Cooperative SLAM Optical Sensor-Based Architecture for Multi-UAVs systems

Simultaneous Localization And Mapping (SLAM) is a well-known problem in a variety of cases, such as ground robots, drones, cars, and underwater robots. To perform efficiently SLAM is even more crucial when autonomous navigation must be guaranteed. UAVs are widely taking the scene in many applications, like search and rescue missions, real-estate applications, agriculture, system logistic, and package delivery, thanks to their dexterity in movements, even in narrow environments. Many of the state-of-the-art SLAM algorithms find a common factor in the usage of Visual - Visual Inertial Odometry (VO-VIO). Indeed, more and more solutions to the problem are developed, bringing new approaches, or improving existing ones. Recent researches include neural-networks too. The search for more information to integrate into SLAM algorithms has brought to investigate systems with multiple agents that collaborate to build a global map and use the shared information to improve the localization process. The objective of this thesis is the development of a strategy to perform multi-agent SLAM for a fleet of drones. Different combinations of inertial sensors, cameras, and range finders are tested, along with sensor fusion algorithms with the idea of improving the localization process as far as possible. Then, general SLAM algorithms are described, analyzed, and tested. The development is thought for a case in which each drone is provided with the same sensor suite, that comprehends a tracking camera module, a depth camera module, and a low-budget onboard computer. Anyway, it is possible to use the same logic to combine data coming from different robots, e.g. system which combines both ground robot and drones, with different sensor suites. This project is carried on in the ROS environment, which allows taking advantage of already existent libraries and communication protocols. Precisely, it is shown how a ROS package has been created and developed to generate a set of nodes that provide features for the multi-agent SLAM, such as data compression and decompression, point cloud filtering, and concatenation. It is shown that, even with limited computational resources and a standard wireless connection, it is possible to share information without loss of data consistency and a global map can be created and then forwarded to each agent.