Multi-Robot-Guided Crowd Evacuation: Two-Scale Modeling and Control Based on Mean-Field Hydrodynamic Models

Tongjia Zheng, Zhenyuan Yuan, Mollik Nayyar, Alan R. Wagner, Minghui Zhu, Hai Lin

Abstract

Emergency evacuation describes a complex situation involving time-critical decision-making by evacuees. Mobile robots are being actively explored as a potential solution to provide timely guidance. In this work, we study a robot-guided crowd evacuation problem where a small group of robots is used to guide a large human crowd to safe locations. The challenge lies in how to use micro-level human-robot interactions to indirectly influence a population that significantly outnumbers the robots to achieve the collective evacuation objective. To address the challenge, we follow a two-scale modeling strategy and explore hydrodynamic models, which consist of a family of microscopic social force models that describe how human movements are locally affected by other humans, the environment, and robots, and associated macroscopic equations for the temporal and spatial evolution of the crowd density and flow velocity. We design controllers for the robots such that they not only automatically explore the environment (with unknown dynamic obstacles) to cover it as much as possible, but also dynamically adjust the directions of their local navigation force fields based on the real-time macrostates of the crowd to guide the crowd to a safe location. We prove the stability of the proposed evacuation algorithm and conduct extensive simulations to investigate the performance of the algorithm with different combinations of human numbers, robot numbers, and obstacle settings.

Introduction

Emergency evacuations are notoriously chaotic, requiring rapid decision-making under stress. While mobile robots offer a promising solution for providing guidance, controlling a small number of robots to effectively influence a massive, panicked crowd presents a significant challenge. Most existing research relies on microscopic models that track every individual's position and velocity. While accurate, these models become computationally prohibitive as crowd size increases. Conversely, macroscopic models treat the crowd as a continuous flow, which scales well but often misses the nuances of individual human-robot interactions.

This paper addresses the challenge of indirect influence: robots do not physically push humans but provide local signals (e.g., "EXIT THIS WAY" signs) that humans respond to. This local response then propagates through the crowd, leading to a collective evacuation.

Two-Scale Modeling and Control

The core innovation is a two-scale hydrodynamic model that bridges the gap between individual behaviors and collective dynamics.

Microscopic to Macroscopic Derivation

The framework begins with "Social Force Models," which describe how individuals are repelled by obstacles and attracted to exits or robot-provided guidance.

Microscopic social force model illustrating how a human (red) is influenced by environmental forces and robot guidance (F).

Using a mathematical "hydrodynamic limit," these microscopic equations are transformed into macroscopic partial differential equations (PDEs). These PDEs describe the evolution of crowd density ($\rho$) and velocity ($u$) over time and space, treating the crowd as a fluid-like continuum.

Dual-Layer Robot Control

The robots are controlled through a two-layer strategy:

  1. Position Control: Robots utilize a potential-field-based coverage algorithm to spread out, ensuring their range of influence covers the human population.
  2. Direction Control: Using a backstepping technique, robots sense the current crowd density and adjust the direction of their navigation force fields (their "arrow signs") to "pull" the density toward the safe zone.
Macroscopic representation of crowd density, which the robots use to dynamically adjust their guidance directions.

To handle unknown environmental factors like moving obstacles, an RBF Neural Network is integrated into the controller, allowing robots to learn and compensate for unknown forces in real-time.

Key Insights and Results

The proposed approach offers several advantages:

  • Scalability: By controlling the "fluid" density rather than individuals, the algorithm's complexity depends on map resolution rather than the number of humans. In fact, the system performs better as the population increases, as the macroscopic approximation becomes more accurate.
  • Efficiency and Robustness: Simulations with 250 humans and only 10 robots achieved over a 90% evacuation rate. The system proved robust against dynamic obstacles and maintained real-time performance, with iterations taking only 40ms on standard hardware.
  • Stability: The authors provided rigorous mathematical proof (using Lyapunov stability) that the crowd density will converge to the target "safe" state as long as the robots' combined influence covers the area where humans are present.

Conclusion

This research demonstrates that global evacuation objectives can be achieved through local, indirect robot interactions. While the current model assumes a centralized "eye in the sky" for density estimation, future work will focus on decentralizing the system and testing in more complex, unstructured building environments. Additionally, incorporating psychological variables like human trust and panic levels remains an important avenue for future exploration.

Summary updated

Abstract

Emergency evacuation describes a complex situation involving time-critical decision-making by evacuees. Mobile robots are being actively explored as a potential solution to provide timely guidance. In this work, we study a robot-guided crowd evacuation problem where a small group of robots is used to guide a large human crowd to safe locations. The challenge lies in how to use micro-level human-robot interactions to indirectly influence a population that significantly outnumbers the robots to achieve the collective evacuation objective. To address the challenge, we follow a two-scale modeling strategy and explore hydrodynamic models, which consist of a family of microscopic social force models that describe how human movements are locally affected by other humans, the environment, and robots, and associated macroscopic equations for the temporal and spatial evolution of the crowd density and flow velocity. We design controllers for the robots such that they not only automatically explore the environment (with unknown dynamic obstacles) to cover it as much as possible, but also dynamically adjust the directions of their local navigation force fields based on the real-time macrostates of the crowd to guide the crowd to a safe location. We prove the stability of the proposed evacuation algorithm and conduct extensive simulations to investigate the performance of the algorithm with different combinations of human numbers, robot numbers, and obstacle settings.
BibTeX
@article{zheng2024multi,
  author = {Tongjia Zheng and Zhenyuan Yuan and Mollik Nayyar and Alan R. Wagner and Minghui Zhu and Hai Lin},
  title = {{Multi-Robot-Guided Crowd Evacuation: Two-Scale Modeling and Control Based on Mean-Field Hydrodynamic Models}},
  journal = {IEEE Trans. Control. Syst. Technol.},
  year = {2024},
  url = {https://openreview.net/forum?id=9nxnxSifQs},
}
Figure