Physical Interaction with Micro Aerial Vehicles: Control and Design

Recent advances in robotics will enable them to do the dirty and dangerous jobs that expose humans to risk. An important area of robotics that gained particular attention recently is aerial robots or Micro Air Vehicles (MAVs). With their ability to freely navigate in 3D space, they can reach inaccessible areas to perform tasks such as inspection and surveillance. This flexibility comes at a cost: MAVs are agile and highly dynamic which poses challenges for control and state estimation. Until today, MAVs have been successfully used in remote sensing applications, where surrounding structures are considered as obstacles to avoid. Giving MAVs the ability to approach and interact with the environment will open doors for many applications that are currently done using scaffolds, rope access or cranes. Among these applications, contact-based inspection, pick and place, cleaning and painting. The main goal of this thesis is give MAVs the capability to perform tasks that require physical interaction with main focus on control and system design. Given the restrictions of MAVs in terms of payload, we focused on designing lightweight manipulation systems that can fit on lightweight MAVs. The first contribution of this thesis is focused on model-based control and a general Model Predictive Controller (MPC) control framework that enables precise trajectory tracking, fail-safe navigation and multi-MAVs reactive collision avoidance. In this thesis we also present the control scheme for an omnidirectional MAV, a hexacopter with tiltable arms that can exert forces and torques in all directions. The overactuated nature of the system poses many challenges regarding the control allocation that needs to be running at high rate on the flight controller microcontroller. A solution to this problem is presented with experimental validation of the system capabilities. Another contribution of this thesis is the design, implementation and validation of various algorithms and mechanical systems to perform tasks that require aerial manipulation. We present visual servoing algorithms that use RGB image data or depth information to accurately position the end-effector on target object. These algorithms have been demonstrated in two scenarios, pick and place of static and moving objects using MAV is demonstrated and tree cavity inspection. Additionally, the mechanical design and modeling of magnetic gripper is presented. The magnetic gripper features very low power consumption as it is based on an elecro-permanent magnet, and we demonstrated the capabilities of the gripper to pick and place static and moving objects. A more advanced actuated manipulator with parallel kinematics is designed and complete system model is derived and validated experimentally in inspection tasks. Finally, we present algorithm that focuses on physical interaction by means of multiple MAVs in communication denied environments. The approach is bio-inspired and focuses on robust interaction between multiple MAVs mechanically coupled. The proposed approach does not rely on global reference frame nor on low-latency communication between robots. Robustness analysis of the proposed method is conducted and we present a method to tune each MAV controller to guarantee stability and performance even in worst case scenario. The presented algorithms have been evaluated in real experiments in various conditions indoor and outdoor