Task dynamics define the contextual emergence of human corralling behaviors

Abstract

Social animals have the remarkable ability to organize into collectives to achieve goals unobtainable to individual members. Equally striking is the observation that despite differences in perceptual-motor capabilities, different animals often exhibit qualitatively similar collective states of organization and coordination. Such qualitative similarities can be seen in corralling behaviors involving the encirclement of prey that are observed, for example, during collaborative hunting amongst several apex predator species living in disparate environments. Similar encirclement behaviors are also displayed by human participants in a collaborative problem-solving task involving the herding and containment of evasive artificial agents. Inspired by the functional similarities in this behavior across humans and non-human systems, this paper investigated whether the containment strategies displayed by humans emerge as a function of the task’s underlying dynamics, which shape patterns of goal-directed corralling more generally. This hypothesis was tested by comparing the strategies naïve human dyads adopt during the containment of a set of evasive artificial agents across two disparate task contexts. Despite the different movement types (manual manipulation or locomotion) required in the different task contexts, the behaviors that humans display can be predicted as emergent properties of the same underlying task-dynamic model.

A Comparison of Dynamical Perceptual-Motor Primitives and Deep Reinforcement Learning for Human-Artificial Agent Training Systems

“…Accordingly, some dyads discover and adopt a novel behavior, whereby instead of responding to the fleeing movements of individual TAs, participants perform coupled, rhythmic movements around the whole herd to form a sort of ‘wall’ around the herd. Interestingly, similar herding-like behavior involving highly coordinated patterns of movement is seen across various animal systems engaging in collaborative hunting and shepherding (D’Vincent et al, 1985; Muro et al, 2011; Nalepka, Silva, et al, 2021).…”

“…Consistent with other recent research exploring the dynamics of complex multiagent behavior (Lamb et al, 2019; Richardson et al, 2015), the current findings provide further evidence that the emergent behavioral dynamics of complex multiagent tasks can be defined using low-dimensional, task-dynamical models consisting of DPMPs. Moreover, such models continue to reveal the low-dimensional nature of goal-directed human behavior (Saltzman & Kelso, 1987), as well as the generic similitude with which simple point-attractive and limit-cycle processes underly functional human behavior across a wide range of task contexts (see also Nalepka, Silva, et al, 2021).…”

“…A total of 164 participants completed the study: 34 participants (17 pairs) in the Full-DPMP condition, 32 participants (16 pairs) in the Limited-DPMP condition, 34 participants (17 pairs) in the DRL condition, 32 participants (16 pairs) in the expert-human trainer condition, and 32 participants (16 pairs) completed the task in the untrained novice (baseline/control) condition (excluding three additional pairs who were excluded from analysis due to equipment malfunction). The sample size was determined following previous research utilizing the multiagent herding task paradigm (Nalepka et al, 2019, 2017; Nalepka, Silva, et al, 2021; Rigoli et al, 2020).…”

“…Further, the TAs are repelled from the HAs when the HAs come within a certain distance, fleeing in the opposite direction to the HA. Thus, HAs can influence and corral the TAs by moving close to them – as is the case in real-life shepherding and herding (Nalepka, Silva, et al, 2021; Strömbom et al, 2014).

Figure 1.Behaviors observed during the multiagent herding task.

…”

“…Perhaps one of the best examples of how DPMPs can successfully model complex human multiagent behavior is the research demonstrating how a simple DPMP model can capture the coordinated behavior of pairs of human agents (and human-AI pairs) performing complex herding tasks (Nalepka et al, 2017, 2019; Nalepka, Silva, et al, 2021; Rigoli et al, 2020). In this task, participants control herding agents (HAs) on a game field and are tasked to work together to corral and contain evasive, autonomous virtual target agents (TAs) within a containment area for a specified period (see Figure 1).…”