Perceived Realism of Crowd Behaviour with Social Forces Stuart O'Connor, Fotis Liarokapis, Chrisina Jayne Coventry University, Masaryk University, Coventry University {oconno13@uni.coventry.ac.uk, liarokap@fi.muni.cz, ab1527@coventry.ac.uk} Abstract This paper investigates the development of an urban crowd simulation for the purposes of psychophysical experimentation. Whilst artificial intelligence (AI) is advancing to produce more concise and interesting crowd behaviours, the number or sophistication of the algorithms implemented within a system does not necessarily guarantee its perceptual realism. Human perception is highly subjective and does not always conform to the reality of the situation. Therefore it is important to consider this aspect when dealing with AI implementations within a crowd system aimed at humans. In this research an initial two-alternative forced choice (2AFC) with constant stimuli psychophysical experiment is presented. The purpose of the experiment is to assess whether human participants perceive crowd behaviour with a social forces model to be more realistic. Results from the experiment suggest that participants do consider crowd behaviour with social forces to be more realistic. This research could inform the development of crowd-based systems, especially those that consider viewer perception to be important, such as for example video games and other media. Keywords--- Crowd Simulation, Psychophysics, Perception, Artificial Intelligence, Agent Behaviour, Virtual Environments. 1. Introduction The process of simulating huge crowds of intelligent agents in real-time is a challenging task due to numerous different considerations, technical or otherwise [12]. When the element of human perception is also taken into account, then this task becomes even more complex. Crowd behaviour is typically simulated with agent artificial intelligence algorithms (AI). The resultant behaviour can differ drastically depending on the specific algorithms that are implemented within a crowd based system, e.g. different pathfinding algorithms can deliver different paths for the same goal. Whilst some may assume that utilising lots of the most sophisticated algorithms would lead to believable behaviour, this is not always the case [16]. In light of this some form of quantification is needed in order to assess the behaviour of agents within crowd simulations. Realism, in the context of a crowd simulation, is the degree of plausibility of the crowd behaviour. Perceived realism in particular is a type of realism that is centered on the perception of humans. This type of realism is highly important for crowd simulations with a purpose geared towards interaction with humans, e.g. video games, and as such can be utilised to determine the degree of success for crowd behaviour [21]. Previous research has already utilised perception as a tool for evaluating crowds [7]. This research suggests a novel perspective of utilising psychophysics when evaluating crowd simulation systems. Psychophysics refers to a broad range of psychology based experiment methodologies, which are used in many ways to evaluate human perception through varying the levels of a stimuli and measuring the response [8]. Such methods when considered with perceived realism in mind can be adapted to perceptually evaluate the crowd behaviour of agents within a simulation. A specific behavioral feature is selected in order to conduct a forced choice psychophysical experiment to assess the perceived realism of a single aspect. The selected behavioral feature is the social forces model [9], an algorithm that has been implemented in many different crowd simulations. To this end an urban crowd simulation is developed to produce varied stimuli for different intensities of the different social forces. These stimuli are utilised under the psychophysics method for comparing the perceived realism between virtual scenes with various intensity of social forces present and scenes without social forces present. This research aims to address these aspects in the specific scope of an urban crowd simulation when considering perceptual realism. Whilst there has been some research towards evaluating crowds [4][23], the use of psychophysical methods to assess perceptual realism is an area open for further research. Typically research towards crowds is focused on further developing algorithms [10][13] and applying current methods to specific scenarios [1][3][26][27], so furthering knowledge for evaluation methods of crowd behaviour will add to the field. 2015 19th International Conference on Information Visualisation 1550-6037/15 $31.00 © 2015 IEEE DOI 10.1109/iV.2015.88 494 2015 19th International Conference on Information Visualisation 1550-6037/15 $31.00 © 2015 IEEE DOI 10.1109/iV.2015.88 494 2015 19th International Conference on Information Visualisation 1550-6037/15 $31.00 © 2015 IEEE DOI 10.1109/iV.2015.88 494 2015 19th International Conference on Information Visualisation 1550-6037/15 $31.00 © 2015 IEEE DOI 10.1109/iV.2015.88 494