Use this URL to cite or link to this record in EThOS:
Title: Optical Tau guidance of Unmanned Aerial Systems
Author: Dadswell, Christopher
Awarding Body: University of Liverpool
Current Institution: University of Liverpool
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Access from Institution:
The use of Unmanned Aerial Systems (UAS) or 'drones' as they are more commonly known, has increased dramatically in recent years. Innovations in electrical power systems as well as improvements in aircraft autonomy technologies have driven a significant consumer surge in the use of small UAS. At the same time, there has been a simultaneous large increase in commercial and military use of UAS across a range of sizes and missions. There is significant evidence that unmanned aircraft experience accidents at a much greater rate than is observed in human-crewed aviation. A number of surveys investigating the general public's perception of drones have also reported a general distrust of the emerging technology, in part due to negative media coverage and high-profile mishaps. Though inadequate reporting of UAS accidents seems to mask the scale of the problem somewhat, analysis of accident rates for military UAS reveals unacceptably high failure rates for their platforms. Common problems include an inability to sense and avoid hazards, unstable communication links, and failures during landing. New technology solutions are required to overcome the problems outlined above. The application of Tau theory to the guidance function of unmanned aircraft offers one such possibility. At the University of Liverpool, ecological Tau theory has been established as the basis to understand and model pilot behaviour, based on the optical parameter 'Tau', or time-to-contact. Tau guidance is inherently reactive to external obstacles and hazards, does not require any external signals or infrastructure to function, and is well suited to soft landing manoeuvres, which have all been identified as problematic functions for UAS. This thesis describes work carried out to implement and analyse Tau-guided UAS in the context of simulated landing manoeuvres. Landing in difficult or dynamic environments are examined in detail to illustrate that the inherent reactivity of Tau guidance is useful in these demanding situations. Two main scenarios are considered: rotary-wing maritime landings, and fixed-wing landings on uneven ground. Existing computer vision techniques to measure time-to-contact were identified and analysed. Three previously established techniques for estimating Tau with monocular video cameras were compared: a dimension-tracking method; an optical flow divergence method, and a direct gradient-based method. The direct gradient method was selected as the most reliable and widely applicable of the three approaches and was thoroughly investigated to find the limits of its operation. It was found that, for the camera used in testing, the method is most effective at measuring time-tocontact when the ground truth value is between 10 and 1 seconds. Above 10 seconds time-to-contact, measurements became too noisy to be usefully applied, and close to the contact point high optical flow causes divergence of the estimate. The accuracy of the method is closely linked to the magnitude of optical flow perceived by the camera, and it was found that manipulating camera frame rate and resolution through subsampling is useful to maintain accuracy throughout the manoeuvres. High resolution cameras can be used to extend the accurate measurement interval above 10 seconds, and high frame-rate cameras can be used to extend the range closer to the contact point. One of the key limitations of the gradient method is that constant image brightness is assumed across video frames. This leads to reduced accuracy of the Tau sensor if the illumination of the visual scene changes, such as when artificial lights are turned on or off, or when shadows cast by a vehicle or cloud encroach on the scene. This thesis reports upon the development of a new extension to the direct gradient method that removes this limitation/assumption. This innovation expands the operational envelope of Tau sensors that use gradient methods by increasing their accuracy when compared to existing methods under changing light conditions. The enhanced Tau perception algorithm was tested in simulation using rotary- and fixed-wing UAS platforms in a number of different scenarios. The algorithm was implemented to use one simulated monocular camera to measure time-to-contact with an approaching object or surface. The rotary-wing platforms were tested in a frigate deck landing scenario, where the targeted landing point is the ship deck. Landings were analysed in a range of sea state conditions that develop different amounts of deck motion for the aircraft to contend with during landing. Tau estimates from the enhanced direct gradient method were used as a feedback control variable to the aircraft autopilot to perform successful landings. Using only the time-to-contact as a control variable, the tau guidance system allowed the simulated aircraft to perform smooth landings with lower touchdown velocities when compared to the more commonly used constant descent rate guidance strategy. For a sea state 4 deck landing, the Tau guidance system delivered an average touchdown velocity of 0.2 m/s over 20 different landings, while the softest landing the constant descent rate controller delivered was 0.5 m/s, with an average of 1.3 m/s over 20 landings. However, it was found that poor Tau estimation performance close to the deck increased touchdown velocity, especially in high sea-state conditions. It was therefore recommended that the optical Tau estimation method be augmented with an ultrasonic sensor for increased time-tocontact measurement accuracy during the last 1-2 seconds of the manoeuvre. The algorithm was sufficiently robust to allow deck landings to be performed in any sea state, so long as the aircraft had sufficient heave control power to deal with the heave demands of the moving deck. A heave dynamics model was combined with analytic expressions for the Tau guide spatial parameters to produce a new tool to predict whether an aircraft would be able to perform a Tau guided landing in a specific sea state. The tool was tested using an SH60B Seahawk helicopter, and correctly predicted that the aircraft would be unable to make a successful deck landing in sea-state 6 or above. Tau guidance systems were also implemented on a simulation model of a small fixedwing aircraft, the 3DR Aero, with a virtual optical Tau sensor integrated into the system. Use of optical Tau to control a fixed-wing aircraft has not previously been demonstrated in operation. Conventional landings were performed on both a standard runway and unprepared sites with changing terrain elevation. Again, it was found that the aircraft was able to consistently make smooth landings using only Tau as a control variable while using a 'perfect' tau measurement, regardless of landing surface. However, when using tau estimates from the virtual tau sensor, tracking performance was too poor for the Tau controller to reliably deliver a safe landing. Further investigation of fixed-wing implementation of optical Tau systems is required. Overall, the research showed that Tau guidance can be usefully and practically applied to address some of the problems identified for UAS operations, chiefly the ability to react to dynamic obstacles and hazards in the environment. This was demonstrated by the landing manoeuvres of a rotary-wing UAS onto a moving ship deck. The system also delivered reduced touchdown velocity in comparison to a common alternative guidance system, reducing the risk of aircraft damage. Similar benefits were exhibited for fixed-wing landings on unknown terrain. The key enabler for these benefits was the development of a new variant of the direct gradient method for Tau perception which expanded the operational envelope of the algorithm.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral