Vision-based control of near-obstacle flight
Lightweight micro unmanned aerial vehicles (micro-UAVs) capable of autonomous flight in natural and urban environments have a large potential for civil and commercial applications, including environmental monitoring, forest fire monitoring, homeland security, traffic monitoring, aerial imagery, mapping and search and rescue. Smaller micro-UAVs capable of flying inside houses or small indoor environments have further applications in the domain of surveillance, search and rescue and entertainment. These applications require the capability to fly near to the ground and amongst obstacles. Existing UAVs rely on GPS and AHRS (attitude heading reference system) to control their flight and are unable to detect and avoid obstacles. Active distance sensors such as radars or laser range finders could be used to measure distances to obstacles, but are typically too heavy and power-consuming to be embedded on lightweight systems. In this thesis, we draw inspiration from biology and explore alternative approaches to flight control that allow aircraft to fly near obstacles. We show that optic flow can be used on flying platforms to estimate the proximity of obstacles and propose a novel control strategy, called optiPilot, for vision-based near-obstacle flight. Thanks to optiPilot, we demonstrate for the first time autonomous near-obstacle flight of micro-UAVs, both indoor and outdoor, without relying on an AHRS nor external beacons such as GPS. The control strategy only requires a small series of optic flow sensors, two rate gyroscopes and an airspeed sensor. It can run on a tiny embedded microcontroller in realtime. Despite its simplicity, optiPilot is able to fully control the aircraft, including altitude regulation, attitude stabilisation, obstacle avoidance, landing and take-off. This parsimony, inherited from the biology of flying insects, contrasts with the complexity of the systems used so far for flight control while offering more capabilities. The results presented in this thesis contribute to a better understanding of the minimal requirements, in terms of sensing and control architecture, that enable animals and artificial systems to fly and bring closer to reality the perspective of using lightweight and inexpensive micro-UAV for civilian purposes.
Keywords: vision-based control ; optic-flow-based control ; obstacle avoidance ; near-obstacle flight ; autonomous unmanned aerial vehicle (UAV) ; micro-air vehicle (MAV) ; contrôle basé sur la vision ; contrôle basé sur le flux optique ; évitement d'obstacle ; vol en environnement encombré ; micro-drônes ; aerial roboticsThèse École polytechnique fédérale de Lausanne EPFL, n° 4456 (2009)
Programme doctoral Informatique, Communications et Information
Faculté des sciences et techniques de l'ingénieur
Institut de microtechnique
Laboratoire de systèmes intelligents
Record created on 2009-05-28, modified on 2016-12-12