated from small parts of the architecture to the complete system.

The basic instance of the framework that we propose includes several modules that can be reused and modified, such as: a basic sequential mission planner, a basic 2D trajectory planner, an odometry state estimator, localization and mapping modules which obtain absolute position measurements using visual markers, a trajectory controller and a visualization module. The basic instance of the framework, which is compatible with the cost-efficient and reliable platform Parrot ARDrone 2.0, achieves successful experimental flight of multi-robot swarms for pure navigation missions and serves as a first building block for more complex solutions.

 

Fig 2. The heterogeneous multi aerial robot system communicates through a WLAN. 

 

Fig 3. Our Implementation of the Modules of the Framework. The architecture is modular and is built using the Robot Operating System (ROS) framework. Each white box represents a module, and the green text inside it are configuration parameters. The localization module fuses the odometry based estimation with the visual markers feedback. This module broadcasts the estimated pose to the mission and trajectory planning modules, to the controller module, to the obstacle detector, and to the other robotic agents. The trajectory planner gives free-collision trajectories to the trajectory controller which also receives yaw commands given by the yaw commander. The trajectory controller generates commands to the drone. The mission planner module monitors the mission given mission points to the trajectory planner. The hypothalamus module receives the estimated position of the other robots and communicates it to the trajectory planner.

 The framework is fully operative, which is shown in the paper through simulations and real flight tests of up to 5 drones, and was demonstrated with the participation in an international micro-aerial vehicles competition. Since we trust in our system and we believe in sharing with the scientific community, we decided to make our framework open-source. This way, anyone interested in working with multi-robot aerial applications can use our framework as a starting point of their research and as a tool to test their own algorithms. The link to the framework's code Git repository is specified in the following website: https://github.com/Vision4UAV/cvg_quadrotor_swarm .

3. Videos

 

More Videos:

3.1 Mission: The Hole

 

 

3.2 Mission: The Pinball

 

 

4. Researchers/Authors

The PhD Students and Researchers that have actively worked on this project are: