A GM-PHD-SLAM algorithm based on pose and map alternating update

Simultaneous localization and mapping (SLAM) is widely used in autonomous driving, augmented reality, and embodied intelligence. In real-world settings, sensor measurements often suffer from substantial clutter (false alarms) and missed detections, which complicate SLAM data association. This complexity manifests as uncertainty in associating observations to landmarks, the possibility of erroneous associations between clutter and landmarks, and the potential absence of landmark observations. Random Finite Set (RFS) theory offers a Bayesian estimation framework well suited to SLAM with uncertain data association and an unknown, time-varying number of landmarks, and has spurred extensive research on RFS-based SLAM methods. Particle-filter-based Probability Hypothesis Density (PHD)-SLAM can effectively estimate the joint probability density of the pose and the map under clutter and missed detections, yielding robust SLAM performance. However, improving the estimation accuracy of particle-filter PHD-SLAM typically requires increasing the number of particles, which rapidly scales the computational cost.

In a recent article featured in the Chinese Journal of Aeronautics (Volume 38, Issue 11, November 2025), Dr. Han Shen-tu and Mr. Zhiyuan Bai from Hangzhou Dianzi University, together with collaborators, propose Pose and Map Alternating Update (PMAU)-SLAM. The method uses Gaussian mixtures (GM) to alternately estimate the posterior pose marginal and the map’s probability hypothesis density. For pose estimation, it directly applies the Unscented Kalman Filter (UKF) to handle nonlinear motion and measurement models, and uses covariance intersection (CI) fusion to integrate multiple hypotheses under multi-hypothesis association. For map estimation, it employs a GM-PHD filter for the posterior update, augments the map with label management, and leverages landmark re-identification to mitigate the long-term accumulation of estimation errors.

The paper reports comparative experiments across multiple simulation scenarios as well as on the Victoria Park dataset—with systematic variations in measurement noise, clutter intensity, and detection probability. Results show that the PMAU-SLAM family achieves localization and mapping accuracy comparable to, or better than, particle-intensive PHD-SLAM, while maintaining computational cost close to lightweight particle configurations. The labeled variant is particularly effective at heading estimation and suppressing cumulative drift. Sensitivity analyses further indicate robustness to changes in detection probability and clutter intensity.

The authors argue that PMAU-SLAM strikes a favorable balance among accuracy, robustness, and computational efficiency in perception environments rife with clutter and missed detections, supporting applications in autonomous driving, virtual/augmented reality, and embodied intelligence. Future work includes coupling PMAU-SLAM with more sophisticated RFS filters, optimizing label-management for long-term operation in large-scale environments, and extending the approach across diverse sensor platforms.

Original Source

Han SHEN-TU, Zhiyuan BAI, Yun CHEN, Yizhen WEI, Jiaxin ZHAO, Zheng CAO. A GM-PHD-SLAM algorithm based on pose and map alternating update [J]. Chinese Journal of Aeronautics, 2025, https://doi.org/10.1016/j.cja.2025.103739.

About Chinese Journal of Aeronautics 

Chinese Journal of Aeronautics (CJA) is an open access, peer-reviewed international journal covering all aspects of aerospace engineering, monthly published by Elsevier. The Journal reports the scientific and technological achievements and frontiers in aeronautic engineering and astronautic engineering, in both theory and practice. CJA is indexed in SCI (IF = 5.7, Q1), EI, IAA, AJ, CSA, Scopus.