In this thesis, two problems are covered. The first one is to estimate the position of a drone out of two different given transformations, which don’t have the same coordinate origin. Furthermore, measurement errors have to be outbalanced so that the position estimation is as accurate as possible. In order to solve this problem, a Kalman Filter was utilized.
The second problem is to direct the drone to given goal positions while avoiding obstacles. For directing the drone to the desired location a vector flight control with temporary goals was created. The vector flight control is working online and is constantly using the current estimated position of the Kalman Filter in order to direct the drone correctly at each time. This thesis is covering the concept, the implementation and evaluation of these algorithms.
Table of Contents
1 Problem definition
2 Related work
3 Basics
3.1 Quadrocopter
3.2 Frames
3.3 Quaternions
3.4 Robot Operating System
3.5 Unified Robot Description Format
3.6 Kalman Filter
3.6.1 Extended Kalman Filter
3.6.2 Unscented Kalman Filter
3.7 Trajectory planning
3.8 Potential Field
4 Concept
4.1 Sensor Fusion of drone odometry and inspection software
4.2 Vector Flight Control
4.3 Architecture Overview
5 Implementation
5.1 Kalman Filter Implementation
5.1.1 Existing libraries
5.1.2 Simulation
5.1.3 Realization
5.2 Potential Field Method Implementation
5.2.1 Existing libraries
5.2.2 Realization
6 Proof of Concept
6.1 Sensor Fusion Evaluation
6.2 Vector Flight Control Evaluation
6.3 Evaluation of the combined architecture
7 Conclusion and Future Work
Objectives & Research Focus
This thesis addresses the automation of drone-based inspection of large structures, such as ship hulls, by developing a robust system for sensor fusion and trajectory planning. The primary research goal is to fuse odometry data from a drone with position data from an inspection software—using a Kalman Filter to handle measurement errors and drift—and to implement a vector flight control system that enables the drone to reach specific inspection points while autonomously avoiding static obstacles and escaping local minima traps.
- Sensor fusion techniques to reconcile drone odometry with external inspection software coordinates.
- Implementation of Kalman Filters (EKF/UKF concepts) to provide accurate, drift-compensated position estimation.
- Development of a vector flight control system utilizing the Potential Field Method for online trajectory generation.
- Integration of the Random Walk method to handle local minima traps during navigation.
Excerpt from the Book
3.8 Potential Field
An approach for path planning is the potential field method. Based on the configuration of the robot, artificial potential forces are computed. These forces determine the movement of the robot [Pet15]. Every point in the world is assigned to an artificial potential field, which is either an attractive or a repulsive field. The robot’s goal is to move to the lowest potential, the target position. The obstacles, which should be avoided by the robot, are marked as high potentials [Sid18]. In order to create a path, two planners are needed: a global and a local planner. The global one selects a path out of the lowest potentials and the local planner modifies the path so that every possible dynamic collision is avoided [HA+92].
Summary of Chapters
1 Problem definition: Outlines the challenge of automating large-scale structure inspections using drones and defines the core requirements for sensor fusion and trajectory planning.
2 Related work: Provides a literature review of existing autonomous inspection projects, sensor fusion approaches, and trajectory planning methods.
3 Basics: Establishes the theoretical foundation, covering quadrocopters, coordinate frames, Quaternions, ROS, Kalman Filters, and the Potential Field Method.
4 Concept: Details the proposed architecture, consisting of a sensor fusion component and a vector flight control system designed for autonomous navigation.
5 Implementation: Describes the technical realization of the Kalman Filter and Potential Field Method using Python and ROS within the RotorS simulation environment.
6 Proof of Concept: Evaluates the system performance through various simulations, including scenarios with sensor drift, jumps, and path navigation challenges.
7 Conclusion and Future Work: Summarizes the findings, confirms the effectiveness of the proposed algorithms, and suggests improvements for future hardware integration.
Keywords
Drone Inspection, Sensor Fusion, Kalman Filter, Vector Flight Control, Potential Field Method, Robot Operating System, ROS, Trajectory Planning, Autonomous Navigation, Quadrocopter, Localization, Path Planning, RotorS, Simulation, Obstacle Avoidance
Frequently Asked Questions
What is the core purpose of this thesis?
The thesis focuses on automating the inspection of large structures using drones by integrating sensor fusion to accurately determine the drone's position and developing a flight control system to navigate to inspection points safely.
Which sensors and software are used in this project?
The project uses drone odometry and simulated inspection software coordinates within the Robot Operating System (ROS) framework, utilizing the RotorS simulator for validation.
What is the primary research question?
How can data from disparate position estimation tools be fused to create a stable, error-tolerant position estimate, and how can a drone be directed to specific goal points in an environment with obstacles while avoiding local minima?
What scientific methods are utilized?
The work employs the Kalman Filter for robust sensor fusion (estimating position while filtering noise and drift) and the Potential Field Method combined with the Random Walk algorithm for collision-free trajectory planning.
What does the main body cover?
The main part of the thesis covers theoretical foundations, the conceptual architecture, the software implementation (using Python and ROS), and an extensive evaluation through simulation scenarios testing drift, signal jumps, and obstacle navigation.
What are the most significant keywords?
Key terms include Drone Inspection, Sensor Fusion, Kalman Filter, Potential Field Method, Trajectory Planning, and Autonomous Navigation.
How is the Kalman Filter adapted for this specific drone model?
The Kalman Filter is implemented as an 8-state predictor that accounts for 3D coordinates, rotation angles, and their respective velocities, with added logic to detect and reject sensor jumps from the simulated inspection software.
How does the drone escape local minima traps?
When the potential field forces cancel out, leaving the drone trapped, the system triggers a Random Walk method, moving the drone in a random direction for a short time to exit the trap before resuming standard trajectory planning.
- Quote paper
- Michelle Bettendorf (Author), 2019, UAV Inspection of Large Components. Adaptive Navigation at Runtime, Munich, GRIN Verlag, https://www.grin.com/document/1162994
