Machine Learning for intelligent avatars in Virtual Reality simulations: Training environment for CBRN procedures

Training in Virtual Reality (VR) has proven to be a much more efficient alternative than traditional methods. It can provide an immersive experience, but it is important to improve the realism of VR simulations in order to achieve a better experience. Most of the times, a simulation involves other avatars that can help the user in his or her task, which must respond correctly according to the situation. This work focused on two main objectives. The first one is to develop a simulation in order to train Air Force personnel on the CBRN (Chemical Bacteriological Radiological Nuclear) procedures, that are procedures to be followed in situations in which CBRN hazards may be present. The second one is to introduce a Machine Learning (ML) algorithm able to control the avatars that will help the player during the simulation. This thesis is a collaboration between Politecnico di Torino, LINKS Foundation and the 3rd Wing of the Italian Airforce.