A high fidelity digital twin framework for testing exteroceptive perception of autonomous vessels

In february 2021 Kjetil Vasstein finished his master thesis at NTNU Cybernetics where he investigated how to create a high fidelity digital twin framework for testing exteroceptive perception for autonomous vessels. The specific case was the NTNU autonomous vessel prototype milliAmpere. The technology Kjetil helped to develop will be used in the continuous developing and testing of the Zeabuz autonomy system. Below is the abstract from Kjetils master thesis.

Abstract

In the future, it is believed that verification of autonomous ships must involve simulation based verification, due to the huge costs involved in real life testings. In contrary to the car industry, where open source digital twin frameworks exists, there is no equivalent solution for the maritime industry. There are also few studies that shows if simulations can be trusted, moreover how they affect algorithms used by autonomous agents such as target trackers. This thesis implements a digital twin framework, capable of simulating camera and lidar data using the Unity game engine, in order to test a simulation verification method, where a real and synthetic dataset with common ground truths are compared using a Hellinger distance on the outputs of a target tracker. The results demonstrate a metric that is capable of measuring fidelity, establishing a quantitative method that can be used for future improvements towards trustworthy simulation based verification for autonomous marine vessels.

Figure 1 Image from simulation.