Vision assisted guidance techniques for quadrotor landing

Abstract

Rotary vehicles like quadrotors, have become an essential part of many applications like mapping, object delivery, tracking, patrolling, and communication relay, etc. A key requirement for these applications is that the vehicle needs to perform these tasks autonomously from takeoff to landing. Landing is the most crucial task in the mission, which has a direct impact on the physical safety of the vehicle. Further, in mapping and object delivery applications, the vehicle landing area is constrained, and hence requires precision landing capability. Additionally, external disturbances like winds and visibility add another dimension to challenges in actual landing scenarios. This thesis focuses on the development of vision-assisted guidance techniques for landing of a quadrotor. The primary concern of a guidance strategy is to enable persistent tracking of the landing pad (referred to as target) which could be either stationary or moving and accurately land on the target. Additionally, for a moving target, the velocity of the UAV and the target must be the same to achieve zero closing velocity at the time of landing. Besides this, it is desired that the convergence profile of the UAV velocity to the target velocity is slower in the beginning and faster towards the end for achieving time efficient landing. To this end, we have developed a novel log polynomial guidance for accurately landing the vehicle on the target, while meeting the required velocity characteristics. The target is persistently detected and the distance to the target is estimated using an onboard camera vision algorithm. Based on the estimated distance, the log polynomial guidance law drives the vehicle towards the target and ensures landing with zero closing velocity. We theoretically prove the convergence of the proposed guidance law. The guidance law is applicable to both stationary and moving targets. We have performed simulations and carried out field experiments to demonstrate the efficacy of the guidance law. The proposed guidance law is analyzed in detail for a vertical landing scenario. The guidance law consists of a log polynomial function, which is dependent on the distance between the UAV and the target, and can be controlled by two design parameters. We have carried out a detailed theoretical analysis of the impact of these design parameters on the guidance characteristics. We show that the decay characteristics of the log function, inherently guarantee zero closing velocity and acceleration at touchdown. Analytic conditions are deduced on the two design parameters for achieving, strictly decreasing, initially accelerating, and hovering velocity profiles, respectively. Closed-form expressions are derived for maximum velocity, and engagement time constraints as a function of design parameters. The results of the vertical landing scenario can be used to derive speed profiles of landing in different dimensions (x, y, z). Additionally, it provides a visual representation of the choice of velocity profile, maximum velocity constraint, and engagement time in two-dimensional design parameter space. These results help us in devising the landing behaviors for a robust landing and can give insights to the overall landing process (rather than just the terminal phase) to suit the needs of the application. Any standard autonomous UAV mission is subject to hostile environmental conditions. Landing being an inherently difficult maneuver becomes more challenging due to environmental uncertainties like wind disturbance, uneven landing surface and constraints imposed by target localization errors, uncertain motion of the target and time critical nature of applications. Thus, it becomes imperative that the guidance strategy is robust and it degrades gracefully even in unforeseen conditions, and a decision to abort and/or reinitialize the landing process can be made accordingly. In order to understand these effects, we have carried out qualitative analysis of the log polynomial guidance law in simulations under varying environmental conditions and movement of the target. The simulations have been performed on a realistic quadrotor simulator, Microsoft Airsim with unreal engine. The robustness of the guidance law is further analyzed through, outdoor experiments. These studies show that the proposed guidance law is capable of guiding the vehicle to land accurately on any target, stationary and moving, in reasonably adverse environmental conditions. In summary, this thesis focuses on developing a novel landing guidance law for a quadrotor using vision to land on stationary and moving targets under wind disturbance and uncertain target motion. We have shown convergence analysis and also experimentally validated the guidance principle through realistic simulations and outdoor hardware experiments.