A Computational Model Of Aggression De-escalation

This master thesis is part of the STRESS project which aims to develop a virtual-reality based training environment for aggression de-escalation. We describe the development of a generic computational model of deescalation that should be able to figure as the training environment’s decision-making component. Based on a literature study we start by designing a conceptual model. Important aspects of this conceptual model are the use of a de-escalation protocol and the distinction between reactive and proactive aggression. The conceptual model is then formalised into a computational model of a multi-agent system with two agents, a de-escalator and an aggressor. Next, the computational model is used to simulate characteristic scenarios that represent the variety of circumstances we expect the model to be able to represent. Finally, we describe an application that uses a support and analysis model to give essential feedback, and that is thereby a first step towards a system that provides de-escalation training. The most important conclusions are that the de-escalation model is able to simulate characteristic behaviour, that it allows for a basic form of de-escalation training to be given, and that it is theoretically possible to use it in the system envisioned by the STRESS project with the advantage of allowing generic script writing with respect to aggressor personalities.