Skip to content
Publications

Talk · Adaptive Agents and Multi-Agents Systems · May 24, 2026

Geometric State Fusion for Autonomous Agents: A Comparative Analysis of Dual Quaternion Observer and Kalman Filters

Abstract

Autonomous agents operating in dynamic 3D environments require robust pose estimation that preserves the geometric structure of rigid-body motion. Traditional filtering approaches often decouple rotation and translation, leading to kinematic inconsistencies. This paper presents a comprehensive benchmark of three geometrically-aware architectures: a Dual Quaternion Geometric Observer (GeoDQ), a manifold-aware Unscented Kalman Filter (UKF-M), and an Error-State Kalman Filter (ESKF). We evaluate these on the complete RoNIN dataset (35 trajectories). Results demonstrate that the proposed GeoDQ method significantly outperforms filtering baselines, reducing RMSE by 3.7x compared to ESKF and 6.1x compared to UKF-M. Furthermore, robustness analysis reveals that the geometric observer maintains superior stability under sparse updates. Despite its mathematical rigor, GeoDQ is the most computationally efficient, executing 10% faster than ESKF and 5x faster than UKF-M, making it ideal for resource-constrained embedded agents.

Inertial Sensor and NavigationTarget Tracking and Data Fusion in Sensor NetworksRobotics and Sensor-Based Localizationpose estimationdual quaternionsunscented Kalman filter

More by Ilya Afanasyev