Efficient Decentralized Sequential Planner for Spatiotemporal Wildfire Mitigation Using Multiple UAVs

Academic Background

Wildfires pose a significant threat to global biodiversity and resource sustainability, especially in their early stages. If not controlled in time, wildfires can rapidly expand, leading to severe ecological damage. In recent years, multi-UAV systems (Unmanned Aerial Vehicles, UAVs) have increasingly been used in wildfire management, primarily to reduce human exposure to hazardous environments and improve emergency response efficiency. However, existing research mostly focuses on single aspects such as search, monitoring, or firefighting, lacking comprehensive studies on multi-UAV cooperative tasks. Efficient task allocation for early wildfire mitigation in resource-limited, partially observable environments with dynamically changing wildfires remains a complex and challenging issue.

This paper proposes a Conflict-aware Resource-efficient Decentralized Sequential Planner (CREDS), which aims to mitigate dynamically spreading wildfires through a team of heterogeneous UAVs, maximizing resource utilization efficiency and minimizing biodiversity loss.

Source of the Paper

This paper is co-authored by Josy John, Shridhar Velhal, and Suresh Sundaram, all affiliated with the Department of Aerospace Engineering at the Indian Institute of Science. The article was published in IEEE Transactions on Automation Science and Engineering in 2025.

Research Process

1. Problem Modeling and Task Assignment

CREDS models the wildfire mitigation problem as a decentralized spatiotemporal task allocation problem, aiming to maximize the success rate of single-UAV tasks while minimizing ecological damage caused by fires. The study assumes that wildfires have dynamically expanding characteristics, and the limited sensing range of UAVs results in a partially observable environment. To address this, CREDS adopts a three-phase framework: Search Phase, Local Trajectory Generation Phase, and Conflict Resolution Phase.

a) Search Phase

UAVs use the Oxyhrris Marina-inspired Search Algorithm (OMS) to search for fires within the mission area. OMS is a multi-level search algorithm that mimics the foraging behavior of marine plankton. When the temperature is below the threshold, UAVs employ Levy search for exploration; when the temperature exceeds the threshold, they use Brownian motion search. When the infrared camera detects a fire, the fire’s location, area, and spread rate are recorded in the UAV’s detection list.

b) Local Trajectory Generation Phase

CREDS uses the Resource-efficient Decentralized Sequential Planner (REDS) to generate local firefighting trajectories for UAVs. REDS introduces a novel Deadline-prioritized Mitigation Cost (DPMC), which efficiently allocates tasks based on fire deadlines. DPMC cost includes temporal deadline cost and task start time cost, the former ensuring task completion before the deadline and the latter minimizing losses caused by fire spread.

c) Conflict Resolution Phase

Tasks generated locally by UAVs may conflict. To resolve this, CREDS employs a Conflict-aware Consensus Algorithm, which synchronizes information with other UAVs via a communication network and assigns tasks based on the lowest cost, ultimately generating a conflict-free global trajectory.

2. Experiments and Performance Evaluation

The study evaluates CREDS’ performance under partial and full observability conditions through Monte Carlo simulations, testing both heterogeneous and homogeneous UAV teams under different fire-to-UAV ratios.

a) Homogeneous UAV Teams

In homogeneous UAV teams, all UAVs have the same speed and firefighting capabilities. Experimental results show that CREDS achieved a 100% success rate for fire-to-UAV ratios up to 4. Even at a fire-to-UAV ratio of 5, CREDS’ success rate was significantly higher than baseline methods.

b) Heterogeneous UAV Teams

In heterogeneous UAV teams, UAVs have varying speeds and firefighting capabilities. Experimental results demonstrate that CREDS performs better in handling tasks with heterogeneous deadlines, achieving an 84% success rate at a fire-to-UAV ratio of 5, significantly higher than the baseline method’s 67%.

3. Scalability and Convergence

The study further evaluates CREDS’ scalability and convergence. Results show that as firefighting capabilities improve, CREDS’ success rate increases significantly for higher fire-to-UAV ratios. Additionally, CREDS achieved 100% convergence in all test scenarios, requiring fewer iterations than the baseline method.

Research Findings

1. Task Success Rate: In extreme cases with a fire-to-UAV ratio of 5, CREDS’ heterogeneous UAV team achieved an 84% success rate, 17% higher than the baseline method.
2. Convergence Rate and Iteration Count: CREDS achieved 100% convergence in all test scenarios, with an average iteration count 26.3% lower than the baseline method.
3. Firefighting Time and Fire Expansion Ratio: CREDS significantly reduced the total firefighting time and fire expansion ratio, demonstrating its efficiency in minimizing ecological losses.

Conclusions and Significance

This paper proposes CREDS, an efficient decentralized sequential planning method that enables multi-UAV cooperation for early wildfire mitigation in resource-limited scenarios. Its innovations include: 1. Task Allocation Strategy: Through the DPMC cost function, CREDS efficiently allocates tasks in dynamic environments, ensuring task completion before deadlines and minimizing fire spread. 2. Advantages of Heterogeneous Teams: Heterogeneous UAV teams perform better in handling tasks with varying deadlines, further improving task success rates. 3. Scalability and Convergence: CREDS demonstrates excellent scalability in resource-limited scenarios and achieves rapid convergence in all test cases.

Research Highlights

1. Efficient Task Allocation: CREDS achieves efficient task allocation through the DPMC cost function, significantly improving task success rates.
2. Excellent Performance in Partially Observable Environments: CREDS maintains high task success and convergence rates even in partially observable environments.
3. Practical Application Value: CREDS’ efficiency and scalability make it highly applicable in real-world wildfire mitigation scenarios.

CREDS provides an innovative solution to the complex problem of multi-UAV wildfire mitigation, offering significant scientific value and application potential.