Do-Hyeok Ahn, Myung-Il Roh, In-Chang Yeo, Do-Hyun Chun, "A Method for Automatic Control of the Block Lifting by an Offshore Floating Crane Based on Deep Reinforcement Learning", Proceedings of ISOPE(International Society of Offshore and Polar Engineers) 2025, Goyang, Korea, 2025.06.01-06
International Conference
2025.02.14 10:40
Do-Hyeok Ahn, Myung-Il Roh, In-Chang Yeo, Do-Hyun Chun, "A Method for Automatic Control of the Block Lifting by an Offshore Floating Crane Based on Deep Reinforcement Learning", Proceedings of ISOPE 2025, Goyang, Korea, 2025.06.01-06
조회 수 705
| 첨부 '1' |
|---|
| Abstract | An offshore floating crane is widely used in shipyards for lifting massive blocks. This crane consists of several components, including crane booms on the barge, block loaders (or equalizers), and wire ropes, which work together to lift blocks. However, due to the complex mechanical interactions between these components, achieving precise control over the movement of blocks has many challenges. Furthermore, since an offshore floating crane operates on the sea, it is highly susceptible to external environmental conditions such as winds, waves, and currents. These factors can induce swaying or cause unexpected movements of the blocks, leading to delays in work, reduced precision, and even accidents. Traditional control methods, such as the Proportional-Integral-Derivative (PID) control and the Sliding Mode Control (SMC), have limitations in achieving precise control in such highly dynamic and unpredictable environments. In this study, we applied the Deep Reinforcement Learning (DRL) to achieve precise control over the movement of the wire ropes, ensuring stability during the block lifting of the crane. By adopting a Hierarchical Reinforcement Learning (HRL) approach, we assigned distinct roles to two policies, each receiving different rewards. The high-level policy oversees the global control of the hoisting speed of wire ropes, while the low-level policy focuses on stabilizing the blocks and alleviating their swaying. This allows the DRL network to effectively learn how to safely lift the blocks to their target positions while minimizing undesired motion despite the external environmental forces. The proposed method successfully performed the block lifting tasks even in the presence of external disturbances in a simulation environment. We also demonstrated that the HRL approach in the crane control not only meets various control objectives in more complex scenarios compared to using traditional control methods or single-policy reinforcement learning, but also results in faster learning speeds and greater adaptability to changing conditions. |
|---|---|
| Publication Date | 2025-06-02 |
