Simulating complex social behaviours of virtual agents through case-based planning

•Introduced Dynamic Institutions combine (i) the normative control of OCMAS systems, with (ii) the expressivity and robustness of BPMN systems, with (iii) planning capabilities of standard planning systems (i.e. STRIPS) and (iv) performance of statically compiled systems.•DI can be employed in various scenarios. Its architecture allows Dis to be deployed in the distributed manner, where high-performance simulation can run on a separate machine to the visualisation machine, improving the system performance and user experience.•Using our version of case-based planning, we have increased the number of dynamically planning agent in real-time at an acceptable framerate from around 40–4000. This also includes almost 10-fold increase to our previous results. [Display omitted] In commercial video games and simulations, non-player characters are capable of quite complex behaviour. Very often though, each class of non-player characters (that we further call virtual agents) is manually programmed or scripted. This means that instead of possessing some level of intelligence, allowing the agent to decide dynamically on the actions it needs to perform, we supply the agent with a list of possible situations that may arise in the game. For each such situation, we give the agent a pre-programmed script that tells it how to behave. Producing such scripts for every role an agent might play in a game or simulation is a very costly exercise. This may be acceptable in commercial game development, where budgets for modern games are sometimes comparable to budgets of Hollywood movies, but not adequate for research simulations and indie games. In this paper, we discuss how indie games and research simulations can be enriched with the sophisticated social behaviour of virtual agents in a semi-automatic manner through the use of AI planning. By supplying agents with roles and developing a computational model of their needs, we can use AI planning (also known as dynamic planning) to increase the complexity of agent behaviour dramatically and at the same time achieve a high degree of automation and reduce the development costs. AI planning is gaining popularity in games development, but it is often discarded due to performance issues. We will show how to improve the performance of planning process through the use of dynamic institutions and case-based planning. We will illustrate the aforementioned ideas on an example of developing a Virtual Reality simulation of everyday life in Anci