← Back ICRA 2026

LST-SLAM: A Stereo Thermal SLAM System for Kilometer-Scale Dynamic Environments

Zeyu Jiang, Kuan Xu, Changhao Chen

PDF

AI summary

Key figure (auto-extracted from paper)

LST-SLAM enables robust, drift-free localization and mapping in large-scale, dynamic, and low-light outdoor environments using a novel stereo thermal SLAM pipeline.

Thermal SLAM Stereo Vision Dynamic Environments Self-Supervised Learning Loop Closure Autonomous Navigation

Problem

Existing thermal SLAM systems suffer from unreliable feature extraction, unstable motion tracking, and accumulated drift in large-scale, dynamic, and illumination-challenging outdoor environments.

Approach

The system integrates a self-supervised thermal feature network, stereo dual-level motion tracking, semantic-geometric dynamic filtering, and an incremental online bag-of-words model for loop closure and global optimization.

Key results

Self-supervised Thermal Point network adapting RGB priors to thermal imagery
Stereo dual-level tracking optimizing photometric and descriptor errors
75.8% and 66.8% lower localization errors than AirSLAM and DROID-SLAM
Incremental online bag-of-words for drift-free thermal loop closure

Why it matters

Enables reliable autonomous navigation and mapping for robots operating in extreme lighting, weather, and dynamic conditions where traditional RGB cameras fail.

Abstract

Thermal cameras offer strong potential for robot perception under challenging illumination and weather condi- tions. However, thermal Simultaneous Localization and Map- ping (SLAM) remains difficult due to unreliable feature extrac- tion, unstable motion tracking, and inconsistent global pose and map construction, particularly in dynamic large-scale outdoor environments. To address these challenges, we propose LST- SLAM, a novel large-scale stereo thermal SLAM system that achieves robust performance in complex, dynamic scenes. Our approach combines self-supervised thermal feature learning, stereo dual-level motion tracking, and geometric pose optimiza- tion. We also introduce a semantic–geometric hybrid constraint that suppresses potentially dynamic features lacking strong inter-frame geometric consistency. Furthermore, we develop an online incremental bag-of-words model for loop closure detection, coupled with global pose optimization to mitigate accumulated drift. Extensive experiments on kilometer-scale dynamic thermal datasets show that LST-SLAM significantly outperforms recent representative SLAM systems, including AirSLAM and DROID-SLAM, in both robustness and accuracy.

Index terms

SLAM Localization