Position estimation without GNSS
2024 (English)Independent thesis Basic level (professional degree), 10 credits / 15 HE credits
Student thesis
Abstract [en]
Localization is a commonly encountered task in a wide variety real-world applications. A well-established solution for tracking the position of land and aquatic vehicles, planes, robots, or even animals is satellite based navigation, such as GPS or other global navigation satellite systems (GNSS). A common problem is the need for a stable satellite connection. This connection can be disturbed by a blocked line of sight between the satellite and the receiver, or even a lack of suitable satellites. Accurate localization of tracked vehicles in mining quarries is a vital part of their autonomous and remote-controlled operation. To improve the robustness of these systems, this study investigates the feasibility of employing statistical state estimation techniques for precise position estimation in temporary GNSS-denied environments. By developing and evaluating multiple motion models tailored to tracked vehicle dynamics, including differential drive- and IMU kinematics, the study identifies an optimal model for accurate position estimation. Sensor fusion techniques, integrating track speed sensors and inertial measurement units, are explored to enhance state estimation and improve its accuracy. Through testing and analysis, robust sensor fusion configurations are identified, enabling reliable estimation even in GNSS-denied environments. Additionally, an evaluation system incorporating metrics like Root Mean Square Error (RMSE) and Normalized Estimation Error Squared (NEES) is utilized to assess estimation quality and reliability. Results demonstrate the feasibility of maintaining sub-meter accuracy for extended durations without GNSS reception, subject to operational constraints and desired accuracy thresholds. Insights gained from the analysis highlight factors influencing estimation quality, such as maneuver complexity at the point of GNSS loss, duration of GNSS loss, and sensor noise, particularly relevant in mining quarries. Track speed and gyroscope readings are found the be the most suitable sensor data, where an extended differential drive motion model incorporating both slippage and gyroscope readings proved the most precise in modeling the machines movement. These findings offer valuable guidance for future optimization efforts, aimed at enhancing GNSS-denied localization for tracked vehicles in challenging mining environments, thereby advancing operational efficiency.
Place, publisher, year, edition, pages
2024.
Keywords [en]
State estimation, Position estimation without GNSS, Motion models, Sensor fusion, Unscented Kalman filter, Differential drive kinematics
National Category
Computer Sciences
Identifiers
URN: urn:nbn:se:oru:diva-115233OAI: oai:DiVA.org:oru-115233DiVA, id: diva2:1887202
Subject / course
Computer Engineering
Supervisors
Examiners
2024-08-072024-08-072024-08-07Bibliographically approved