Pilot 3: Autonomous Ships in Short Sea Transport
Autonomous, electric barges will be used in transporting goods across Norwegian fjords, replacing millions of annual diesel-powered road kilometres. These barges will be controlled from remote operation centres, which will be equipped with systems for maritime traffic surveillance. To aid the operation of these barges in a secure manner, the pilot will provide relevant traffic information in a proactive manner. By estimating routes and schedules of surrounding traffic, barge-crossings can be planned with low risk of running into close quarters with other vessels.
Use Cases
During the first evaluation period of the project, the focus of the pilot was on the following use cases
Predict Routes of Surrounding Traffic
By using predicted routes, the maritime surveillance system can calculate optimal timeslots for barege-crossings that reduces chances of meeting other vessels.
Monitor Traffic
By detecting unexpected behaviour of surrounding traffic during a crossing, the operators of the autonomous barges can be alerted, so that they can take proactive measures if required.
Vessel Collision Risk Assessment and Forecasting
By identifying encountering vessels and assessing/predicting their risk of collision in the short term, we enable the operator to make quick decisions (e.g., evasive manoeuvres), thus improving maritime mobility awareness.
Pilot-participating Partners
Pilot Technological Highlights
The technological highlights of this pilot were the following
The pilot uses historical and current AIS traffic data. This data is combined with data from the Norwegian maritime single window, which contains voyage information for larger vessels in the same area.
Data
Historical and current AIS data is collected in a database and used for training of machine learning (ML) models developed using PyTorch and Scikit-Learn. Anomaly Detection and Collision Risk Assessment models are developed to warn operators of unexpected and/or potentially dangerous behaviour, either at the present or in the future via route prediction. These models are exposed through a REST-API that is called from the maritime surveillance system when predictions are needed. This API can be hosted externally or locally via a Docker container.
Components
- Accurate long-term route predictions (up to one hour) that improve the situational awareness of Shore Control operators by seeing into the future.
- Anomalous vessel behavior are predicted in due time for a Shore Control operator to take proper action.
- Route suggestion for unmanned vessel that minimizes collision risk and reduces evasive maneuvers.
Highlights
Pilot Results
Compared to the current state-of-the-art, ML-based models are found to provide more accurate and faster results. This is conducive with the needs of operators to get accurate live predictions of surrounding traffic. Route suggestions will also be provided to the operators based on the predicted traffic. It is expected the developed functions will result in increased situation awareness for operators.