Lloyd's Register
The American Club
Panama Consulate
London Shipping Law Center
Home HRCommunication MarISOT – Neural network

MarISOT – Neural network

by admin
In our Posidonia miniseminar, technical lead Timo Haavisto – picured above, will tell more about our technical innovations

Neural network

MarISOT backend has been developed with data gathering capabilities tailored for neural network training and validation data collection. At high level, the data can be separated into two distinct categories: scenario physics data and user behaviour data.

Scenario physics data contains the information of other vessels in the scenario, their current locations and identifiers, logged at 100 millisecond intervals. All vessels in the scenario, including the one operated by the user, are recorded at the same time in order to create a snapshot of the environment situation at any point.

The paths of the vessels are affected by the ocean dynamics as well as other vessels. The user’s ship is operated by the user, and therefore has different path and course depending on the actions of the user.

With the snapshots as a time series, it is possible to train a neural network to determine if the ship’s course in relation to the other vessels in the scenario indicates a completed scenario: pass or fail of the training. It is also capable of measuring the similarity of the path user is taking, in comparison to previously recorded training set’s passing or failing paths.

If the neural network inference is run in the middle of the scenario, it may provide information on how well the user has executed their scenario so far.

Behaviouristic data contains tracking of user’s individual actions and tracked properties: the user’s movements in the command bridge, the user’s gaze targets and eye pupil size and the user’s operations of the environment controls. Using the collected behaviour data, it is possible to create time series of the user’s operations in the environment, which could be used to determine their problem-solving process inside the scenario.

Neural network could be trained to detect patterns of behaviour, that often predict failure of the scenario. These could be visualized as indicators in the training phase. In upcoming Posidonia 2022 exhibition next week, we will finally reveal these achievements. In MarISOT – Next Generation Training in Maritime Safety seminar organized Friday 10th, we will have three presentations and panel discussion where you can meet our specialists. In addition, we will be demonstrating our innovations in stand 2.554.

You may also like

Leave a Comment