Remote control of a fleet of robotsAs part of the fourth year study project we had the chance to work on this project in collaboration withMITSUBISHI ELECTRIC R&D CENTRE EUROPE. The aim of this project is to develop the tools for controlling afleet of robots. Robots must be able to follow a dynamically calculated path and their position will be measured bycombining local information (odometry, lidar, etc.) as well as that from a video system. The system is developedfrom TurtleBot3 Burger robots because they integrate a lidar, an odometry system at the wheels and arecompatible with ROS. ROS is often used in this type of project because it is a set of open source IT tools fordeveloping programs for robotics. To view the points of the lidars of each robot and the virtual maps created withSLAM (simultaneous localization and mapping) we also used Rviz. In the current circumstances that preventteam members from interacting with each other and with robots, we have chosen to use Gazebo software tosimulate the robots and their environment, to test our programs and film the demonstrations.Our goal is to organize a robot network capable of delivering packages within it. This network makes itpossible to overcome the problem of battery of the robots, in fact each one is limited to a perimeter around itshome or it can be recharged, the robots pass the packages in an order established by a global controller. Thiscontroller monitors the status of each robot and decides the optimal path to deliver packages from one point toanother, it arranges the tasks so as to reduce the overall delivery time using as few robots as possible. Thebehavior of each robot is framed by a state machine, in this way the robot manages its movement from one pointto another, it is positioned in space by odometry and takes into account global and local mapping determined bySLAM in order to avoid dynamic walls and obstacles. The local path taken by the robot is calculated using thepurepursuit algorithm using the Simulink model. Thanks to this architecture, in the event of loss of contact withthe controller, the robot returns to its home (while keeping its usual behavior) while waiting for the resolution ofthe problem.We have also developed a part based on image processing to determine the position of each robot andthus correct the trajectory errors of the latter due to the odometry of the wheels which can diverge. The difficultyencountered is to find correspondences between points in the real environment and their 2D image projection tocalculate distances. For this we use the free OpenCV graphic library and in particular the ArUco library which isbasically used for augmented reality. The markers provided by this library have enough correspondence todetermine the position of the camera. Thus the algorithm that we have developed makes it possible to calculatethe real position of a marker, placed on a robot, in real time, in the repository of reference markers which areplaced on the ground. The latter is very robust and takes into account the perspective, the inclination of thecamera and works with one or more reference markers on as many robots as desired. From this algorithm, wewere able to set up other applications that use distance calculation such as the autonomous parking and vehicletracking function.This parking algorithm allows the robot to find a parking space in a given area. To do this, it detects anArUco marker placed in front of the location which will allow it to detect that it is a parking space and to orientitself to park autonomously.Finally, to manage all these developments we have implemented a man-machine interface thanks toQtCreator. It allows you to send the order to a robot to retrieve or drop a package somewhere with contact detailsor to transmit it to a robot in the network (not necessarily to a neighbor). It also allows you to view the status ofeach robot in real time and to know which package they are transporting.In conclusion this project lays the foundations for distributed control and still has a lot of potential forevolution. Indeed, one can imagine thereafter not only moving on a flat surface but on reliefs or even replacingthe TurtleBot with drones and thus illustrate the current developments of companies like Amazon which hope toset up parcel deliveries by drone in several countries.
Tutors : Rheda Hamouche, Remy Kocik, Genevieve Baudoin