The full autonomous control of a rotary-wing Micro Aerial Vehicle (MAV) relies on measurement inputs from sensors to allow the vehicle to maintain a constant position in a 3D environment. Monocular Visual SLAM (VSLAM) is a particularly efficient sensing method in terms of payload cost. A single camera can be used to provide a full 6-DOF pose measurement, however, this is at the cost of increased communication bandwidth and computational requirements, often resulting in low-rate and delayed feedback measurements. This thesis presents an investigation into the use of VSLAM feedback to stabilise the full 3D position of a MAV. To reach this goal, experimental work is conducted using small rotary-wing platforms in indoor environments. Platforms include the static 3-DOF Quanser and the 6-DOF AR.Drone2.0 quadrotor. Theoretical dynamic models are developed and simplified into decoupled linear sub-systems. The sensing properties of VSLAM are also experimentally identified for both static and dynamic flight scenarios. The control challenge of low-rate delayed measurements is overcome by applying discrete-time LQG control design methods. A discretisation method is detailed which can represent arbitrary delays in a discrete state-space form whilst preserving the dynamic behaviour. A method of tuning the LQG to preserve the reference tracking response is presented, making use of the discrete delayed model structure. Initially the presented control design is experimentally applied to SISO systems to assess the performance, making use of high accuracy external tracking systems. The result is then extended to the full 6-DOF helicopter where free flight is demonstrated, using VSLAM as the only pose measurement, with no additional sensors required to initialise VSLAM.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:687191 |
| Date | January 2014 |
| Creators | Wood, Kieran Thomas |
| Publisher | University of Bristol |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
Page generated in 0.0122 seconds