Abstract
Deep-sea mining aims to extract mineral resources from the ocean seabed, necessitating advanced vehicles for efficient operations. These vehicles, essential for exploiting the vast underwater resources, require sophisticated navigation. The primary challenge in deep-sea navigation is the absence of satisfactory algorithms that are capable of handling the complexity and unpredictability in deep-sea environment. This research deploys advanced deep reinforcement learning algorithms, to enable dynamic control in the deep-sea navigation - which was previously challenging when using conventional methods. These algorithms with detailed optimization of the hyperparameters have been implemented on a four-track deep-sea mining vehicle, demonstrating good performance in dynamic avoidance of obstacles that are randomly deployed.
| Original language | English |
|---|---|
| Article number | 117199 |
| Journal | Ocean Engineering |
| Volume | 298 |
| Early online date | 29 Feb 2024 |
| DOIs | |
| Publication status | Published - 15 Apr 2024 |
Funding
This study was supported by the State Key laboratory of Ocean Engineering , the Yazhou Bay Institute of Deepsea SCI-TECH and the Institute of Marine Equipment . This work was also supported by the Major Projects of Strategic Emerging Industries in Shanghai, the Fundamental Research Funds for the Central Universities and China Scholarship Council . The authors are grateful for the financial support.
Keywords
- artificial intelligence
- deep reinforcement learning
- deep-sea mining
- dynamic obstacle avoidance
- hyperparameter optimization