Ph.D. (Engg): Development of Scalable UAV Swarm-based Cooperative Search and Mitigation Approaches for Wildfire Management

Climate change has significantly exacerbated the wildfire seasons, increasing their frequency, duration, and scale of destruction. Globally, wildfires destroy approximately 400 million hectares of land annually, resulting in significant biodiversity loss, degradation of soil nutrients, and other ecological consequences. The fire locations are often inaccessible for ground-based interventions due to the challenging terrain, and current human-centered firefighting strategies are both dangerous and unreliable, primarily due to limited situational awareness of evolving wildfire scenarios. Additionally, wildfire scenarios frequently involve rapidly spreading clusters of fires that surpass available resources. The wildfire scenarios also have large fires that require simultaneous action from multiple resources for mitigation. Unmanned Aerial Vehicles (UAVs) have emerged as an effective solution for enhancing situational awareness and facilitating interventions during wildfires. This thesis develops UAV swarm-based strategies for wildfire detection, monitoring, and mitigation in resource-constrained and dynamic environments.
The thesis first focuses on the early mitigation of clustered fires by assigning and scheduling firefighting UAVs under resource limitations. The objective is to reduce biodiversity loss through early mitigation of fires as Single UAV Tasks (SUTs) before they escalate into complex multi-UAV coordination tasks. The problem is reformulated as a shortest-schedule-route optimization and solved using two centralized approaches: Genetic Algorithm-based Routing and Scheduling with Time Constraints (GARST) and Hybrid Particle Swarm Optimization-based Routing and Scheduling with Time Constraints (HPSO-RST). GARST and HPSO-RST evaluated on homogeneous and heterogeneous UAV teams under full observability conditions show that HPSO-RST outperforms GARST, with a higher success rate, reduced mean fitness values, and minimized burned areas. However, the centralized nature of GARST and HPSO-RST limits scalability and convergence in dynamic environments with continuously evolving task demands. These challenges are further compounded in real-world firefighting scenarios by partial observability, limited UAV sensor capabilities, and physical constraints of UAVs related to payload and endurance.

Next, the complexities of non-stationary wildfire scenarios, including growing fires, emerging new fires, partial observability, and heterogeneous temporal and physical constraints, are addressed in the SUT mitigation. The problem is reformulated into a sequential spatiotemporal task assignment framework with non-stationary cost functions under partial observability. The Conflict-aware Resource-Efficient Decentralized Sequential planner (CREDS) is developed to address the challenges for early wildfire suppression using heterogeneous UAV teams. CREDS employs a three-phase approach: fire detection using a search algorithm, local trajectory generation with an auction-based Resource-Efficient Decentralized Sequential planner (REDS) incorporating a novel Deadline-Prioritized Mitigation Cost (DPMC) function, and a conflict-aware consensus algorithm to establish global trajectories for mitigation. CREDS achieves high success rates under various conditions, handling diverse fire-to-UAV ratios with scalability and robustness. The CREDS is robust against physical constraints, managing resource limitations through increased UAV capacity, additional UAVs, and efficient refueling strategies. In resource-constrained wildfire scenarios, the evolving nature of the wildfire may result in multiple spatially distributed larger fires, which require simultaneous and coordinated mitigation efforts from multiple UAVs. The single swarm mission with a decentralized approach has less likelihood of multiple UAVs detecting the same target. The multi-swarm missions with distributed solutions lead to the collective action of swarm members in the search and mitigation of larger fires in large unknown areas.

Finally, the thesis develops the Multi-Swarm Cooperative Information-Driven Search and Divide-and-Conquer Mitigation Control (MSCIDC) approach for large-scale wildfire scenarios. This methodology employs cooperative UAV swarms to enhance fire detection and mitigation efficiency. A two-stage search process combines exploration and exploitation, guided by thermal sensor data, for rapid identification of fire locations. Dynamic swarm behaviors, including regulative repulsion and merging, minimize detection and mitigation times, while local attraction accelerates the response of non-detector UAVs. The divide-and-conquer strategy ensures effective, non-overlapping sector allocation for fire mitigation. The simulations for a pine forest environment show that MSCIDC reduces the average burned area and mission time considerably compared to existing multi-UAV methods, providing faster and more efficient wildfire management.

Overall, the thesis presents scalable UAV swarm-based solutions to address clustered and large-scale wildfire management challenges. The UAV swarm-based solutions integrate decentralized spatiotemporal task assignment and multi-swarm strategies to effectively minimize ecological damage and provide robust solutions for real-world disaster management applications.

Speaker: Josy John

Research Supervisor: Dr. Suresh Sundaram