Degeneration-Aware Multisensor Fusion for Robust Indoor SLAM via Texture–Geometry Coupling

his paper presents a degeneration-aware Simultaneous Localization and Mapping (SLAM) framework designed to enhance localization robustness and mapping accuracy in perceptually degenerate indoor environments. The proposed system fuses measurements from a ground texture camera, LiDAR, and an inertial measurement unit (IMU) to specifically combat the large-scale drift errors that LiDAR-based systems suffer in geometrically degenerate environments, such as long, featureless corridors. A unified degeneration descriptor is introduced by combining orientation consensus and directional entropy derived from LiDAR surface normals, enabling quantitative modeling of environmental degeneracy and adaptive residual weighting in sensor fusion. The framework employs a tightly coupled multi-modal registration strategy, incorporating <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">SE</i>(2)-constrained initialization based on texture features and subsequent <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">SE</i>(3) optimization using both geometric and visual residuals. Furthermore, a degeneracy-aware covariance damping mechanism is embedded within an Iterated Error State Kalman Filter (IESKF), ensuring consistent state estimation under poor observability. Extensive experiments on challenging indoor datasets validate the proposed method’s effectiveness, demonstrating an average reduction of 61.89% in End-to-End positioning error over the leading general-purpose fusion framework FAST-LIVO2 and confirming its superior robustness against geometric degeneration.