UAV Autonomous Trajectory Planning in Target Tracking Tasks via a DQN Approach

For unmanned aerial vehicles (UAVs) to find safe and valid paths in a complex environment is a challenging and complicated work. However, many existing approaches of that are based on the specific scenarios. When the scenario is unknown, the algorithm would be noneffective. To improve the efficiency of the path planning method, the DQN approach is utilized in the target tracking tasks. The issue is modeled as follows: the UAV can be seen as the agent, who needs to find a safe and valid way to the destnation. The process of that can be designed as a Markov decision problem. As a deep reinforcenment learning way, DQN has deep networks to approximate the Q value accurately, and the Q(λ) algorithm makes it possible to make the right decisions in path planning process. It is a remarkable fact that even if the environment is unknown, DQN approach is also effective. Numerical simulations are provided to show the effectiveness of the suggested system.