Drone-to-drone interception path planning by Deep Q-network with Graph Neural Network based (DQN-GNN) model.

With the increasing ubiquity of drones, criminals may also start using drones more frequently. This may necessitate catching these hostile drones with friendly drones. This paper proposes using reinforcement learning to train an agent involving a deep Q-network (DQN) with an underlying GNN-based model to move and intercept a moving target drone. The results from this approach seem to suggest that the use of GNNs with deep reinforcement learning is a feasible way to tackle this problem. With the method proposed in this work, the chaser drone can be trained to intercept the target drone. Furthermore, the chaser drone can catch the target in a reasonable amount of time.