Gottschalg, Grischa (2022)
Data Fusion Architecture with Integrity Monitoring for State Estimation in Automated Driving.
Technische Universität Darmstadt
doi: 10.26083/tuprints-00021764
Ph.D. Thesis, Primary publication, Publisher's Version
Text
Grischa_Gottschalg_Diss_final.pdf Copyright Information: CC BY-NC 4.0 International - Creative Commons, Attribution NonCommercial. Download (16MB) |
Item Type: | Ph.D. Thesis | ||||
---|---|---|---|---|---|
Type of entry: | Primary publication | ||||
Title: | Data Fusion Architecture with Integrity Monitoring for State Estimation in Automated Driving | ||||
Language: | English | ||||
Referees: | Becker, Prof. Dr. Matthias ; Winner, Prof. Dr. Hermann ; Eichhorn, Prof. Dr. Andreas | ||||
Date: | 2022 | ||||
Place of Publication: | Darmstadt | ||||
Collation: | xxix, 174 Seiten | ||||
Date of oral examination: | 19 July 2022 | ||||
DOI: | 10.26083/tuprints-00021764 | ||||
Abstract: | In the advent of automated driving, numerous system architectures to reach this goal are currently being developed. In this process, safety and modularity criteria are becoming increasingly important. This opens up new opportunities in the system design, but also creates new challenges. All functions or services in an automated vehicle are affected by this, including the estimation of the vehicle's dynamic state. Requirements for the Vehicle Dynamic State Estimation demand integrity measures and high system reliability, which can often only be achieved by redundant structures. A federated data fusion architecture with integrity monitoring is presented in this work to fulfill the aforementioned requirements for the Vehicle Dynamic State Estimation in automated driving. It inputs several redundant multi-sensor data fusion filters in a first fusion layer whose results are combined in a second fusion layer. This second layer implements plausibility checks and combines voting and data fusion, both utilizing the integrity measures of the redundant data fusion filters in the first layer. To compute such integrity measures, three integrity monitoring concepts are developed which differ substantially in their error modeling and complexity. The first concept represents the traditional approach and relies on the fusion filter's estimated covariances assuming that the error is normally distributed. The second concept models the errors in the fusion filter's output as the sum of errors from the filter's inputs. The error of each input is modeled as a multi-variate Student distribution and propagated through the filter. Utilizing the resulting error distribution, the integrity measures are computed. The third concept is based on the principle of Multiple Hypothesis Solution Separation. Subsets consisting of parts of the available filter inputs are formed and the fusion for each of these subsets is computed. Integrity measures are deducted by comparing the subset's results with each other. Each of the three concepts outputs a protection level, which is compared to an alert limit for integrity monitoring. Additionally, one of the redundant multi-sensor data fusion filters used in the federated fusion architecture is presented in this work. An Error-State Extended Kalman Filter is implemented to fuse the observations of three sensor types: a dual-antenna Global Navigation Satellite System Receiver, an Inertial Measurement Unit and wheel odometry sensors measuring wheel speeds and steering angles. The implementation includes sensor error models providing suitable error covariances for the filter's measurement updates and Fault Detection & Exclusion methods to increase the filter's robustness. Besides the measurement updates from the mentioned sensor types, also zero updates, i.e., zero velocity and zero angular rate updates, are part of the implementation, which are executed when a standstill of the vehicle is detected. In order to evaluate the performance of the presented fusion filter, integrity algorithms and fusion architecture, an extensive set of measurements with a total duration of more than 23 hours is used. The measurements are divided into four categories according to their environmental and satellite reception conditions, since these have a strong influence on the filter's performance. All in all, the integrity requirements are met by all implemented algorithms in favorable satellite reception conditions. However, only the second concept using multi-variate Student distributions for error modeling does so in all measurement categories, from ideal satellite reception conditions in the test track category to very challenging environments in the urban category. The federated data fusion architecture also fulfills the integrity requirements with only one minor exception, maintaining or even reducing the empirical integrity risk depending on the measurement category. Additionally, the accuracy and availability is improved significantly in most cases, compared to the fusion filters in its first layer. Finally, the usability of the developed concepts is demonstrated by applying them in a prototype vehicle of the research project UNICARagil. The integration in the system architecture and the interaction with other services is presented. In measurements from the commissioning tests the functionality is illustrated. |
||||
Alternative Abstract: |
|
||||
Status: | Publisher's Version | ||||
URN: | urn:nbn:de:tuda-tuprints-217646 | ||||
Classification DDC: | 500 Science and mathematics > 550 Earth sciences and geology 600 Technology, medicine, applied sciences > 620 Engineering and machine engineering |
||||
Divisions: | 13 Department of Civil and Environmental Engineering Sciences > Institute of Geodesy 13 Department of Civil and Environmental Engineering Sciences > Institute of Geodesy > Physical and Satellite Geodesy |
||||
TU-Projects: | Bund/BMBF|16EMO0286|UNICARagil | ||||
Date Deposited: | 30 Aug 2022 08:35 | ||||
Last Modified: | 01 Sep 2022 11:00 | ||||
URI: | https://tuprints.ulb.tu-darmstadt.de/id/eprint/21764 | ||||
PPN: | 498738388 | ||||
Export: |
View Item |