Graph-Optimized Encoder–IMU Fusion for Robust Pipeline Robot Localization in Confined Spaces

Abstract

Localization in confined spaces presents significant challenges, as conventional vision-based and LiDAR-based methods often exhibit limited performance due to environmental constraints. These limitations underscore the urgent need for enhanced inertial navigation systems with improved accuracy. To address the persistent issue of noise interference in the traditional inertial localization, this study introduces an enhanced encoder-inertial measurement unit (IMU) framework, specifically designed to provide a cost-effective localization solution for short-to-medium range tasks in enclosed environments. The proposed architecture adopts a dual-component design: 1) a front-end module that integrates data from the wheel encoder and IMU to estimate the robot pose, leveraging an error-state Kalman filter (ESKF) and 2) a back-end module that initializes the IMU data through graph optimization and performs large-scale local optimization of historical poses and inertial parameters. Finally, extensive experimental evaluations demonstrate the effectiveness of the proposed method.