Skip to content
Publications

Talk · Conference of Open Innovations Association · April 28, 2026

Beyond Kalman Filtering: a Dual Quaternion Geometric Observer for Robust 6DoF Pose Estimation

Abstract

Autonomous robots and wearable devices operating in dynamic 3D environments require robust pose estimation that preserves the geometric structure of rigid-body motion. Traditional filtering approaches, such as Extended Kalman Filters (EKF), often decouple rotation and translation, leading to linearization artifacts and kinematic inconsistencies. This paper presents a comprehensive benchmark of a Dual Quaternion Geometric Observer (GeoDQ) against state-of-the-art manifoldaware filters (UKF-M) and standard error-state filters (ESKF). We introduce an optimized architecture featuring a Screw Linear Interpolation (SCLERP) fusion mechanism, augmented with a heuristic tilt correction, PID-based feedback loops, and Numbabased JIT compilation for real-time performance. Evaluated on 35 distinct trajectories from the RoNIN dataset under wearable smartphone configurations with unstable IMU orientations, the proposed GeoDQ (JIT) achieves a Root Mean Square Error (RMSE) of 0.0043 m. By interpolating directly on the$S E(3)$manifold, GeoDQ outperforms the standard ESKF (0.1409 m) and UKF-M (0.2312 m) by orders of magnitude. Furthermore, robustness analysis demonstrates that the geometric observer exhibits smooth error growth under severe measurement decimation, maintaining sub-meter accuracy where filter-based methods rapidly diverge. Despite its mathematical rigor, the JIToptimized GeoDQ executes in just 3.0 s per trajectory, making it significantly faster than standard Python-based filters and highly suitable for edge computing applications, wearable devices, and resource-constrained platforms.

Inertial Sensor and NavigationTarget Tracking and Data Fusion in Sensor NetworksRobotics and Sensor-Based Localization6DoF pose estimationdual quaternionsextended Kalman filter

More by Ilya Afanasyev