Noting the limitations of conventional navigation algorithms, the ARION R&D team proposes to create autonomous navigation systems that integrate new methods of interpretation and understanding of the environments in which autonomous vehicles travel. These methods are based on the use of new information analysis techniques such as machine learning and more specifically deep learning. One of the approaches targeted in the research proposal to increase the quality of the interpretation of the environment is semantic scene analysis. This is an innovation that is considered particularly promising for making autonomous vehicles capable of driving in changing and highly dynamic environments, for example on snowy roads in winter where an autonomous shuttle will be driving or in agricultural fields where the growth of crops strongly modifies the environment perceived by an autonomous tractor. This approach could also be used for other vehicles of interest to the ARION programme, such as industrial handling and specialised heavy transport vehicles. In addition, the R&D team also intends to use deep learning to improve the understanding of the environment in which autonomous vehicles operate. Prediction algorithms will be implemented and tested to, for example, predict the trajectory of pedestrians in the vicinity of autonomous vehicles or to anticipate the dynamic behaviour of the vehicle in a given environment.