Environment-Driven Social Force Model: Lévy Walk Pattern in Collective Behavior

Danyan Lv, Zhaofeng Li, Yichuan JiangIJCAI 2015

Introduction

In the study of collective animal behavior, researchers often encounter a "behavioral conflict" paradox. Individual animals must navigate toward resource-rich patches to feed (environmental effect) while simultaneously maintaining proximity to their group for protection and social benefits (conspecific effect). Traditional models often struggle to balance these needs; social interaction models like the Vicsek model produce cohesive groups but lack realistic foraging paths, while optimal foraging models often ignore the social constraints that keep groups together.

This paper introduces an Environment-Driven Social Force (ESF) model that bridges this gap. By implementing a "cohesion-first" rule, the model enables groups to maintain integrity while individuals perform efficient "Lévy walk" search patterns—a specific type of random walk characterized by many short steps and occasional long "flights."

Methodology: The ESF Model

The proposed model expands the classic Social Force Model by incorporating environmental attractions and a specific decision-making hierarchy. Each agent's movement is governed by three primary forces:

  1. Friction ($F_f$): Simulates the "stickiness" of food, making it harder for agents to leave resource-rich pixels.
  2. Pull Force ($F_p$): An attractive force toward nearby abundant food sources based on the agent's perception range.
  3. Repulsion-Attraction Force ($F_e$): The social component, managing inter-agent distances to avoid collisions while preventing group dispersal.
Agent perception zones and social distance parameters.

A critical component of the ESF model is its switching strategy between two distinct behavioral paradigms based on the agent's current state (velocity $v_i$ and food availability $A_i$).

The switching strategy between searching and feeding paradigms.
  • Searching Paradigm: Used for transit between patches, where agents follow Newtonian mechanics derived from the resultant of all three forces.
  • Feeding Paradigm: Triggered within a food patch, where agents prioritize localized movement and social "comfort zones" over global force calculations.

Key Insights and Results

The model was tested in a synthetic $600 \times 600$ pixel environment with randomly distributed food patches and a group of 10 autonomous agents. Key findings include:

  • Lévy Walk Emergence: The model successfully generated movement patterns following a power-law distribution ($P(l) \sim l^{-\mu}$), characteristic of Lévy walks. The simulation's flight lengths matched real-world data from species like sheep and albatrosses with high statistical significance ($R^2$ up to 0.95).
  • Density-Dependent Strategy: The Lévy walk exponent ($\mu$) was found to be dynamic. At low food densities, agents performed more long-range "flights" ($\mu \approx 2$), whereas high densities led to more localized, intensive searching ($\mu \approx 3$).
  • Scarcity and Consensus: Interestingly, food scarcity was found to facilitate group consensus. When resources are limited, the group exhibits higher "polarization" (aligned movement). Upon reaching a rich patch, the group becomes more disordered as individuals focus on localized feeding.
  • Comparative Performance: In benchmarks against the Vicsek, standard Social Force, and pure Lévy Flight models, the ESF model was the only one capable of maintaining both high group polarization and realistic individual search patterns.

Conclusion

The "cohesion-first" rule effectively resolves the conflict between individual foraging efficiency and collective stability. While the current model uses a small group of identical agents in a simplified 2D environment, it provides a robust framework for understanding how complex individual search patterns can emerge from simple social and environmental rules. Future research will explore scaling these dynamics to larger groups and incorporating agent diversity and more complex, dynamic terrains.

Summary updated

BibTeX
@inproceedings{lv2015environment,
  author = {Danyan Lv and Zhaofeng Li and Yichuan Jiang},
  title = {{Environment-Driven Social Force Model: Lévy Walk Pattern in Collective Behavior}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2015},
}
Figure