Approach

Simulation of life in 3D reconstructed historical cities is a costly and time-consuming process, comparable in cost and efforts to development of a commercial video game (involving years of development and millions of dollars in funding (Gauder 2013)).

Costs and effort can be decreased with automatic generation of population. This is a two-fold process, in which we need to generate the unique appearance and the behaviour of each individual. Unique appearance can be generated by mimicking the biological reproduction, as for example in Trescak et al. (2012). One way of automa- tisation of behaviour is to represent individuals as autonomous virtual agents that can generate their goals and act upon them (Vosinakis and Panayiotopoulos 2001). To generate such goals, we propose to use motivation, and in particular physiological motivation, such as hunger, thirst, fatigue and comfort. In this case agents generate their goals upon physiological trigger, e.g. getting hungry. If needed, other types of motivation can be employed, such as safety, love, or self-realisation (Maslow et al. 1970; Alderfer 1969).

The problem with classical approaches to agents driven by physiological motivation is that in a historical simulation all such agents would follow the same circadian rhythm (get hungry, thirsty at the same time), what leads to undesired, uniform behaviour. To avoid this, in our methodology, we propose to configure motivational modifiers, which affect the decay rate of a given motivation. For example, a hunger modifier affects the pace in which an agent gets hungry. If such modifiers are different for every agent—then every individual follows its own circadian rhythm, executing goals at various time intervals, increasing believability of the simulated population.

In classical Artificial Intelligence (AI), in order to achieve a goal each agent needs a plan.

Such plans can be automatically generated using traditional planning techniques (Shehory et al. 1999; Braubach et al. 2005). Such planning techniques normally model perfectly rational behaviour, which is not always suitable for simulating humans as this results in emotionless, “robotic” behaviour. To avoid it, in our methodology, we enrich agents with personalities and emotions, which affect their decisions when creating a plan for a current goal. This approach may even lead to emergent agent behaviour that appears to be closer to human-like reasoning. As an example, imagine a fisherman agent with no personality and emotions, that catches fish when it’s hungry. The agent will fish until it succeeds, or until it dies of hunger, unless we manually specify a possible change of plans when hunger level raises to a critical value. In contrast, the same fisherman having personality and emotions may get frustrated when being hungry and unsuccessful. This agent may “decide” to stop fishing when frustration level overwhelms the rational decision for fishing and will search for alternatives to feed, such as begging or stealing food. The decision whether to beg or steal would depend on agent’s personality.

In the previous example, fisherman represents a specific social group of the simulated population. Social groups combine certain classes of individuals that fulfill their goals in a similar way. Combining individuals into social groups allows us to define and program actions on a group level, rather than having to do this on individual level, reducing effort in defining crowd behaviour.

In human societies, it is not uncommon for members of different social groups to interact with each other and even cooperate in order to fulfill their goals. For example, imagine a fisherman who has to trade fish with a spear-maker in order to replace his broken fishing spear (see Sect. 14.4). A common technique being used in AI to facilitate the kind of interactions between different social groups as in the example above is to employ Organisation-Centred Multi-Agent System (OCMAS). The OCMAS approach is to explicitly formalise social norms of the agent population and connect those norms to the social roles, which represent different population groups. Such social norms capture rules and protocols that drive agent interactions. As a result, agents can use these norms in reasoning to create plans for their current goal. This provides agents the ability to automatically perform their actions depending on their assigned social group.

14.3

<< | >>

↑

Source: Barcelo Juan A., Del Castillo Florencia (eds.). Simulating Prehistoric and Ancient Worlds. Springer,2016. — 410 p.. 2016

Approach

More on the topic Approach: