Source-linked AI summary
A Distributed Control Framework of Multiple Unmanned Aerial Vehicles for Dynamic Wildfire Tracking
Huy Xuan Pham, Hung Manh La, David Feil-Seifer, Matthew Dean
TL;DR
Wildfire response needs continuous fire-front information, but existing approaches can be centralized, static-environment oriented, or dependent on modeled fire shapes. This paper develops decentralized control for UAV teams that track expanding fire boundaries while maintaining coverage and safety. Simulations demonstrate wildfire coverage, boundary following, collision avoidance, altitude adaptation, and cooperation with neighbors.
Problem
Wildfire tracking needs continuous information about evolving fire fronts, while prior approaches may centralize decisions, assume static environments, or depend on modeled fire shapes.
Method
The paper proposes decentralized multi-UAV control that tracks spreading fire boundaries without dependency on wildfire modeling while coordinating coverage, resolution, communication, collision avoidance, and altitude.
Results
Simulations show UAVs can cover spreading wildfires, follow fire borders, adapt altitude and field of view, maintain safe separation, and cooperate with physical and sensing neighbors.
Takeaways & Limitations
The framework supports distributed monitoring of changing wildfire environments while preserving whole-fire coverage and emphasizing higher-resolution observation near fire boundaries.
Abstract
from arXiv · showhide
Wild-land fire fighting is a hazardous job. A key task for firefighters is to observe the "fire front" to chart the progress of the fire and areas that will likely spread next. Lack of information of the fire front causes many accidents. Using Unmanned Aerial Vehicles (UAVs) to cover wildfire is promising because it can replace humans in hazardous fire tracking and significantly reduce operation costs. In this paper we propose a distributed control framework designed for a team of UAVs that can closely monitor a wildfire in open space, and precisely track its development. The UAV team, designed for flexible deployment, can effectively avoid in-flight collisions and cooperate well with neighbors. They can maintain a certain height level to the ground for safe flight above fire. Experimental results are conducted to demonstrate the capabilities of the UAV team in covering a spreading wildfire.
I. INTRODUCTION
Wildfires are dangerous and time-sensitive because incomplete information about fire-front evolution can endanger firefighters and hinder evacuation planning. The paper motivates distributed multi-UAV control to monitor spreading fires while avoiding the scalability and modeling limitations of prior approaches.
- A. Wildfire monitoring and tracking: Wildfires require continuous fire-front information because unexpected propagation can endanger firefighters and complicate evacuation and containment planning.
- A. Wildfire monitoring and tracking: UAVs can assist hazardous fire tracking and reduce operating costs compared with manned helicopters, but wildfire-focused UAV research remains limited.
- B. Multiple robots in sensing coverage: Prior multi-UAV work includes fire mapping, disaster-management decision frameworks, and perimeter tracking, but often relies on centralized decisions or modeled fire shapes.
- B. Multiple robots in sensing coverage: Most existing sensing-coverage methods assume static environments, whereas wildfire monitoring requires coverage of a changing environment.
- B. Multiple robots in sensing coverage: The paper proposes decentralized UAV control that tracks spreading fire boundaries without wildfire-model dependency while sharing coverage, avoiding collisions, increasing border resolution, and maintaining safe altitude.
II. WILDFIRE MODELING
The paper models wildfire evolution on a discrete grid using simplified fire-front propagation and intensity fields. The simplification prioritizes evaluating distributed UAV control while allowing alternative spreading models without changing the controller’s intended operation.
- II. WILDFIRE MODELING: The wildfire model uses a discrete grid, precomputed steady-state spreading rates, and propagation of only fire-front points.
- II. WILDFIRE MODELING: The fire-front geometry is simplified from elliptical propagation equations to retain the center of each newly developed fire front.
- II. WILDFIRE MODELING: Fire-front displacement uses Finney’s equations, with spreading determined by the steady-state rate and wind-related quantities.
- II. WILDFIRE MODELING: Each heat source contributes a multivariate-normal-like intensity field, and total wildfire intensity is the linear sum of contributions from k sources.
- II. WILDFIRE MODELING: The controller is intended to remain effective when the wildfire spreading model changes, although the model determines the resulting fire shape.
III. PROBLEM FORMULATION
The problem formulation asks multiple UAVs to deploy to a known wildfire location, cover and track its evolving fire fronts, and maintain collision and fire safety. UAVs use camera fields of view, local communication, and altitude adjustments to balance coverage with boundary resolution.
- III. PROBLEM FORMULATION: The team must deploy from its depot to the estimated wildfire location before beginning collaborative coverage and tracking.
- III. PROBLEM FORMULATION: UAVs spread out to cover the wildfire, follow expanding fronts, lower altitude for boundary resolution, and preserve a complete view of the fire.
- III. PROBLEM FORMULATION: The formulation requires collision avoidance and a ground-distance safety margin, using localization and downward-facing rectangular-FOV cameras.
- III. PROBLEM FORMULATION: Each UAV pose contains lateral coordinates and altitude, while its visible region is represented by the points inside its rectangular field of view.
- III. PROBLEM FORMULATION: The rectangular camera field of view is bounded by four edges with outward-facing normal vectors and records the covered area as image pixels.
- III. PROBLEM FORMULATION: UAVs communicate with nearby vehicles within a communication range and maintain a designed safe separation while identifying physical and sensing neighbors.
A. Deployment objective
The deployment objective sends UAVs toward a rendezvous point and formulates wildfire coverage as a time-varying, resolution-aware sensing problem. The framework prioritizes fire-front regions while balancing image quality, overlap, communication, collision avoidance, and altitude safety.
- Deployment: UAVs move toward an initially estimated rendezvous point until the wildfire enters their camera fields of view.The rendezvous point can be supplied by distributed forest depots or a firefighting department center.
- Safety and coordination: UAVs coordinate through nearby physical neighbors, maintain a designed separation distance, and enforce a minimum altitude to avoid collisions and fire exposure.Neighbor identification depends on communication range, while altitude safety is represented by zmin.
- Coverage objective: The formulation adapts optimal coverage to a wildfire region Q(t) that changes over time, while treating intensity as locally constant because fire spread is assumed slower than UAV motion.This assumption supports computing coverage over short time periods.
- Coverage objective: Higher altitude increases coverage cost and lowers image resolution, creating a sensing trade-off that motivates altitude adjustment during tracking.The image-quality cost is derived from camera geometry and field-of-view area.
- Coverage objective: The coverage objective weights fire-front regions more heavily than intensely burning centers using infrared heat-intensity measurements.The importance function is based on detected intensity within the cameras' range, with the border assigned greater tracking importance.
IV. CONTROLLER DESIGN
The controller architecture divides each UAV's operation into coverage-and-tracking position generation and potential-field motion control. This separation lets UAVs pursue wildfire coverage while avoiding collisions and maintaining ground clearance.
- IV. CONTROLLER DESIGN: Each UAV controller combines a coverage-and-tracking component with a potential-field component.The first computes desired positions for wildfire monitoring; the second executes those positions while enforcing safety behaviors.
- IV. CONTROLLER DESIGN: The potential-field component moves UAVs toward desired positions while avoiding other UAVs and maintaining a safe distance from the ground.The coverage-and-tracking component updates the desired position after the UAV reaches the wildfire region.
A. Coverage & tracking control
The coverage-and-tracking controller updates each UAV's desired pose by descending the wildfire coverage objective. Its lateral motion reduces poorly covered regions and overlap, while altitude trades field of view against fire-front image resolution.
- A. Coverage & tracking control: Each UAV uses negative gradient descent of objective H with respect to its pose to compute distributed coverage-and-tracking control.The pose includes lateral position and altitude, and the resulting desired virtual position is passed to potential-field control.
- A. Coverage & tracking control: The lateral derivative moves UAVs toward wildfire areas with larger intensity variation while reducing coverage intersections with neighboring UAVs.The neighbor contribution is computed from the coverage-neighbor set excluding the current UAV.
- A. Coverage & tracking control: The altitude derivative trades larger wildfire field of view against improved fire-front propagation resolution.Lower altitude improves border resolution, whereas higher altitude covers a larger field of view.
- A. Coverage & tracking control: The controller estimates changing wildfire coverage by discretizing each UAV's field of view and checking sampled-point fire intensity with its detection system.The computation assumes the paper's environment-intensity model holds.
B. Potential field control
The potential-field controller converts desired coverage positions into safe UAV motion. It combines attraction to rendezvous and desired positions with repulsion from nearby UAVs and the ground, switching from rendezvous attraction to fire tracking once the fire is sensed.
- B. Potential field control: An attractive potential-field force moves each UAV toward the desired virtual position produced by the coverage-and-tracking component.This force is combined with the other potential-field terms in the final motion update.
- B. Potential field control: Repulsive forces push a UAV away from neighboring UAVs when their separation falls below the designed safe distance d.The repulsion is obtained from the gradient of the summed neighbor potential fields.
- B. Potential field control: A ground-repulsion term enforces a minimum altitude zmin by activating when the UAV approaches its ground image too closely.The ground safety condition is incorporated into the general potential-field control law.
- B. Potential field control: The general control law combines rendezvous, desired-position, neighbor-repulsion, ground-safety, and fire-detection-dependent potential terms.The resulting controller updates the UAV position through its control input.
- B. Potential field control: The controller initially attracts UAVs to the rendezvous point, then cancels that attraction when fire sensing begins so they can track growing fire fronts.The switching variable ζi is zero during transit and becomes one in the sensing waypoint region.
C. Stability analysis
The stability analysis uses Lyapunov functions to show that the coverage, tracking, and potential-field controllers converge to stable equilibria.
- The coverage and tracking controller is analyzed with H as a positive-definite Lyapunov candidate, where H represents weighted total FOV area.The analysis states that H is positive definite for all UAV positions.
- The derivative satisfies V̇ = 0 only at equilibrium positions corresponding to local minima of H.
- The equilibrium point is asymptotically stable according to Lyapunov’s stability theorem.
- Both attractive and repulsive potential-field phases are shown to be asymptotically stable under the stated equilibrium conditions.The proof uses positive-definite, radially unbounded Lyapunov candidates for the two phases.
D. Overall algorithm for decentralized control of UAVs
The decentralized algorithm combines local sensing and communication with discretized FOV processing to update UAV positions for wildfire tracking and coverage.
- Each UAV uses localization, neighbor communication, and heat-intensity sensing within its FOV to compute a new position.
- Simulation figures show UAV FOVs following newly developed fire fronts while altitude changes produce different coverage areas.
- FOV integrals are converted into sums over small particles after discretizing the FOV and its edges.
- The algorithm discretizes each rectangular FOV into interior points and four edge sets before evaluating their relation to the fire.
- The controller also represents UAV trajectories in three dimensions during tracking of a wildfire spreading north-east.
V. SIMULATION
The simulations evaluate ten UAVs tracking a spreading wildfire in MATLAB under specified sensing, safety, communication, and fire-spread settings.
- The simulation starts 10 UAVs on the ground near [300, 300]T and uses safe distance d = 10 and safe altitude zmin = 15.
- Each UAV uses identical cameras with focal length b = 10, pixel area S1 = 10^-4, half-angles θ1 = 30° and θ2 = 45°.
- The UAV communication range is r = 500, and the study tests both border-focused and non-focused coverage scenarios.
A. Scenario: Wildfire coverage with specific focus on border of the fire
The simulations show UAV teams covering and tracking a spreading wildfire, with border-focused control adapting altitude and FOV while normal coverage maintains broad coverage without border prioritization.
- A. Scenario: Wildfire coverage with specific focus on border of the fire: Across 6000 time steps, UAVs spread from a rendezvous point, fragment as the fire expands, and follow its border regions.The initial rendezvous point is pr = [500, 500, 60]T.
- A. Scenario: Wildfire coverage with specific focus on border of the fire: UAVs change altitude from zi ≈60 to different levels, producing different FOV areas while attempting to follow fire-front propagation.
- B. Scenario: Normal wildfire coverage: In normal coverage, the UAVs covered the spreading fire with no uncovered space, and their altitudes were almost equal.
- B. Scenario: Normal wildfire coverage: The two scenarios use the same controller parameters, but normal coverage sets φ(q) = κ because it does not prioritize higher-resolution border images.