The EMISTRAL project aims to design a controller for autonomous sailboats combining machine learning methods, reinforcement learning, transfer learning, domain adaptation, Internet of Things and modeling and simulation. It is a technology independent of a particular vessel model, with the ability to autonomously adapt to the characteristics of the vessel where it is deployed.
To this end, a simulator will be created for a specially built sailing ship, representing the different combinations of atmospheric and sea conditions, as well as the control characteristics of modern sailing ships, and also interpreting the ship's interactions with the wind and the sea.
EMISTRAL will allow understanding the potential of learning transfer methods to adapt a pre-trained model to different vessels. It will partner with work developed at the Universidade Federal Fluminense and Universidade Federal do Rio Grande do Norte, both in Brazil. In particular, it seeks to collaborate in the development of the latest generation prototypes of the N-Boat type, which will serve to create a first deployment of the project results. Later on, the extension of results is also proposed by creating a package of IoT sensors specially designed for this purpose, which will increase the capabilities of obstacle detection and recognition, a better understanding of the state of the sea and environmental and ecological monitoring.
EMISTRAL is a project aimed at the construction of unmanned vessels. In particular, it is expected to be used in coastal eco-surveillance and in general to obtain marine data such as salinity samples, chlorophyll quantity, microplastics concentration and water quality in regions with marine pollution, among other applications.
The initiative, financed by the STIC AMSUD program, is made up of researchers from Inria Chile; the Inria ACO and SCOOL project teams; the Universidad de la República, Uruguay; and the Universidade Federal Fluminense and the Universidade Federal do Rio Grande do Norte, both in Brazil.
Visit EMISTRAL website here.