An Underwater, Fault-Tolerant, Laser-Aided Robotic Multi-Modal Dense SLAM System for Continuous Underwater In-Situ Observation

2025-04-30

AI Summary

To address SLAM tracking failures and sparse mapping caused by underwater texture scarcity and geometric degradation, this paper proposes a fault-tolerant, LiDAR-enhanced multimodal dense SLAM system. Methodologically, it introduces three key innovations: (1) the first underwater binocular structured-light module (UBSL); (2) a fault-resilient tri-subsystem architecture integrating DP-INS, Water-UBSL, and Water-Stereo; and (3) an asynchronous multimodal factor graph backend that tightly fuses ESKF-based navigation (DVL, pressure sensor, IMU), UBSL–INS joint estimation, and stereo vision–aided initialization. Experimental results demonstrate a trajectory RMSE of only 0.039 m, 100% tracking continuity under partial sensor failure, and dense mapping density of 6922.4 points/m³—approximately tenfold higher than state-of-the-art methods—significantly enhancing in-situ continuous observation capability.

Technology Category

• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Computer Vision: Multi-modal Vision
• Machine Learning: Multimodal Learning

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Existing underwater SLAM systems are difficult to work effectively in texture-sparse and geometrically degraded underwater environments, resulting in intermittent tracking and sparse mapping. Therefore, we present Water-DSLAM, a novel laser-aided multi-sensor fusion system that can achieve uninterrupted, fault-tolerant dense SLAM capable of continuous in-situ observation in diverse complex underwater scenarios through three key innovations: Firstly, we develop Water-Scanner, a multi-sensor fusion robotic platform featuring a self-designed Underwater Binocular Structured Light (UBSL) module that enables high-precision 3D perception. Secondly, we propose a fault-tolerant triple-subsystem architecture combining: 1) DP-INS (DVL- and Pressure-aided Inertial Navigation System): fusing inertial measurement unit, doppler velocity log, and pressure sensor based Error-State Kalman Filter (ESKF) to provide high-frequency absolute odometry 2) Water-UBSL: a novel Iterated ESKF (IESKF)-based tight coupling between UBSL and DP-INS to mitigate UBSL's degeneration issues 3) Water-Stereo: a fusion of DP-INS and stereo camera for accurate initialization and tracking. Thirdly, we introduce a multi-modal factor graph back-end that dynamically fuses heterogeneous sensor data. The proposed multi-sensor factor graph maintenance strategy efficiently addresses issues caused by asynchronous sensor frequencies and partial data loss. Experimental results demonstrate Water-DSLAM achieves superior robustness (0.039 m trajectory RMSE and 100% continuity ratio during partial sensor dropout) and dense mapping (6922.4 points/m^3 in 750 m^3 water volume, approximately 10 times denser than existing methods) in various challenging environments, including pools, dark underwater scenes, 16-meter-deep sinkholes, and field rivers. Our project is available at https://water-scanner.github.io/.

Problem

• Achieve uninterrupted dense SLAM in complex underwater environments
• Overcome texture-sparse and geometrically degraded underwater conditions
• Integrate multi-sensor fusion for robust continuous in-situ observation

Innovation

• Laser-aided multi-sensor fusion for dense SLAM
• Fault-tolerant triple-subsystem architecture with ESKF
• Multi-modal factor graph back-end for dynamic fusion