Multi-Agent Systems for Disaster Response

Overview

This survey investigates how Multi-Agent Systems (MAS) can support collaborative decision-making and resource coordination in complex disaster response. Agents provide a way to represent responders, resources, or decision units that act with different capabilities and information.

I compared six complementary studies and organized their methods into a proposed conceptual framework: risk assessment, resource allocation, dynamic dispatching, task scheduling, and simulation-based validation. The contribution is the synthesis: identifying how these decision layers relate, where their assumptions differ, and what their integration would require.

Developed as part of a graduate-level Multi-Agent Systems course voluntarily undertaken during my undergraduate studies.

Research form

Literature review

Studies compared

Six perspectives

Synthesis

Five decision stages

Research Challenges

My interest in this research is shaped by Taiwan's disaster-response challenges. Its mountainous terrain and geographical complexity, together with frequent earthquakes and typhoons, make access and coordination important concerns. Limited emergency personnel and equipment can become harder to deploy when transport routes are disrupted and communications infrastructure is damaged. Taiwan's strategic geographical position and regional geopolitical tensions further reinforce my interest in infrastructure resilience and emergency preparedness.

These considerations led me to explore how Multi-Agent Systems could support collaborative disaster response through decentralized coordination and adaptive resource allocation. I am interested in how distributed response units might coordinate when information is incomplete, access changes, and resources must be reassigned. Taiwan provides the motivation and a potential application context for this inquiry; the findings discussed below come from the reviewed studies and their respective settings. The survey examines what their methods could contribute, alongside the information exchange, computation, and synchronization that coordination requires.

Objectives guiding the review

• Understand how game-theoretic, coalition, and market mechanisms coordinate changing task demands under time and information constraints.
• Compare the flexibility and resilience sought in distributed coordination with the planning and consistency available from centralized decisions.
• Identify computational, modeling, and evaluation limitations, then examine how complementary methods could be connected.

Methodology & Literature Review

The review uses a requirements-oriented comparison. It begins with the response problem, separates functional needs such as prediction and assignment from qualities such as resilience and scalability, and compares each study by decision architecture, prediction capability, resource constraints, scheduling, and evaluation setting.

The six selected studies form a focused narrative review. Their problems and evaluation settings differ, so the analysis compares methodological roles and assumptions rather than treating their results as a common benchmark.

Expand a study for its problem, approach, contribution to the framework, and limitations. Reference numbers stay consistent throughout the page.

Integrated MAS Framework

The framework gives the six studies a common analytical structure. Prediction and threat assessment inform strategic resource decisions; dispatch and scheduling translate those decisions into coordinated actions; simulation offers a setting in which to examine their behavior.

The architectural insight: placement and dispatch are different decisions

Predictive allocation asks where resources should be positioned before incidents occur [5]. Dynamic dispatch asks which responder should act given the current state [6]. Treating these as connected but distinct layers makes their different time horizons visible. Coalition and market-based scheduling add the feasibility constraints that a placement or assignment alone cannot resolve [2] [3].

Integrating the layers would require compatible representations of incidents, priorities, responder availability, travel, and time. In this survey, these are relationships to investigate rather than interfaces assumed to exist between the original methods.

Key Findings & Research Insights

Findings in the reviewed literature

• Allocation quality includes coordination. The game-theoretic study connects threats and contribution-based sharing [1]; coalition and market approaches make cooperation, deadlines, and task order explicit [2] [3].
• Adaptive decisions need state and time. Continuous-time prediction supports prioritization and placement [5], while sequential dispatch accounts for incident arrivals and service completion [6].
• Evaluation is a modeling problem too. DrillSim makes participant roles and behavior part of the evaluation environment [4]. Evidence remains tied to the scenarios, information, and objectives used.

Trade-offs and limits

Resource scarcity creates competing objectives: coverage, urgency, balanced workloads, and travel effort may favor different decisions. The survey identifies increasing computational demands as agents and tasks grow, alongside reliance on forecasts and resource-state data. Heuristics can make decisions tractable, while changing what can be claimed about optimality.

The studies span threat allocation, law enforcement, urban incident response, and disaster simulation. They offer complementary evidence, but their reported results do not constitute a joint evaluation of the integrated framework. Communication-constrained operation and transfer from simulation to field conditions remain important research questions.

Future Directions

A useful next step would turn the conceptual connections into testable interfaces and experiments:

• Build a modular prototype. Define a shared incident and responder model, then connect prediction, allocation, dispatch, and scheduling as replaceable policies.
• Compare coordination strategies. Evaluate centralized and decentralized variants under the same incident streams and resource budgets.
• Study scale and uncertainty. Vary agent counts, incident density, partial observability, and forecast quality; examine response delays, coverage, and computational cost.
• Stress communication. Simulate delayed messages, disconnected units, and infrastructure failures to investigate coordination recovery and conflicting assignments.
• Ground scenarios in realistic data. Assess suitable disaster datasets and operational assumptions before extending the framework to more realistic exercises.

The survey's limitations shape this agenda: a focused six-study selection, heterogeneous evidence, expensive decision models, uncertain forecasts, and the gap between simulation and operational conditions. A more systematic review and reproducible scenario comparisons would strengthen the evidence for future integration.

References

The research repository and its reference guide document the survey and its synthesis. Original publication links below support the methodological descriptions; no figures from the papers are reproduced.

1. A Game Theory Approach for Multi-agent System Resources Allocation against Outside ThreatsC.-K. Wu and X. Hu. Journal of Risk Analysis and Crisis Response · 9(3), 2019 issue.
2. Multi-Agent Routing and Scheduling Through Coalition FormationL. Capezzuto, D. Tarapore, and S. D. Ramchurn. arXiv:2105.00451 · 2021.
3. Dynamic Multi-Agent Task Allocation with Spatial and Temporal ConstraintsS. Amador, S. Okamoto, and R. Zivan. AAAI · 2014.
4. Multi-Agent Simulation of Disaster ResponseD. Massaguer, V. Balasubramanian, S. Mehrotra, and N. Venkatasubramanian. Agent Technology for Disaster Management workshop · 2006.
5. Prioritized Allocation of Emergency Responders based on a Continuous-Time Incident Prediction ModelA. Mukhopadhyay, Y. Vorobeychik, A. Dubey, and G. Biswas. AAMAS · 2017.
6. A Decision Theoretic Framework for Emergency Responder DispatchA. Mukhopadhyay, Z. Wang, and Y. Vorobeychik. AAMAS · Industrial Applications Track · 2018.

Research topics

• Multi-Agent Systems
• Resource Allocation
• Game Theory
• Task Scheduling
• Sequential Decisions
• Research Synthesis