A Hybrid Three Layer Architecture for Fire Agent Management in Rescue Simulation Environment

Alborz Geramifard, Peyman Nayeri, Reza Zamani-Nasab, Jafar HabibiarXiv 2006 · OpenReview

Abstract

Disaster response simulations present a "wicked" challenge for artificial intelligence, requiring agents to operate in high-dimensional, dynamic environments with limited communication. This paper introduces the Fire Agent Integrated System (FAIS), a three-layer hybrid architecture that manages autonomous fire-fighting agents. By combining low-level heuristics with high-level structured decision-making and a "Core" safety layer, FAIS successfully balances individual agent autonomy with global team coordination. The architecture's efficacy was validated during the 2003 RoboCup Rescue competition, where the Arian team secured a first-place ranking.

Introduction

In a simulated city engulfed in fire, agents such as Fire Brigades and Fire Stations must make critical decisions under pressure. Purely heuristic systems often lack the sophistication for complex team coordination, while purely structured mathematical models can become computationally overwhelmed by the chaotic nature of disaster zones. The FAIS architecture addresses this gap by creating a system that processes massive sensory data into actionable, coordinated plans without excessive complexity.

The FAIS Architecture

The authors propose organizing agent logic into three distinct layers to manage the transition from raw data to high-level strategy.

The three-layer hybrid architecture of FAIS.

1. Heuristic Module (Lower Layer)

This layer serves as a data compressor and "common sense" engine. It handles fundamental tasks such as path planning (using Breadth-First Search), collision detection, and monitoring internal agent states (e.g., water levels and injury status). By transforming raw sensory data into a simplified, abstract world model, it prunes the search space for higher-level modules.

2. Structured Decision Making (Middle Layer)

The middle layer utilizes the abstracted model to make high-level strategic choices. For the Fire Station coordinator, this involves calculating a Building Value ($V(b)$) for every fire using a weighted formula:

$$V(b) = \beta N(b) + \gamma HV(b) - \alpha \sum d_i(b)$$

Where $N(b)$ is the number of unfired neighbors (potential for spread), $HV(b)$ is the "Harm Value" based on the building's location, and $d_i(b)$ is the distance to available agents. A Linear Programming (LP) solver then optimally assigns agents to buildings to minimize total travel distance.

3. Core Layer (Top Layer)

The Core Layer acts as a final safety filter. It applies predefined rules to handle edge cases where underlying algorithms might fail—for instance, when an agent is trapped or when no fires are currently visible.

Key Insights and Strategies

A defining feature of the FAIS approach is the "Critical Section" Strategy. Recognizing that early fire suppression prevents exponential damage spread, the system implements logic that prioritizes aggressive, immediate action at the start of a simulation, overriding long-term optimal planning.

Furthermore, the architecture emphasizes Centralized Advice with Decentralized Action. While Fire Brigades maintain autonomy, the Fire Station acts as a central decision-maker. This prevents redundant efforts and ensures vital areas are not ignored.

Comparison of agent distribution: centralized coordination (bottom) vs. decentralized/uncoordinated (top).

The system also includes specific Traffic Control logic. By monitoring mission assignments, the Fire Station provides "traffic advice" to prevent road blockages, a common bottleneck in disaster simulations.

Evaluation and Results

The FAIS architecture was implemented by the Arian team for the 2003 RoboCup Rescue competition. The results highlighted a significant performance advantage:

  • Competition Ranking: The Arian team placed 1st in Round 1.
  • Quantitative Performance: Arian achieved a score of 62.35, significantly outperforming the runner-up, S.O.S. (56.08), and nearly doubling the scores of mid-tier teams.
  • Efficiency: Centralized decision-making led to more efficient water consumption and superior agent distribution compared to purely decentralized logic.

Conclusion

The FAIS architecture demonstrates that a hybrid approach—combining heuristics, structured optimization, and rule-based safety layers—is highly effective for complex multi-agent coordination in disaster scenarios. Future work involves extending this framework to include police and ambulance agents for a fully integrated response team.

Summary updated

Abstract

This paper presents a new architecture called FAIS for imple- menting intelligent agents cooperating in a special Multi Agent environ- ment, namely the RoboCup Rescue Simulation System. This is a layered architecture which is customized for solving fire extinguishing problem. Structural decision making algorithms are combined with heuristic ones in this model, so it's a hybrid architecture.
BibTeX
@misc{geramifard2006hybrid,
  author = {Alborz Geramifard and Peyman Nayeri and Reza Zamani-Nasab and Jafar Habibi},
  title = {{A Hybrid Three Layer Architecture for Fire Agent Management in Rescue Simulation Environment}},
  year = {2006},
  eprint = {cs/0601055},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/cs/0601055},
}
Figure