Skip to Content

A Vision-based Quadrotor Swarm for the participation in the 2013 International Micro Air Vehicle Competition (Toulouse, France)

1. Abstract

We present a completely autonomous solution to participate in the 2013 International Micro Air Vehicle Indoor Flight Competition (IMAV2013). Our proposal is a modular multi-robot swarm architecture, based on the Robot Operating System (ROS) software framework, where the only information shared among swarm agents is each robot's position. Each swarm agent consists of an AR Drone 2.0 quadrotor connected to a laptop which runs the software architecture.

In order to present a completely visual-based solution the localization problem is simplified by the usage of ArUco visual markers. These visual markers are used to sense and map obstacles and to improve the pose estimation based on the IMU and optical data flow by means of an Extended Kalman Filter localization and mapping method. Taking into account the other swarm agents' positions a free-collision trajectory for each drone is generated by using  a combination of state of the art trajectory planning algorithms: probabilistic road maps, a potential field map algorithm and an A-Star algorithm. The last element of our autonomous agent is a robust mid-level controller which executes the generated trajectory commands.

This paper also presents a discussion of the performance of our architecture on various simulated and experimental flights on a replica of the IMAV2013 environment. The presented solution and the performance of the CVG_UPM team were awarded with the First Prize in the Indoors Autonomy Challenge of the IMAV2013 competition.

The presented solution has been designed using AR Drone 2.0 quadrotors, see (up-left),
and a replica of the IMAV2013 indoors challenge environment where the map and the obstacles were marked using ArUco markers, see (b). This environment, used during experimental flights, consists of a small window, a big window and 8 poles. The position of the wall is previously known except for the positions of the windows along it are unknown. The positions of the 4 corner poles are previously known, however, the positions of the 4 poles in the middle are unknown. The windows and poles represent obstacles that must be avoided during flight. The subfigure (c) shows a experimental flight where one of the drones is crossing the unknown poles area. The unknown poles are robustly located on previous laps, where the drone performs laps around the known poles, ensuring a good estimation of their positions.  Our framework also allows to test partner collision avoidance during experimental flights, as shown in (d), where a drone is waiting until the path to cross the big window is clear. The flights shown in (c) and (d) are shown at the end of this entry of our website.


2. Motivation

The motivation of this work is the design of a solution to participate in the 2013 edition of the International Micro Air Vehicle Flight Competition (IMAV2013). The IMAV Flight Competition is the most relevant European competition in the fields of Autonomous Aerial Robotics and Small Remotely Piloted Air Systems (sRPAS). Our research group, the Computer Vision Group (CVG), was awarded for its performance in the 2012 edition of the {IMAV} competition showing the potential of our group in the development of autonomous Unmanned Aerial Systems (UAS). The learning experience obtained from the indoor dynamics competition encouraged us to keep working in the same direction and also to try a swarming approach in the 2013 edition. Our motivation for participating in such competitions is to develop autonomous systems which can be later modified to perform civilian applications. The 2013 edition's rules are significantly different with respect to former edition's. In IMAV2013 there was only one indoors competition which requires a high level of autonomy. The scenario has some fixed and previously known obstacles (a wall and four fixed poles) and several obstacles located at unknown positions (two windows and four pole obstacles). The indoor competition includes various challenges, including flying through a window, flying through an obstacle zone, target detection and recognition, path following and precision landing, among others.

3. Videos



4. Researchers/Authors

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