Reinforcement Learning-Based Vessel Route Optimization with Continuous Obstacle Avoidance by Using DDPG Algorithms


Creative Commons License

Nazlıgül Y. E., Yazır D., Hocek H.

Global Maritime Congress, İstanbul, Türkiye, 20 - 21 Mayıs 2024, ss.215-218

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Basıldığı Şehir: İstanbul
  • Basıldığı Ülke: Türkiye
  • Sayfa Sayıları: ss.215-218
  • Karadeniz Teknik Üniversitesi Adresli: Evet

Özet

This paper presents a novel approach to vessel route optimization by employing machine learning techniques, specifically the Deep Deterministic Policy Gradient (DDPG) algorithm. The research aims to address the challenge of continuous obstacle avoidance for vessels while optimizing their routes. The proposed methodology utilizes DDPG algorithms to develop an intelligent decision-making system that allows vessels to navigate efficiently while continuously avoiding obstacles. The study begins with the collection and pre-processing of data, including vessel characteristics, meteorological information, and obstacle positions. The preprocessed dataset is then used to train the DDPG model, enabling it to learn the optimal vessel behavior in different scenarios. To validate the proposed methodology, extensive simulations are conducted with varying obstacle positions and environmental conditions. The results demonstrate improved route efficiency and obstacle avoidance rates compared to traditional navigation systems, showcasing the effectiveness of the machine learning-based approach. Furthermore, the trained DDPG model exhibits enhanced generalization capabilities, effectively adapting to dynamic situations and unencountered obstacles. In conclusion, this paper presents a novel application of DDPG algorithms in vessel route optimization, introducing an intelligent decision-making system capable of continuously avoiding obstacles. The findings contribute to the advancement of autonomous vessel navigation, providing a more efficient and safe maritime transportation system. Future research avenues include optimizing the DDPG algorithm further, integrating realtime sensor data, and considering multi-vessel interactions in route planning.