top of page
Screenshot 2024-03-01 150848.png

Autonomous Marine Navigation


Thrace Marine is a start-up developing a hydrogen-electric recreational vessel. Early in the design process, Thrace Marine identified that fully electric marine vehicles have opportunities for features that would be difficult or impossible to implement in a typical gasoline powered boat. The motor response time and control accuracy of a fully electric drive system opened the possibility of an autonomous drive system, similar to Tesla’s AutoPark System.


The majority of marine accidents happen in marinas, with inexperienced owners at the helm of large boats that are difficult to manoeuver. The majority of accidental damage in marinas is caused by these inexperienced owners hitting the wharf or other vessels in the marina. The autonomous drive system in Thrace Marine’s hydrogen-electric vessel focuses on preventing these accidents and saving money by fully automating the parking process.

System Platform

Motus Design started by determining a development platform that would allow us to use a mix of off-the-shelf tools and custom solutions. Whenever possible we used existing technologies to shorten development time and cost. 


A key system that allowed for the integration of off-the-shelf solutions was Robot Operating System (ROS). ROS is the de-facto standard for robotics middleware. Middleware handles the low level communications, logic, and data recording for the system. While custom solutions are frequently used in production robotics, ROS makes it quick and easy to get a full system up and running.

Screenshot 2024-03-01 151523.png

Real-Time Object Detection

Identification and classification of objects in the environment is critical for making informed real-time decisions on pathing and collision avoidance. Recent advances in AI have made it possible to create adaptable object detection and tracking systems that can be run on edge devices.


We built a custom convolutional neural network (CNN) marine object detection model using YOLO v7, a high performance real time object detection and tracking algorithm that can run on edge devices. We chose YOLO v7 due to its high performance, ease of use, ability for customization, and reliability when running edge devices. Images were sourced from Roboflow and labelled by our team using CVAT, an open source data annotation platform.

Path Planning

An optimised model of Rapidly-exploring Random Trees (RRT*) was selected as the best path planning approach to use. RRT* is statistically guaranteed to find a path, and continually improves the path between system operations. A custom implementation gave us much greater flexibility around how the algorithm explored the map and optimised its search for the goal.

Screenshot 2024-03-01 151030.png

Completed System

The complete Thrace Autonomous Navigation Proof of Concept solution is an effective harbour navigation system that has the capability to reduce low speed accidents in marinas. It can map the local environment, navigate to a goal, identify objects in the environment, make decisions based on the objects, and avoid collisions. The dual electric stern drive powertrain allows for high manoeuvrability. The electric control system and drivetrain enables fast response times that would not be possible in a typical internal combustion drivetrain. 


The full autonomous system was tested over a range of marina traffic densities and wave/wind conditions. In each scenario, the vessel was able to map the environment and navigate to the destination without colliding into any objects or features.

The next phase of Thrace Marine will be creating a proof of concept demonstration vessel. This may take the form of a scale vessel that can be used to validate the sensors, data processing, and control systems. With the core navigation system functional, we will also spend time refining some of the more critical precision system functions, such as station keeping and docking. 


While the physical demonstrator is being built, we will continue with control system updates. Edge cases will be introduced, tested, and solved in future development cycles. Having control protocols and systems in place for events like obscured cameras, 3rd party fault collisions, and extreme weather are essential for moving from a proof of concept to commercial product.

bottom of page