Abstract:

Unmanned Aerial Vehicles (UAVs or

drones) play a crucial role in surveillance missions,

especially where ground infrastructure is damaged or

inaccessible. Ensuring reliable and simultaneous coverage

of critical zones by UAVs, known as k-coverage,

remains a significant challenge. Traditional methods

require UAVs to cover an entire area which leads to high

energy consumption, this is problematic, especially in

environments where battery recharge or replacement

is difficult. To overcome these challenges, Only a set

of targets should be monitored instead of monitoring

the entire area. This paper proposes DQTCP (Deep

Q-learning-based Target Coverage Protocol), a new

deep reinforcement learning approach to continually

cover a maximum number of stationary targets. In

DQTCP, the UAV acts as an autonomous Deep QNetwork

(DQN) agent, with discrete actions and individualized

learning parameters balancing exploration

and exploitation. Through iterative training and environment

interaction, UAV adopts policies that optimize

the target coverage effectiveness. Simulations

show that DQTCP using based on the Reinforcement

learning theory, is very efficient in terms of coverage

performance and stability.