Unmanned vehicles have become indispensable tools in various fields such as agriculture, environmental monitoring, and disaster response. A crucial component of these platforms is sensor fusion. As measurements are inherently noisy, certain variables cannot be captured by a single sensor. Sensor fusion integrates multiple, often redundant, and thus at least partially controllable sensor measurements to deliver accurate estimates of position, attitude, and other relevant variables.

While adding new sensors to a platform can reduce the errors in these estimates, space and cost limitations often make this option impractical, particularly for unmanned vehicles. Therefore, alternative approaches must be explored. One such approach is the Vehicle Dynamic Model (VDM), which models the effects of control inputs and external factors (e.g., wind) on the vehicle's acceleration. By leveraging this relationship as an additional constraint in sensor fusion, the benefits of the model are obtained similarly as new additional sensor. Hence, this method requires no extra sensors, making it ideal for platforms constrained by space and budget. Previous studies have applied VDM-based navigation primarily to fixed-wings and delta-wings, using Kalman Filters. While these approaches have shown promise, they fail to capture certain long-term dependencies and are optimally suited mainly for real-time. In contrast, factor graph-based methods for sensor fusion offer a more versatile and robust framework, capable of simultaneously processing all available information. This makes them particularly advantageous for handling spatio-temporal constraints, accommodating long-term memory in stochastic models, and managing non-linear observation relationships.

This thesis extends prior work by applying VDM to factor graph-based sensor fusion for fixed-wing aircraft and pioneering its use with quadcopters and ground vehicles, such as cars, within the mentioned factor graph framework. This integration allows the exploitation of the unique strengths of factor graphs in VDM applications, especially as those related to precise mapping or sensing without the real-time constraint, particularly due to their ability to incorporate spatio-temporal constraints, such as camera observations. In a nut shell, this research explores the potential of VDM in various applications, such as wind sensing for unmanned aerial vehicles in challenging environments like mountainous terrain, optimizing platform costs for airborne laser scanning, and improving positioning accuracy in GNSS-denied scenarios, especially for unmanned ground vehicles operating or mapping in urban canyons or similar environments. Ultimately, this thesis advances cost-effective, reliable trajectory determination and sensing systems for unmanned aerial and ground vehicles, providing innovative solutions for environmental monitoring, mapping, and autonomous navigation.