Source-linked AI summary
A Distributed Control Framework for a Team of Unmanned Aerial Vehicles for Dynamic Wildfire Tracking
Huy X. Pham, Hung M. La, David Feil-Seifer, Matthew Deans
TL;DR
Limited information about fire-front evolution makes wildfire fighting dangerous, motivating UAV-based tracking. The paper proposes distributed control for UAV teams to cover and track spreading fires, and reports simulations in which UAVs spread out, follow expanding borders, and change altitude while covering the wildfire.
Problem
Wildfire fighting is dangerous, and a lack of information about the current state and dynamic evolution of fire contributes to accidents.
Source: §I · p. 1Method
The paper proposes decentralized UAV control to collaboratively track spreading fire boundaries without depending on a wildfire model, while sharing field vision and avoiding collisions.
Source: §I · p. 2Results
In 6000 Matlab simulation time steps, UAVs spread out over the wildfire and followed its expanding border regions while changing altitude.
Source: §V · p. 8Takeaways & Limitations
The authors conclude that the control design can track expanding wildfire borders while maintaining coverage, avoiding collisions, and supporting flexible deployment.
Source: §VI · p. 8Hardware implementation, changing network topology, and sensing endurance in hazardous environments remain topics for future work.
Source: §VI · p. 8
Abstract
from arXiv · showhide
Wildland fire fighting is a very dangerous job, and the lack of information of the fire front is one of main reasons that causes many accidents. Using unmanned aerial vehicle (UAV) to cover wildfire is promising because it can replace human in hazardous fire tracking and save operation costs significantly. 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 collision as well as cooperate well with other neighbors. Experimental results are conducted to demonstrate the capabilites of the UAV team in covering a spreading wildfire.
I. INTRODUCTION
Wildfire response is dangerous, and insufficient information about fire-front evolution contributes to accidents. The paper proposes distributed UAV control to track spreading boundaries while coordinating coverage and avoiding collisions.
I. INTRODUCTION
- Insufficient information about fire-front evolution contributes to accidents, while more information about spreading areas can support evacuation and protection strategies.The introduction reports an average of 70,000 wildfires annually burning around 7 million acres and destroying more than 2,600 structures.Source: §I · p. 1
- UAVs can assist hazardous fire tracking and replace manned helicopters while saving sizable operation costs compared with traditional methods.The paper notes that research on UAV assistance for firefighting remains limited.Source: §I · p. 1
- Prior approaches include UAV fire detection and cooperative fire-front sensing, as well as coordination architectures and checkpoint-based tracking.The introduction also identifies centralized decision-making in some approaches and model-shape dependence in another.Source: §I · pp. 1–2
- Most cited prior work does not address system behavior during fire spread; the proposed decentralized control tracks spreading boundaries without relying on a wildfire model.The framework also shares field vision, maintains safe distance to avoid collisions, and can increase image resolution at the fire border.Source: §I · p. 2
II. WILDFIRE MODELING
The paper simplifies wildfire propagation to provide a changing environment for evaluating multi-UAV control rather than to produce an accurate fire-growth model. It models front movement and fire intensity over time.
II. WILDFIRE MODELING
- The model assumes a discrete grid, known steady-state spreading rates for each grid, and propagation only at fire-front points.Source: §II · p. 2
- Fire-front movement uses an elliptical spread formulation, with wind direction and spreading rate informing the updated front location after time step δt.The paper simplifies the original differential equations to retain the center of the newly developed fire front.Source: §II · p. 2
- Fire intensity is modeled as a linear sum of multivariate-normal-like contributions from fire-front sources, with intensity decaying over time.Figure 3 depicts the simulated fire at successive time steps; brighter red indicates lower-intensity outer and boundary regions, while darker red indicates higher-intensity fire.Source: §II · pp. 2–3, Figure 3 · p. 3
- The fire model is intended to simulate a changing environment for UAV coverage; its accuracy is stated not to affect the distributed control algorithm's performance.The authors modify FARSITE propagation to focus on multi-UAV control rather than accurate fire-growth modeling.Source: §II · pp. 2–3
III. PROBLEM FORMULATION
The problem is to deploy UAVs to an estimated wildfire location, cover the fire and track its spreading fronts, and avoid in-flight collisions. Coverage should retain a whole-fire view while allowing higher-resolution observation near the boundary.
III. PROBLEM FORMULATION
- The team must reach the initially estimated wildfire location, spread out to cover the fire, track its developing fronts, and avoid in-flight collisions.The formulation assumes the fire occurs in a known forest section and its estimated location is reported to the UAVs.Source: §III · p. 3
- While tracking expanding fronts, some UAVs may lower altitude to improve boundary-image resolution as the team tries to maintain a complete view of the wildfire.Source: §III · p. 3
- Each UAV is modeled with a pose, a rectangular camera field of view, and access to communication with nearby UAVs within a radius r.The formulation defines pose as lateral coordinates and altitude, and assumes localization devices and downward-facing fire-detecting cameras.Source: §III · p. 3
A. Deployment objective
UAVs approach an estimated rendezvous point and continue until they detect the wildfire within their field of view. The deployment point can be served from forest depots or a firefighting center.
A. Deployment objective
- The rendezvous point is the initially estimated wildfire location toward which UAVs move after receiving a fire report.Source: A. Deployment objective · p. 3
- UAVs keep moving toward the rendezvous point until the wildfire is detectable inside their field of view.Source: A. Deployment objective · p. 3
- A UAV camera captures the area beneath its rectangular field of view.The field of view is specified by half-angles θ1 and θ2.Source: Figure 4 · p. 4
B. Coverage and tracking objective
The coverage objective uses camera field-of-view information to guide UAV coverage of the changing wildfire. It weights the fire border more heavily than the center, which may include regions at maximum intensity that are not tracking targets.
B. Coverage and tracking objective
- Because each camera has a limited pixel count, covering a larger field of view yields lower image resolution; the objective seeks higher-resolution fire-border images.The coverage formulation adapts an objective that minimizes information captured per pixel across UAVs.Source: B. Coverage and tracking objective · p. 4
- Two UAVs are coverage neighbors when their fields of view share one or more points.Source: B. Coverage and tracking objective · p. 4
- The UAV objective weights the fire-border region more heavily than the center, using detected heat intensity to characterize the regions.The cameras are assumed to detect intensity over a shared range.Source: B. Coverage and tracking objective · p. 4
- Some center regions at maximum intensity are treated as unimportant because they may burn out quickly and are not tracking targets.Source: B. Coverage and tracking objective · p. 4
C. Collision avoidance objective
Collision avoidance relies on nearby UAVs communicating while maintaining designed spacing. The framework distinguishes the communication range from the safe distance between team members.
C. Collision avoidance objective
- Each UAV communicates only with nearby UAVs inside communication range r.Source: Figure 5 · p. 4
- UAVs that satisfy the communication-range condition become physical neighbors, and must remain at least the designed distance d apart.Source: C. Collision avoidance objective · p. 5
- Figure 6 labels the distance between two UAVs with poses as d_ij.Source: Figure 6 · p. 5
D. Identify neighbors
The neighbor-identification scheme distinguishes physical neighbors, who can exchange signals within communication range, from sensing neighbors. The condition for sensing neighbors is necessary but not sufficient, and range selection balances connectivity against communication overload.
D. Identify neighbors
- Physical neighbors are UAVs within communication range, which allows them to exchange position signals.Source: D. Identify neighbors · p. 5
- Sensing neighbors must first be physical neighbors and also satisfy a geometric condition involving their fields of view.The stated condition is necessary; rectangular FOV geometry means it does not guarantee sensing-neighbor status.Source: D. Identify neighbors · p. 5
- Choose communication radius r large enough to connect UAVs with overlapping fields of view, but limit it to avoid communication overload from too many neighbors.Source: D. Identify neighbors · p. 5
IV. CONTROLLER DESIGN
The controller separates wildfire coverage and tracking from potential-field motion and collision avoidance. Its coverage component adjusts UAV positions using an objective-function gradient, including an altitude trade-off between field of view and fire-front resolution.
IV. CONTROLLER DESIGN
- The upper-level component sets wildfire coverage and tracking positions, while the lower-level potential-field component moves UAVs to those positions and avoids collisions.Source: §IV · p. 5
A. Coverage & tracking control
- Each UAV’s coverage and tracking controller uses negative-gradient descent on the objective function with respect to its pose.The resulting lateral-position and altitude controls use partial derivatives of the objective.Source: A. Coverage & tracking control · p. 5
- Figures 8 and 9 show ground field of view and UAV altitude, respectively, at t = 1000, 3000, 4000, and 6000.Source: Figure 8 · p. 6, Figure 9 · p. 6
- The lateral control moves toward wildfire regions with larger intensity change while reducing coverage overlap; altitude control trades larger FOV against better fire-front resolution.The desired virtual position is passed to the lower-level potential-field controller.Source: A. Coverage & tracking control · p. 6
B. Potential field control
The lower-level controller combines attraction toward rendezvous or virtual target positions with repulsion from nearby UAVs. A coverage gate switches the team from rendezvous travel to fire-front tracking.
B. Potential field control
- A quadratic distance potential pulls each UAV toward the initial rendezvous point, and its gradient yields the attractive force.Source: B. Potential field control · p. 6
- The coverage and tracking component supplies a desired virtual position pdi, which the UAV approaches using an attractive force.Source: B. Potential field control · p. 6
- Neighbor UAVs within the designed safe distance d generate repulsive forces, obtained from the gradient of their summed potential fields, to avoid collisions.The safe-distance condition is ||pj − pi|| < d.Source: B. Potential field control · p. 6
- The lower control component's control equation is derived from the preceding force and gating equations.Source: B. Potential field control · p. 6
- The coverage-control component is inactive during travel to the wildfire region and activates upon arrival in the waypoint region, allowing fire-front tracking to replace rendezvous attraction.The passage states that ζi is 0 during travel and becomes 1 when the UAVs reach a region where they can sense the fire.Source: B. Potential field control · p. 6
V. SIMULATION
In Matlab simulations, the UAVs spread out over the wildfire and followed its expanding border while changing altitude and field of view. Some maximum-intensity regions remained uncovered.
V. SIMULATION
- The simulation began with 10 ground-based UAVs, a rendezvous point at [500, 500, 60]T, a 30-unit safe distance, and a 100-unit communication range.The wildfire started from five fire-front points near [500, 500]T.Source: §V · p. 8
- Across 6,000 Matlab time steps, UAVs spread out to cover the wildfire and followed its expanding border regions.They changed altitude from about 60 to different altitudes as the fire developed, producing different FOV areas.Source: §V · p. 8
- Some regions at maximum fire intensity remained uncovered, while UAV altitude and field of view varied with local intensity.A UAV covering a low-intensity region could fly at low altitude.Source: §V · p. 8
VI. CONCLUSION
The authors conclude that their distributed UAV controller can track expanding wildfire borders while maintaining coverage and avoiding collisions. They identify hardware implementation and related operational questions as future work.
VI. CONCLUSION
- The distributed design lets UAVs collaboratively follow an expanding wildfire border while trying to maintain coverage and avoid collisions.The authors also describe the team as flexible in deployment.Source: §VI · p. 8
- The authors propose extending the approach to other dynamic environments, including oil spills and floods.Source: §VI · p. 8
- Future work includes hardware implementation, communication under changing network topology, sensing endurance in hazardous environments, and synchronization with wildfire spread speed.The authors also mention multi-drone cooperative sensing and cooperative learning for wildland-fire mapping.Source: §VI · p. 8