Non-linear interactions of perceived behaviours in collective flow

Abstract

In everyday life our visual system is continuously exposedto a wide range of motionflow patterns. Our knowledgeof low-level motion processing is substantial, while higher-level processes such as future state prediction, motion con-stancy, and behavioural property estimation remain poorlyunderstood. Here, we concentrate on behavioural cues ofcollectiveflow. Collectiveflow exists of a body of individualagents that show both collective and individual behavioursfollowing a coordinated set of rules. In nature there aremany occurrences of collectiveflow on various scales, var-ious levels of complexity, across both animate and inani-mate systems (e.g., swarms of insects, cars on highways).Using a real-time browser-based simulator of a relativelysimple six-dimensional parametric model we displayed arange of collective behaviours. In a variety of experimentswith free naming, name selection, similarity judgements,and rating tasks we started exploring the parametricspace and its perceived behavioural dimensions. Wefindthat observers can name a wide range of behaviours despitethe abstraction of the simulations. Observers found thewords expressing the spacing between agents to be themost descriptive. However, in the rating experiment it wasfound to be a challenge to differentiate between more dis-tinct definitions of this spacing such as grouping or dispersal.Moreover, the six-dimensional parametric space containedmultiple instances of the same perceived behaviour, makingdirect mappings between the parametric space and percep-tual space even more complex. The challenge will be toclearly tease apart the perceived behaviours with their non-linear interactions across the explored parametric space.[This work was supported by a Marie-Skłodowska-CurieActions Individual Fellowship (H2020-MSCA-IF-2019-FLOW, Project ID: 896434) and a Marie-Skłodowska-CurieActions Innovative Training Network (MSCA-ITN-ETN,grant number 765121, 2017) DyViTo.]