Smart digital port of the future

Using deep learning to train a robust lineair regression model for autonomous driving vehicles in the port of Rotterdam

Harbour and Transport

The port of Rotterdam is the largest port and industrial complex in Europe. Between 1962 and 2004, the port of Rotterdam was even the largest port in the world. Rotterdam’s ambition is to be the ‘smartest port’ and with this, the city is taking the lead in the digital transformation of port and logistics. In addition, the port aims to have ships sailing through the port autonomously by 2030.

This project focused on the transport of containers by autonomous vehicles.


Enabling autonomous driving of vehicles and sailing of ships is a key goal to make work at the port more efficiently. This will save costs and allow more work can be done. But autonomous driving of vehicles and sailing of ships must be safe. Tricky: who is responsible if something goes wrong?

In this project, we were tasked with making a vehicle to drive autonomously safely in a digital environment. The autonomous vehicle can collect data while driving, using two types of sensors, a LIDAR and a SONAR. Manual moving of the vehicle with a joystick is also possible and give the opportunity to collect data from the car to use it to train a neural network so that the car can drive around the simulation environment independently and as safely as possible.

Robust Lineair Regression

Lineair regression is a supervised machine learning algorithm with inputs called feautrues and outputs called labels. PyTorch 2, a deep learning library for Python 3, has been used to train a robust lineair regression model with sensor data. With sixteen features from LiDAR and three features from SONAR, creating labels for speed and steering angle.

As it turned out the complexity wasn’t in the model, which in the end is a rather straight forward regression model, but in the tuning of the many parameters. Even overfitting wasn’t the biggest issue, as long as the vehicles keep moving on the same closed tracks in the port.

MLflow has been used for recording of the many tests that were performed, displaying the results in a nice dash board.


Thanks to this model, vehicles can move autonomously on a closed track to transport containers from the ships to the lorries for further transport. Further investigation is needed to have more vehicles on the track and thereby improving efficiency. Also a closer look at the safety regulations is necessary as technology normally develops faster than jurisdiction.