Header Search Widget
SOcietal-scale systems: Learning, Incentives, and Dynamics
Home > Coordinated Autonomous Drones for Human-Centered Fire Evacuation in Partially Observable Urban Environments

Coordinated Autonomous Drones for Human-Centered Fire Evacuation in Partially Observable Urban Environments

Autonomous drone technology holds significant promise for enhancing search and rescue operations during evacuations by guiding humans toward safety and supporting broader emergency response efforts. However, their application in dynamic, real-time evacuation support remains limited. Existing models often overlook the psychological and emotional complexity of human behavior under extreme stress. In real-world fire scenarios, evacuees frequently deviate from designated safe routes due to panic and uncertainty.
To address these challenges, this paper presents a multi-agent coordination framework in which autonomous Unmanned Aerial Vehicles (UAVs) assist human evacuees in real-time by locating, intercepting, and guiding them to safety under uncertain conditions. We model the problem as a Partially Observable Markov Decision Process (POMDP), where two heterogeneous UAV agents—a high-level rescuer (HLR) and a low-level rescuer (LLR)—coordinate through shared observations and complementary capabilities. Human behavior is
captured using an agent-based model grounded in empirical psychology, where panic dynamically affects decision-making and movement in response to environmental stimuli. The environment features stochastic fire spread, unknown evacuee locations, and limited visibility, requiring UAVs to plan over long horizons to search for a human and adapt in real time. Our framework employs the Proximal Policy
Optimization (PPO) algorithm with recurrent policies to enable robust decision-making in partially observable settings. Simulation results demonstrate that the UAV team can rapidly locate and intercept evacuees, significantly reducing the time required for them to reach safety compared to scenarios without UAV assistance.

Problem

Wildfires increasingly threaten urban populations, where dense infrastructure and limited escape routes make timely evacuation critical. Yet, current evacuation systems largely overlook the human side of the problem—how panic, confusion, and poor visibility impair people’s ability to reach safe zones. While UAVs are already used for post-disaster mapping and monitoring, they are rarely deployed in real-time to actively assist evacuees. Existing models often assume rational human behavior and full observability, ignoring the psychological complexity of disaster scenarios. This work addresses that gap by designing a coordinated team of drones that can locate, intercept, and guide evacuees to safety—despite unknown human locations, evolving fire hazards, and panic-induced behavior.

Challenges

This problem presents several key challenges that must be overcome for UAVs to effectively assist during fire evacuations:

  • C1 – Human Behavior under Panic: Evacuees often act irrationally under stress, deviating from safe routes due to fear, limited information, or emotional overwhelm.
  • C2 – Long-Term Planning under Uncertainty: UAVs must make decisions without full knowledge of the environment or evacuee location, requiring planning over long horizons.
  • C3 – Time-Varying Environment: Fires evolve stochastically and unpredictably, affecting both human and UAV actions in real time.
  • C4 – Multi-Agent Coordination: The two UAVs must share partial observations and coordinate to successfully locate and intercept the evacuee.
  • C5 – Heterogeneous Agent Capabilities: The high-level and low-level drones have different sensing, movement, and interception capabilities, complicating joint policy learning.
  • C6 – Unknown Evacuee Location: The rescuers must search for the evacuee from scratch without prior location information, while avoiding wasted effort or missed opportunities.

Approach

We consider an urban environment affected by fire, involving three agents: a human agent (referred to as the evacuee), who aims to reach a designated safe zone while avoiding the fire, and a team of UAVs (referred to as rescuers), whose objective is to locate and guide the evacuee to safety. The rescuers operate collaboratively, with one responsible for high-level surveillance and the other for ground-level interception. We describe each rescuer below:

  • High-Level Rescuer (HLR): Operates at high altitude with a wide field of view. It maps the environment, infers the evacuee’s location, and tracks their movement using a downward-facing camera. However, it cannot physically intercept the evacuee.
  • Low-Level Rescuer (LLR): Operates at low altitude with a narrower field of view. It is the only agent capable of physically intercepting the evacuee. The LLR navigates closer to the ground and detects obstacles that may be hidden from the HLR, using a front-facing camera

Urban environment modeled as a discrete-time, 2D grid world, represented by a tuple G = (X , A, O, B)

Agent-Based Modeling with Panic Behavior

More details coming soon…