This paper presents a mathematical modeling framework designed for air cargo operations of a combination airline, focusing on intercontinental passenger aircraft. This framework optimizes the palletization problem (assigning cargo to Unit Load Devices (ULDs)) and the weight and balance problem (positioning ULDs within the belly of the aircraft). Unlike the traditional approach of solving these problems in sequence, our framework employs a feedback loop between an integrated one-dimensional bin packing with weight and balance and a three-dimensional bin packing. This feedback loop iteratively refines the solution by backpropagating information when items do not fit, reducing the solution space and computation time. The palletization problem is addressed with a three-dimensional bin packing heuristic, while the integrated onedimensional bin packing and weight and balance problem is solved using a mixed integer linear programming formulation. The model employs a lexicographic optimization approach to prioritize maximizing the percent mean aerodynamic chord at zero fuel weight, leading to significant fuel reduction. Additionally, operational objectives aligned with our partner airline’s practices are optimized to reflect actual operations. Scenario analyses validate the model, demonstrating potential fuel savings across various intercontinental routes and aircraft types, with an average fuel saving of 0.53% and reaching up to 1.90%, yielding substantial economic
and environmental benefits.

Neglecting actuator dynamics in nonlinear control and control allocation can lead to performance degradation, especially when considering fast dynamic systems. This thesis provides a novel method to account for actuator dynamics in the control allocation solution, dynamic incremental nonlinear control allocation, or D-INCA. The incremental approach allows for the implementation of a first order discrete-time actuator dynamics model in the quadratic programming (QP) solver. This model is used to find the optimal command inputs in addition to the desired physical actuator deflections, hereby compensating for actuator dynamics delays. Whereas, the baseline incremental nonlinear control allocation (INCA) approach requires pseudo-control hedging of the outer loop reference to increase closed loop stability margins under actuator dynamics delays. To its advantage, D-INCA does not require feedback of higher order output derivatives than INCA and can be used with nonlinear non-control affine systems. Furthermore, with adaptive D-INCA, or AD-INCA, an actuator dynamics parameter estimator is introduced to adapt the actuator model online, minimizing actuator tracking errors after actuator failures. The proposed methods are applied to a fighter aircraft model with an over-actuated innovative control effectors suite and results are compared to the baseline INCA controller.

Climate change poses a serious threat to ecosystems and increases the need for accurate and rigorous monitoring of ecosystems. Current monitoring solutions are often bulky, expensive, and lack critical functionalities such as on-board inference capabilities, robust wireless connections, and a diverse sensor suite. Ecological monitoring projects often suffer from inefficiencies caused by the large time delays between collecting data and analyzing said data, as well as having to spend large amounts of time in the field setting up the sensors manually. This thesis addresses many of these issues by designing a sensor with an extensive sensor suite, robust wireless capabilities and an on-board audio classifier able to perform real-time inference. Furthermore, attention is paid to making the system extendable in the future and allow for potentially integrating the sensors with a drone delivery- and retrieval system. The system tests performed indicate that the system has great potential given more time to tweak some of its identified shortcomings.

We conduct a simulation study of an insect-inspired navigation method that combines visual learning in a small area around a home location with path integration to successfully navigate over distances 8 to 10 times larger than the learning radius, while only requiring 6MB of memory. Our simulations show that the method is capable of learning with on-board data only and handling the noise levels and errors it may encounter in real-world conditions. Furthermore, we investigate the integration of online learning with a structured buffer to mitigate catastrophic forgetting, allowing continuous learning during flight. Additional experiments assess the system's generalization capabilities beyond the learned area, its robustness to small variations in pitch and roll, and the relationship between network size and the learnable area size.

Autonomous drone racing has gained attention for its potential to push the boundaries of drone navigation technologies. While much of the existing research focuses on racing in obstacle-free environments, few studies have addressed the complexities of obstacle-aware racing, and approaches presented in these studies often suffer from overfitting, with learned policies generalizing poorly to new environments. This work addresses the challenge of developing a generalizable obstacle-aware drone racing policy using deep reinforcement learning. We propose applying domain randomization on racing tracks and obstacle configurations before every rollout, combined with parallel experience collection in randomized environments to achieve the goal. The proposed randomization strategy is shown to be effective through simulated experiments where drones reach speeds of up to 70 km/h, racing in unseen cluttered environments. This study serves as a stepping stone toward learning robust policies for obstacle-aware drone racing and general-purpose drone navigation in cluttered environments. Code is available at https://github.com/ErcBunny/IsaacGymEnvs.

Aerial physical interaction opens the door for many operations at height to be automatised using aerial robots. This research presents a novel manipulator design mounted on a traditional quadrotor, which utilises both mechanical and software compliance to perform physical interaction on vertical walls and overhanging surfaces, such as those found under bridges. A centralised impedance control scheme allows direct control of the end-effector pose without needing separate modes for free-flight and contact. A spring-loaded prismatic joint provides passive compliance while doubling as a force-feedback for the impedance controller through measuring the spring displacement. Simulation and flight experiments prove the feasibility and robustness of this approach for exchanging high forces at height, with a total of 44 successful experiments carried out in four sets. An average maximum force of 5.66 N or 19.3\% of the system's weight was achieved over one set of 11 experiments.

Self-supervised deep learning methods have leveraged stereo images for training monocular depth estimation. Although these methods show strong results on outdoor datasets such as KITTI, they do not match performance of supervised methods on indoor environments with camera rotation. Indoor, rotated scenes are common for less constrained applications and pose problems for two reasons: abundance of low texture regions and increased complexity of depth cues for images under rotation. In an effort to extend self-supervised learning to more generalised environments we propose two additions. First, we propose a novel Filled Disparity Loss term that corrects for ambiguity of image reconstruction error loss in textureless regions. Specifically, we interpolate disparity in untextured regions, using the estimated disparity from surrounding textured areas, and use L1 loss to correct the original estimation. Our experiments show that depth estimation is substantially improved on low-texture scenes, without any loss on textured scenes, when compared to Monodepth by Godard et al. Secondly, we show that training with an application's representative rotations, in both pitch and roll, is sufficient to significantly improve performance over the entire range of expected rotation. We demonstrate that depth estimation is successfully generalised as performance is not lost when evaluated on test sets with no camera rotation. Together these developments enable a broader use of self-supervised learning of monocular depth estimation for complex environments.