Reinforcement Learning for Autonomous UAV Navigation Using Function Approximation

Unmanned aerial vehicles (UAV) are commonly used for search and rescue missions in unknown environments, where an exact mathematical model of the environment may not be available. This paper proposes a framework for the UAV to locate a missing human after a natural disaster in such environment, using a reinforcement learning (RL) algorithm. A function approximation based RL algorithm is proposed to deal with a large number of states representation and to obtain a faster convergence time. We conducted both simulated and real implementations to show how the UAVs can successfully learn to carry out the task without colliding with obstacles. Technical aspects for applying RL algorithm to a UAV system and UAV flight control were also addressed.