Synchronous Observer Design for Landmark-Inertial SLAM with Magnetometer and Intermittent GNSS Measurements

A team of researchers including Arkadeep Saha has published a paper on arXiv introducing a novel 'Synchronous Observer' designed to solve a fundamental problem in Landmark-Inertial Simultaneous Localization and Mapping (LI-SLAM). Traditional LI-SLAM, which uses camera-detected landmarks and an Inertial Measurement Unit (IMU), cannot observe the robot's absolute position in the world, its yaw (heading) angle, or the landmark positions in a global frame. This leads to drift and ambiguity. The new observer architecture directly addresses this by incorporating two additional, commonly available sensor streams: intermittent Global Navigation Satellite System (GNSS) position fixes and continuous magnetometer (compass) measurements.

The proposed nonlinear observer fuses these data sources to achieve full-state estimation. Crucially, the team provides a rigorous mathematical proof showing the observer's error dynamics are both almost-globally asymptotically stable and locally exponentially stable. This means the estimated robot pose and map converge to the true values from almost any starting point and do so robustly. The method was validated in simulations, demonstrating its practical viability. The work is submitted for presentation at the 2026 Conference on Decision and Control (CDC), indicating its significance in the control systems community.

This advancement is particularly relevant for robots operating in environments where GPS signals are unreliable or blocked for periods (like urban canyons or indoors near windows), but occasional satellite fixes are available. By formally guaranteeing stability with intermittent aiding data, the observer provides a more reliable and accurate foundation for long-term autonomy in hybrid indoor-outdoor scenarios, moving beyond the limitations of purely vision-and-inertial-based systems.