UAV-based Firefighting by Multi-agent Reinforcement Learning

Wildfire in the forest leads to a huge amount of financial and human losses. That is it causes damage to the forest and the life of firefighters. To reduce the amount of such damage, unmanned aerial vehicles are among the best options to do the job of firefighting. Their orchestration of effective fire extinguishing is a must and essential. This work proposes a multi-agent reinforcement learning method using an actor-critic structure to extinguish the fire. While the learning is centralized, their execution is decentralized. Moreover, we define two models of fire spreading based on the discrete-time Markov process. The results show that the proposed method outperforms the state-of-the-art methods.