/.R. Marx et al.
The automation and control group of the Technical University of
Denmark (DTU) publishes the continuous progress in methods and tech-
21ologies for the Danish autonomous ferry initiative, the GreenHopper
‘Hansen et al., 2022; Enevoldsen et al., 2023; Blanke et al., 2024). Green-
Hopper is a 12.2 m long zero-emission ferry crossing the Limfjorden near
city of Aalborg (Denmark) on a distance of 700 m. The initiative aims to
demonstrate a decision support system for the validation of experiments
with temporally unattended navigation, whereas the task of human op-
erators are represented in the functionalities of the software modules.
The safe, automatic navigation, collision and grounding avoidance cor-
responding to the COLREGs is based on situational awareness and multi-
nodal perception using additional diverse optical sensor technology for
differing weather conditions. In Dittmann and Blanke (2022), a resilient
design of the automation system for autonomous operation has been de-
:ailed to navigate safely in default situations and to request human assis-
:ance when the automation system is unable to handle a situation. This
contribution also included considerations on how to respond to failures
of instruments, sensors and their signals as well as cyber-attacks. A two-
year, continuous test phase of the GreenHopper with all the technologies
involved is planned under the supervision of a professional navigator.
Another advanced project initiative is the development of the au-
:onomous surface platform Roboat in a cooperation of institutions from
Zambridge, Massachusetts, USA and from Amsterdam, Netherlands. The
4m long Roboat is available in different versions depending on the ap-
plication as a water taxi (with up to six persons) or for waste collection
[Wang et al., 2023). The boats are equipped with real-time kinematic
‘RTK) global positioning system (GPS), lidar, a camera, and an inertial
neasuring unit (IMU). Teleoperation is enabled through a 5G mobile
outer and a private network connection is reserved for remote access.
The autonomy framework of Roboat contains a path planner, adaptive
1onlinear model predietive controller (NMPC) as well as object detec-
tion and tracking. The path planning provides a graph that reflects the
observed obstacles in the current field-of-view of the sensors (Shan et al.,
2020). The following motion planning uses this graph of the receding
horizon planner for a lexicographic optimization. The NMPC works with
a nonlinear motion model and parametric cost function which are adapt-
able to the current, estimated payload and the resulting varying dynamic
of the vessel (Wang et al., 2021). There were already many successful
system tests in simulation and in real-world. In future, the system is to
De further improved by considering traffic rules in the planner, inte-
grating disturbance models in the controller and enabling multi-robot
formations.
The Norwegian University of Science and Technology (NTNU) estab-
ıished a autonomous prototype for small, electrical ferries that serve as
a development and test platform (Brekke et al., 2022). The 8.5m long
milliAmpere2 is an improved version of the 5m long milliAmpere, also
equipped with an extended configuration of optical sensors. For situa-
zional awareness, the data from radar and lidar are fused. The automatic
control with extensive redundancies is realized with an industrial DP
system controlling the four azimuth thrusters. The ferry crosses a short
distance on a canal in Trondheim, Norway.
A comprehensive German approach was presented with the 8.5 m
:ong platform Solgenia of the University of Konstanz, Germany describ-
ing the latest development toward energy-efficient autonomous ship-
ping on Lake Constance (Homburger et al., 2025). The platform is
2quipped with lidar, camera, radar, RTK GPS, IMUs and actuator sen-
sors. A motion model was identified that includes the effects of current
Äelds, wind and different actuator configurations. The trajectory plan-
aing can be optimized according a balance between time and energy
consumption. An NMPC approach is applied to track this reference tra-
jectory smoothly and energy-efficiently. Collision avoidance is based on
he detection and tracking of objects using the fused sensor data from
the camera and lidar.
In addition to the initiatives mentioned above, which have been
the subject of scientific publications, there are a number of industrial
projects aimed at advancing autonomous shipping such as Yara Birke-
Ocean Engineering 343 (2026) 123388
and and ASKO maritime in Norway, ZEAM ferry in Stockholm, Sweden
or larger projects in Asian region.
As a rule, the projects mentioned involve the automation of a ship
‘hat moves safely in its environment and automatically avoids detected
and classified objects. For cost and legal reasons, smaller boats under
20 m are usually used in university research projects. In Asia, the focus
s on the immediate industrial implementation of autonomous shipping
.n the construction or retrofitting of conventional ships. The comprehen-
;ive approach to automatic avoidance in ports presented in this paper
‚s characterized by the involvement of three simultaneously operating
ships, whose avoidance paths are calculated centrally cooperative, and
‘he involvement of a conventional ship with a crew and a length of more
han 50m. As usual, model predietive control is performed separately
‘or each vehicle, based on the same motion models as the trajectory op-
imization. The automation system is designed to be so transparent that
‘he captain of the DENEB, in cooperation with the officer on watch on
‘he ship’s bridge of DENEB, can command and monitor the automatic
naneuvering of all three vehicles.
1.2. Own preliminary work
The authors are pursuing the approach of step-by-step maneuver au-
:;omation, which automates individual maneuvers based on assistance
unctions and subsequently combines them into automatic maneuver
;equences (Schubert et al., 2018). One of the advantages of this ap-
’roach is the transparency towards the nautical officers. There is a clear
ıandover procedure between manual and automatic control and vice
versa, similar to the use of an autopilot. The correct functioning of the
automatic control can be watched via a monitoring system by compar-
ng the nautically valid maneuver plan with the actual and model-based
prediction.
The established Maneuver Assistance System (MAS) was developed
during the successive phases in the project GALILEOnautic, funded
»y German Federal Ministry of Economic Affairs and Climate Action
'BMWK). MAS has already been used on several vessels, on the hybrid
‚erry BERLIN, Scandlines, to support manual maneuvering (Schubert
et al., 2019; Baldauf et al., 2024), and additionally for automatic maneu-
vering on the research vessel DENEB, BSH Germany, and the unmanned
surface vehicle (USV) MESSIN, University of Rostock.
Automatic maneuvers with different operation modes and con-
rollers were successfully performed with the research vessel DENEB
‘Rethfeldt et al., 2021; Hahn et al., 2022). The switching between the
nodes for in-port transit and dynamic positioning (DP) operation de-
pends on the velocity and the distance to target position. In the course
of the project, several types of vehicle motion models were tested as
a basis for the control system and the allocation methods. Initial ap-
»roaches to automatic control were based on a parameter space model
‘hat represents the motion behavior as a first-order system in each de-
zree of freedom (DoF), with the non-linearities are contained in look-up
:ables (Schubert et al., 2022). Other approaches related to identifying
he maneuvering behavior directly from regular operations of the ves-
sel, in which the acceleration periods are detected and weighted (Hahn
et al., 2023). The first allocation applied corresponded to the inverted
jarameter space model, published in Hahn et al. (2022). Later, the
]uadratic programming method was used on the research vessel to al-
‚ocate the commanded forces with minimum resulting energy consump-
tion (Koschorrek et al., 2018; Rethfeldt and Jeinsch, 2024; Schubert
ot al., 2024b). Tha automatic maneuvering with the USV MESSIN was
used during a survey campaign in shallow waters, published in Kurowski
et al. (2019b).
A model for wind disturbances for the research vessel DENEB was
‚ntroduced and used to estimate the varying power consumption during
naneuvers (Schubert et al., 2024a). MPC approaches have so far been
applied to the model of unmanned vehicles that are to be used as con
voys on inland waterways as part of the project A-SWARM (Marx et al.,
2023). The motion behavior of the vehicles at low speeds is influenced
