조로만, 노명일, "강화 학습 기반의 다 선박 충돌 회피 알고리즘 연구", 2019년도 대한조선학회 춘계학술발표회, 제주, pp. 186, 2019.05.15-17
Abstract | Developing a collision avoidance system which can operate in an unpredictable environment is a challenging task. Especially in congested sea areas, each ship should continuously make decisions to avoid collisions with other ships while complying with the Convention on the International Regulations for Preventing Collisions at Sea (COLREGs). In this study, we proposed a robust and efficient method to collision avoidance for multi-ships based on the Deep Reinforcement Learning (DRL). The proposed method directly maps the states of the encountered target ships to an own ship’s steering commands related to rudder angle using the Deep Neural Network (DNN). This DNN is trained over the multi-ships on various situations using the policy gradient based DRL algorithm. However, the typical feedforward DNNs used in the domain require a fixed-dimensional input. To handle the multi-ship collision avoidance problem, we category the encountered target ships into four regions in terms of COLREGs for the purpose of fixing the input size. We performed a variety of simulation scenarios to verify the effectiveness and efficiency of the proposed method. The simulation results show that the proposed method has the capabilities to guarantee the collision avoidance of multi-ships while ensuring following their predefined paths. |
---|---|
Publication Date | 2019-05-16 |
조로만, 노명일, "강화 학습 기반의 다 선박 충돌 회피 알고리즘 연구", 2019년도 대한조선학회 춘계학술발표회, 제주, pp. 186, 2019.05.15-17