Time-optimal obstacle avoidance of autonomous ship based on nonlinear model predictive control

Ming Zhang, Shuai Hao, Defeng Wu, Ming-Lu Chen, Zhi-Ming Yuan

Research output: Contribution to journalArticlepeer-review

9 Citations (Scopus)
19 Downloads (Pure)


Autonomous shipping has been identified as the way forward in the maritime transport. However, the time-optimal path planning, anti-disturbance trajectory tracking and obstacle avoidance are still ongoing challenging problems, which have not been properly addressed for autonomous ship. To fill the knowledge gap, we propose a novel nonlinear model predictive control (MPC), which integrates the time-optimal path planning, anti-disturbance tracking and obstacle avoidance. The proposed controller is designed as a 2-level hierarchical controller. In the high level, a planned path considering time minimum and obstacle avoidance is generated by nonlinear MPC in spatial formulation. A spatial reformulation is adopted to express manoeuvring time as a function mathematically. In the spatial coordinate, the manoeuvring time is minimised by MPC to generate a reference path for tracking. In the low level, vessel tracks the time-optimal planning trajectory by nonlinear MPC with extended Kalman filter in temporal formulation. The deviation of the tracking path in longitudinal direction is 15% ship length and the deviation in width direction is 1% under disturbances. The obstacle avoidance is implemented by using the proposed control method, and the tolerance of obstacle avoidance is 2L (ship length) to meet the safety requirement.

Original languageEnglish
Article number112591
Number of pages16
JournalOcean Engineering
Issue numberPart 1
Early online date6 Oct 2022
Publication statusPublished - 15 Dec 2022


  • autonomous ship
  • extended Kalman filter
  • nonlinear model predictive control
  • obstacle avoidance
  • path planning
  • time-optimal control
  • trajectory tracking


Dive into the research topics of 'Time-optimal obstacle avoidance of autonomous ship based on nonlinear model predictive control'. Together they form a unique fingerprint.

Cite this