Three-dimensional path planning is instrumental in path decision making and obstacle avoidance for deep-sea mining vehicles (DSMV). However, conventional particle swarm algorithms have been prone to trapping in local optima and have slow convergence rates when applied to underwater robot path planning. In order to secure a safe and economical three-dimensional path for the DSMV from the mining area to the storage base in connection with innovative mining system, this paper proposes a multi-objective optimization algorithm based on improved particle swarm optimization (IPSO) path planning. Firstly, we construct an unstructured seabed mining area terrain model with hazardous obstacles. Consequently, by considering optimization objectives such as the path length, terrain undulation, comprehensive energy consumption, and crawler slippage rate, we convert the path planning problem into a multi-objective optimization problem, constructing a multi-objective optimization mathematical model. Following that, we propose an IPSO algorithm to tackle the multi-objective non-linear optimization problem, which enables global optimization for DSMV path planning. Finally, we conduct a comprehensive set of experiments using the MATLAB simulation platform and compare the proposed method with existing advanced methods. Experimental results indicate that the path planned by the IPSO exhibits superior performance in terms of path length, terrain undulation, energy consumption, and safety.
Accurate and real-time environmental perception is essential for the safe and efficient execution of deep-sea mining operations. Semantic segmentation of forward-looking sonar (FLS) images plays a pivotal role in enabling environmental awareness for deep-sea mining vehicles (DSMVs), but remains challenging due to strong acoustic noise, blurred object boundaries, and long-range semantic dependencies. To address these issues, this study proposes DSM-Seg, a novel hybrid segmentation architecture combining Convolutional Neural Networks (CNNs) and Receptance Weighted Key-Value (RWKV) modeling. The architecture integrates a Physical Prior-Based Semantic Guidance Module (PSGM), which utilizes sonar-specific physical priors to produce high-confidence semantic guidance maps, thereby enhancing the delineation of target boundaries. In addition, a RWKV-Based Global Fusion with Semantic Constraints (RGFSC) module is introduced to suppress cross-regional interference in long-range dependency modeling and achieve the effective fusion of local and global semantic information. Extensive experiments on both a self-collected seabed terrain dataset and a public marine debris dataset demonstrate that DSM-Seg significantly improves segmentation accuracy under complex conditions while satisfying real-time performance requirements. These results highlight the potential of the proposed method to support intelligent environmental perception in DSMV applications.
With the increasing significance of deep-sea resource development, Forward-Looking Sonar (FLS) has become an essential technology for real-time environmental mapping and navigation in deep-sea mining vehicles (DSMV). However, FLS images often suffer from a limited field of view, uneven imaging, and complex noise sources, making single-frame images insufficient for providing continuous and complete environmental awareness. Existing mosaicking methods typically rely on external sensors or controlled laboratory conditions, often failing to account for the high levels of uncertainty and error inherent in real deep-sea environments. Consequently, their performance during sea trials tends to be unsatisfactory. To address these challenges, this study introduces a robust FLS image mosaicking framework that functions without additional sensor input. The framework explicitly models the noise characteristics of sonar images captured in deep-sea environments and integrates bidirectional cyclic consistency filtering with a soft-weighted feature refinement strategy during the feature-matching stage. For image fusion, a radial adaptive fusion algorithm with a protective frame is proposed to improve edge transitions and preserve structural consistency in the resulting panoramic image. The experimental results demonstrate that the proposed framework achieves high robustness and accuracy under real deep-sea conditions, effectively supporting DSMV tasks such as path planning, obstacle avoidance, and simultaneous localization and mapping (SLAM), thus enabling reliable perceptual capabilities for intelligent underwater operations.
Underwater robots require fast and accurate localization results during challenging near-bottom operations. However, commonly used methods such as acoustic baseline localization, dead reckoning, and sensor fusion have limited accuracy. The use of forward-looking sonar (FLS) images to observe the seabed environment for pose estimation has gained significant traction in recent years. This paper proposes a lightweight front-end FLS odometry to provide consistent and accurate localization for underwater robots. The proposed direct FLS odometry (DFLSO) includes several key innovations that realize the extraction of point clouds from FLS images and both image-to-image and image-to-map matching. First, an image processing method is designed to rapidly generate a 3-D point cloud of the seabed using FLS image, enabling pose estimation through point cloud matching. Second, a lightweight keyframe system is designed to construct point cloud submaps, which utilize historical information to enhance global pose consistency and reduce the accumulation of image-matching errors. The proposed odometry algorithm is validated by both simulation experiments and field data from sea trials.
This study investigates the dual-task control challenge of path following and obstacle avoidance for deep-sea mining robots operating in complex, unstructured environments. To address the limitations of traditional training strategies, we propose an optimized training framework that integrates environmental design enhancements and algorithmic advancements. Specifically, we develop a Dual-Task Training Environment by combining the Random Obstacle Environment with a newly proposed Obstructed Path Environment, ensuring a balanced learning approach. While agents trained solely in the Random Obstacle Environment exhibit unilateral obstacle avoidance strategies and achieve a 0% success rate in randomized obstacle scenarios, those trained in the Dual-Task Environment demonstrate 85.4% success under identical test conditions and acquire more complex bilateral avoidance strategies. Additionally, we introduce a Dynamic Multi-Step Update mechanism, which integrates immediate rewards with long-term returns to enhance deep reinforcement learning (Twin Delayed Deep Deterministic Policy Gradient, TD3) performance without increasing computational complexity. Under the optimal multi-step setting (n = 5), the Dynamic Multi-Step Update mechanism significantly improves path following accuracy, reducing trajectory deviations to 0.128 m on straight paths and 0.195 m on S-shaped paths, while achieving nearly 100% success in multi-directional obstacle avoidance tests. These improvements collectively enhance the adaptability, robustness, and operational performance of deep-sea mining robots, advancing intelligent control strategies for autonomous deep-sea exploration and resource extraction.