The use of computer simulations as a tool for training qualified professionals on the skills needed in their jobs has a long tradition in many areas (healthcare, military, emergencies, transport, etc.). In general, these are areas where reproducing the training scenario in the real world is costly, dangerous or directly unfeasible. Recently, the interest in these kinds of simulations has greatly increased due to the improvement and cost reduction of immersive technologies like Virtual Reality (VR) and Augmented Reality (AR) that reduce the distance between reality and the simulation. By increasing the immersion of the participants in the simulation, we also increase the effectiveness of the training simulation.

Many organisations around the world currently use platforms that enable the definition of 3D emergency scenarios for training their
personnel. During the execution of the training exercise, one or several trainees can participate in the simulation that is controlled by
human operators.

These kinds of platforms present two important limitations. First, they do not allow having multiple synchronised scenarios at different scale levels. This means that it is not possible to have large settings that combine different operational and tactical levels. The reason is that the simulation is built directly on top of the graphical engine and attached to the simulation scale fixed by the detail in the 3D environment. It is not feasible to have large training scenarios at the level of detail necessary for low level operational activities. Second, they rely strongly on human operators to conduct the simulations. To execute even medium-size simulations, you need several qualified operators to deal with all the manual things that need to be controlled during the training session. The level of autonomy of the different simulation parts is very low. As a result, performing training sessions is costly in terms of organisation and execution and there is a limit in the complexity of the training simulations that can be carried out.
With the framework proposed in this project (the RHYMAS framework) we want to overcome these two main drawbacks and allow for large training simulations that combine different tactical and strategic levels and that at the same time are easy to enact, easy to control and with a high training capacity.

Starting hypothesis: The performance of humans in emergency situations may be improved by enabling sophisticated human-in-the-loop agent-based simulations with strong AI foundations.

The objectives of the project are the following:
OBJECTIVE 1. To improve human response in complex emergency scenarios by developing an AI-based training framework.
OBJECTIVE 2. To go beyond the state of the art in simulations involving humans by developing mechanisms that adapt to dynamic scenarios, based on norms and driven by values and goals.
OBJECTIVE 3. To evaluate the improvement in human response through two use cases that test our work in a real-life setting of
firefighters training (in collaboration with the Institut de seguretat pública de Catalunya (ISPC)).