Calibration and estimation for aerial robots with application to additive manufacturing
File(s)
Author(s)
Choi, Christopher
Type
Thesis
Abstract
Additive Manufacturing (AM) has recently entered the construction industry, enabling structures to be built layer by layer. However, existing frameworks often rely on bulky static gantry systems that hinder worker safety and operational flexibility. Ground mobile robots provide an alternative but are constrained by height and maneuverability. This thesis introduces an untethered Aerial Additive Manufacturing (Aerial-AM) framework using autonomous aerial robots to overcome these challenges. A proof-of-concept cylinder structure was successfully printed, demonstrating the feasibility of drone-based large-scale construction.
Autonomous aerial navigation depends on precise state estimation and Simultaneous Localization and Mapping (SLAM), with Visual-Inertial (VI) sensors being a preferred choice for their affordability and compactness. However, accurate VI sensor calibration is crucial and often challenging. Traditional calibration tools focus on algorithmic refinement without addressing the need for specific sensor measurements. This thesis proposes an active sensing approach using Next-Best-View (NBV) for camera calibration and Next-Best-Trajectory (NBT) for camera-IMU calibration. Experiments show this method provides faster, more accurate, and consistent calibration, enhancing Visual-Inertial Odometry (VIO) and SLAM performance.
Additionally, active sensing addresses limitations in passive perception for state estimation and SLAM. Rigidly mounted cameras on aerial robots often experience motion blur during sudden maneuvers or fail in featureless environments, causing tracking errors. To mitigate these issues, a dual VI-sensor rig was developed, with one sensor rigidly mounted and the other stabilized by a 3-axis gimbal. The stabilized sensor actively reduces motion blur, providing smoother and more reliable measurements. Indoor experiments confirmed that gimbal-stabilized VIO achieves lower trajectory errors compared to body-fixed setups. These advancements in calibration and sensing techniques highlight the potential of active sensing in aerial robotics and additive manufacturing.
Autonomous aerial navigation depends on precise state estimation and Simultaneous Localization and Mapping (SLAM), with Visual-Inertial (VI) sensors being a preferred choice for their affordability and compactness. However, accurate VI sensor calibration is crucial and often challenging. Traditional calibration tools focus on algorithmic refinement without addressing the need for specific sensor measurements. This thesis proposes an active sensing approach using Next-Best-View (NBV) for camera calibration and Next-Best-Trajectory (NBT) for camera-IMU calibration. Experiments show this method provides faster, more accurate, and consistent calibration, enhancing Visual-Inertial Odometry (VIO) and SLAM performance.
Additionally, active sensing addresses limitations in passive perception for state estimation and SLAM. Rigidly mounted cameras on aerial robots often experience motion blur during sudden maneuvers or fail in featureless environments, causing tracking errors. To mitigate these issues, a dual VI-sensor rig was developed, with one sensor rigidly mounted and the other stabilized by a 3-axis gimbal. The stabilized sensor actively reduces motion blur, providing smoother and more reliable measurements. Indoor experiments confirmed that gimbal-stabilized VIO achieves lower trajectory errors compared to body-fixed setups. These advancements in calibration and sensing techniques highlight the potential of active sensing in aerial robotics and additive manufacturing.
Version
Open Access
Date Issued
2024-06-27
Date Awarded
01/01/2025
License URL
Advisor
Leutenegger, Stefan
Davison, Andrew
Publisher Department
Computing
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)