Distributed State Estimation and Control of Autonomous Quadrotor Formations Using Exclusively Onboard Resources

The navigation of unmanned aerial vehicles operating in environments without global positioning systems, including global navigation satellite systems and motion capture systems, is a recent research topic, without much work reported in the literature. In indoor applications, particularly, small-scale vehicles are subjected to severe power and weight constraints, limiting their overall navigation capabilities. In such scenarios, multi-vehicle systems can be used in order to mitigate the impact of limited capabilities at the individual vehicle level. If, additionally, a group of vehicles has to maintain a specific spatial topology, well-established formation control algorithms can be used as long as information about mutual inter-vehicle positioning is available. This information can be directly acquired using relative positioning systems on each vehicle. This solution enables the multi-vehicle system to reduce its dependency on absolute localization systems and explicit inter-vehicle communications. Additionally, multi-vehicle formation control can be achieved in either fully distributed or decentralized fashion, reducing the need for external and/or centralized units supervising the system. This thesis introduces two novel relative positioning systems for multi-vehicle formations, focusing on maximizing the number of detected team members while remaining accurate and light enough to allow their deployment on small-scale vehicles: i) a camera-based system that enables a scalable deployment on multiple vehicles; ii) an infrared-based system that provides several hardware and software enhancements with respect to systems reported in the literature using the same technology. The camera-based sensor model can be leveraged as a tool for optimizing the design parameters to meet specific accuracy requirements and allows the system to achieve highly accurate relative localization measurements using low-resolution cameras. The infrared-based system uses miniature omni-directional infrared beacons deployable in small sets on each vehicle which, together with dedicated estimation and calibration algorithms, ensures a adaptability to any 3D geometry of the carrying vehicle. Such innovative design principles result in a system which enables a direct measurement of the relative attitude, and is more flexible, lighter, and less power-hungry than state-of-the-art devices, while providing similar accuracy. Novel formation control methods that tackle limitations arising from the exclusive use of relative positioning systems are an extra contribution of the thesis. A graph-based formation control algorithm has been extended so that sensing constraints could be taken into account when a vehicle has to observe multiple neighbors. This extension consists of enabling each vehicle to control the occupied area of the limited field of view of its sensor, while it moves to the right place in the formation. This in turn provides additional flexibility for the formation topology despite the inter-vehicle sensing limitations. A formation steering algorithm capable of providing a consistent and simultaneous motion direction to all team members has also been developed. This was achieved without requiring artificial landmarks in the environment and/or additional communication overhead between the vehicles. The proposed steering algorithm increases the reactiveness of the formation control when compared to canonical methods relying on leader vehicles.