The growing demand for high-altitude, complex interaction tasks has driven the evolution of Uncrewed Aerial Vehicles (UAVs) from passive perception platforms to systems capable of active physical interaction. This shift has catalyzed the development of Uncrewed Aerial Manipulators (UAMs), which integrate aerial mobility with manipulation capabilities to perform contact-rich tasks such as inspection, repair, and object handling in dynamic environments. In this thesis, we explore how to advance toward general aerial manipulation: enabling a single system to robustly execute a wide range of physical interaction tasks across diverse and challenging scenarios.

To address this challenge, we propose a unified aerial manipulation framework that consists of three key components: (i) a hybrid motion-force control module for dynamic and compliant physical interaction, (ii) a contact-aware trajectory planning module for safe and feasible motion generation under contact constraints, and (iii) an end-effector-centric control interface that separates platform-specific low-level control from high-level decision-making and policy learning. This modular framework aims to enhance system versatility and generality across different manipulation tasks and hardware platforms.

First, we develop a hybrid motion-force controller specifically tailored for aerial manipulation tasks. By explicitly regulating both motion and contact force during flight, this controller enables UAVs to interact compliantly and robustly with their environment, handling disturbances and uncertainties inherent in aerial interaction. This capability provides the foundation for executing complex tasks such as surface contact inspection and manipulation without compromising flight stability.

Second, we extend the scope of aerial manipulation beyond static contacts to dynamic surface interaction. We propose a contact-aware trajectory planner that generates motion plans accounting for varying contact forces and surface motions. Coupled with our hybrid controller, this allows the aerial manipulator to track both force and motion trajectories simultaneously, enabling dynamic tasks such as drawing, sliding, and surface exploration during flight.

Third, we design an end-effector-centric control interface that decouples low-level whole-body control from high-level task planning and policy learning. By focusing control objectives directly at the end-effector, our framework abstracts away platform-specific dynamics, allowing intuitive human teleoperation and facilitating the development of learning-based high-level policies. This design supports cross-task and cross-platform generalization, paving the way for more scalable and modular aerial manipulation systems.

Through real-world experiments across a variety of tasks, we demonstrate that our unified framework improves the precision, robustness, and generality of aerial manipulation capabilities. We believe this work takes a significant step toward advancing autonomous aerial robots as capable and reliable physical agents in complex real-world environments.