Dynamic navigation: integrating GL-STGCNN and MPC for collision avoidance with future awareness

Weiqiang Liao, Yuegao Wu, Peilin Zhou, Haibin Wang, Wanneng Yu, Changkun Zhang, Chenghan Luo

Research output: Contribution to journalArticlepeer-review

2 Downloads (Pure)

Abstract

Existing ship dynamic collision avoidance methods mostly rely on the instantaneous motion information of surrounding ships to make decisions. This makes it difficult to adapt to changes in the motion states of surrounding ships, which may lead to collisions between ships. To improve the safety of dynamic collision avoidance methods, this paper combines the multi-ship trajectory prediction model GL-STGCNN with model predictive control for ship dynamic collision avoidance tasks. Firstly, the interaction between ships is extracted through GL-STGCNN to predict the future trajectories of surrounding ships. Then, the objective function based on the artificial potential field method and the velocity obstacle method is optimized to control the ship to complete the dynamic collision avoidance task. The performance of the dynamic collision avoidance method is verified and analyzed in the ship navigation scenario simulated by AIS data. The experiments show that the new ship dynamic collision avoidance method not only complies with the COLREGs, but also can flexibly select the collision avoidance method according to different scenarios. In addition, the theoretical collision avoidance threshold distance based on the MPC objective function shows a high degree of fit with the actual collision avoidance trigger distance observed in the simulation verification.
Original languageEnglish
Article number118416
Number of pages18
JournalOcean Engineering
Volume309
Issue numberPt. 1
Early online date14 Jun 2024
DOIs
Publication statusE-pub ahead of print - 14 Jun 2024

Keywords

  • trajectory prediction
  • model predictive control
  • dynamic collision avoidance
  • multi-ship interaction

Fingerprint

Dive into the research topics of 'Dynamic navigation: integrating GL-STGCNN and MPC for collision avoidance with future awareness'. Together they form a unique fingerprint.

Cite this