The new era of autonomous tugboats and intelligent docking for large ports | MOSES project
MOSES project
The new era of autonomous tugboats and intelligent docking for large ports
Author: Manthos Kampourakis
January 11th 2023
Nowadays, hub port operations are becoming less efficient due to congested waterways, manoeuvring and berthing processes that are error-prone, time-consuming, costly, vulnerable to disruptions (e.g., strikes), and accidents with significant environmental impact. The MOSES project tries to tackle all these challenges by adopting an autonomous vessel manoeuvring and docking scheme that provides operational independency from the availability of port services. This innovative scheme is based on the cooperation of a coordinated swarm of autonomous tugboats that automates manoeuvring and docking where CORE Innovation is in charge.
The first part for which CORE Innovation was responsible was the design and development of a swarm of intelligent virtual tugboat agents, capable of both performing accurate docking of a large containership and coordinating their actions. The goal of the swarm was to pull off simulations of two virtual scenarios which demonstrate the added swarm intelligence factor. For the first scenario, starting from an initial distance of about 80m from the dock, the push tugboat agent’s mission, was to assist on berthing while the large containership would apply corrective movements using its bow thruster. The pull tugboat agent aimed at maintaining the mother vessel’s yaw angle close to zero. For the second scenario, a swarm of four tugboat agents in total learnt to assist on the berthing operation (two push agents) while again maintaining the mother vessel’s yaw angle (two pull agents).
In this video, the sped-up demonstration of the first simulation scenario, the accurate docking of a large containership, is displayed. The upper right corner displays the distance between a marked position on the tip of the dock and a point at approximately the middle of the starboard of the vessel. The bow thruster applies corrective movements to the yaw angle using a custom script. The whole system covers about 70m until it reaches the dock at a predefined distance. At this point, the two tugboats decelerate and then an automated system can take over for docking.
To achieve this, the Unity3D simulation environment and the 3D models of all involved actors were used. Real-world performance was achieved by calibrating the environment physics and the development of custom Reinforcement Learning algorithms led to the successful swarm training. All relevant information needed to accomplish each agent’s task were given as inputs; the location, acceleration, and distance of virtual LiDAR sensors. Propulsion and steering control outputs enabled agents’ navigation. Learning was achieved by employing tailor-made reward signals that directed the learning process in their policies and at the same time penalized undesired tugboat actions e.g., collisions.
