Acoustic–Visual Sensor Fusion for Robot Localization in Buried Pipes

Autonomous, untethered robots require robust localization algorithms to perform tasks such as inspection and monitoring in buried pipes. This task is challenging due to environmental constraints, such as the absence of a global positioning system underground and limited visual perception. Vision sensing can provide effective motion estimation from visual odometry; however, its position accuracy degrades over time due to drift. Conversely, acoustic echo sensing can detect distant features such as manholes, providing a position estimate that does not accumulate error. The complementary nature of these sensing methods suggests that multisensor data fusion would be a more robust approach for localization. Therefore, this paper presents a novel acoustic–visual sensor data fusion algorithm for robot localization in buried pipes. The method operates directly on acoustic measurements, bypassing the need for feature extraction and data association, and integrates with a direct optic‐flow‐based approach for visual odometry. Trajectory estimation is performed using both pose‐graph optimization and particle filtering. We demonstrate the approach using experimental data collected from a robot platform in a real‐world buried sewer pipe environment. Our results show that sensor data fusion outperforms each sensor used separately, and the approach consistently localizes the robot with a final estimate error of less than 0.5 m, a critical requirement for accurate defect repair.