Leveraging Past Experience for Path planning of Marine Vessel: A Docking Example

Abstract

Path planning before maneuvering is crucial for the safe and efficient operations of marine vessels. The past successful human maneuvering experience can be leveraged to enable the safe and efficient path planning of vessels. In this paper, the previous successful maneuvering operations from ship operators are leveraged to find the optimal path. A deep conditional generative model is used to learn the distribution from those experiences. The model is then combined with the sampling-based RRT* planning algorithm to guide the search process. In this way, the theoretical guarantee of RRT* is preserved while the sampling process is more efficient. The docking operation is used as an example to validate the method. Experimental results show that the presented method not only improves the success rate and convergence speed to the optimal cost but also generalizes well to starting points beyond maneuvering experience.

Keywords

• path planning
• rrt*
• conditional variational autoencoder
• learning from past experience