Position estimation for autonomous navigation
Location
Ada, Ohio
Start Date
9-12-2025 12:50 PM
End Date
9-12-2025 1:00 PM
Description
This research compares single versus dual Extended Kalman Filter (EKF) approaches for vehicle localization, examining the impact of IMU sensor configuration and feedback between local and global filter stages in ROS2 navigation systems. Single EKF implementations fuse odometry, IMU, and GPS data in one filter, exhibiting accumulating drift over time. Dual EKF architectures employ separate local and global filters (odom and map frames in ROS2) with a feedback mechanism through local-versus-global displacement calculations, enabling cross- validation and error correction between stages. Four configurations were evaluated: (1) single EKF with one IMU, (2) dual EKF with one IMU, (3) single EKF with two IMUs on opposite vehicle sides in opposite orientations, and (4) dual EKF with two IMUs. Results show dual EKF architectures reduce position error by approximately 40% compared to single implementations, with feedback effectively lowering accumulation of position estimation drift from ground truth. The addition of a second IMU produces divergent outcomes: single EKF systems struggle with compounding error from opposing IMUs, increasing drift, while dual EKF systems can leverage IMU redundancy through opposing feedback, achieving superior accuracy. The local-versus-global displacement calculation enables this setup to distinguish sensor noise from actual motion. These findings directly apply to ROS2's robot localization package, where dual EKF naturally aligns with the transform tree architecture. While requiring increased computational overhead and careful parameter tuning, the accuracy improvements justify this complexity for applications requiring precise long-term localization in GPS-denied environments. Future work should investigate adaptive tuning and integration with additional sensor types and configurations.
Recommended Citation
Spinner, Elijah, "Position estimation for autonomous navigation" (2025). College of Engineering Student Research Colloquium. 5.
https://digitalcommons.onu.edu/eng_student_research_colloquium/2025/Presentations/5
Position estimation for autonomous navigation
Ada, Ohio
This research compares single versus dual Extended Kalman Filter (EKF) approaches for vehicle localization, examining the impact of IMU sensor configuration and feedback between local and global filter stages in ROS2 navigation systems. Single EKF implementations fuse odometry, IMU, and GPS data in one filter, exhibiting accumulating drift over time. Dual EKF architectures employ separate local and global filters (odom and map frames in ROS2) with a feedback mechanism through local-versus-global displacement calculations, enabling cross- validation and error correction between stages. Four configurations were evaluated: (1) single EKF with one IMU, (2) dual EKF with one IMU, (3) single EKF with two IMUs on opposite vehicle sides in opposite orientations, and (4) dual EKF with two IMUs. Results show dual EKF architectures reduce position error by approximately 40% compared to single implementations, with feedback effectively lowering accumulation of position estimation drift from ground truth. The addition of a second IMU produces divergent outcomes: single EKF systems struggle with compounding error from opposing IMUs, increasing drift, while dual EKF systems can leverage IMU redundancy through opposing feedback, achieving superior accuracy. The local-versus-global displacement calculation enables this setup to distinguish sensor noise from actual motion. These findings directly apply to ROS2's robot localization package, where dual EKF naturally aligns with the transform tree architecture. While requiring increased computational overhead and careful parameter tuning, the accuracy improvements justify this complexity for applications requiring precise long-term localization in GPS-denied environments. Future work should investigate adaptive tuning and integration with additional sensor types and configurations.