In context
In 2025, simultaneous localization and mapping (SLAM) remained a cornerstone for autonomous robots in industrial inspection, mining, and service applications. However, complex environments such as post-earthquake buildings, coal mine tunnels, and urban utility corridors—characterized by low light and weak texture—posed persistent challenges to traditional visual SLAM, leading to loop-closure failures and degraded trajectory accuracy. This paper addressed that gap by fusing complementary SLAM paradigms.
What was reported
Researchers at Soochow University proposed a heterogeneous SLAM fusion method combining ORB-SLAM2, a feature-based visual SLAM known for high-precision mapping, with RatSLAM, a bionic SLAM inspired by hippocampal neurodynamics that excels in scene recognition under complex illumination. The system runs both SLAM threads in parallel, synchronized via keyframe alignment, and integrates their outputs using fuzzy neural networks (FNNs).
A standard FNN handles loop-closure decision-making when the two SLAM results conflict, evaluating point-cloud structure similarity, image color similarity, and geometric consistency to select the more reliable detection. A Takagi-Sugeno (T-S) FNN optimizes trajectory fusion by assessing the reliability of inter-frame transformations from each SLAM, based on relative reprojection error and relative pose error, then adaptively blending the trajectories.
Experiments on self-collected and public datasets showed that the proposed method achieved higher loop-closure recall and lower absolute trajectory error (ATE) compared to ORB-SLAM2 and RatSLAM individually, demonstrating robust performance in complex illumination scenarios.
Why it mattered
This work offered a practical pathway to enhance autonomous robot navigation in challenging industrial environments, where reliable localization is critical for tasks such as inspection and mapping. By intelligently fusing traditional and bionic SLAM, it addressed the limitations of fixed-weight multi-sensor fusion, potentially improving the robustness and accuracy of robotic systems in real-world low-light conditions.
“The proposed heterogeneous SLAM fusion method has higher recall rate of loop-closure detection and lower absolute trajectory error (ATE) on self-collected and public datasets, shows strong robustness in complex scenarios.”
Source: 《机器人》期刊 (robot.sia.cn) · Published 2025-08-19 · “面向复杂光照场景的异质SLAM融合方法”