Neural Q Learning Algorithm based UAV Obstacle Avoidance

In the paper, Neural Q Learning algorithm (NQL) was involved to solve the obstacle avoidance problem for UAV path planning. Q learning was good at online learning and BP network provided excellent function approximation. The combination of two methods can provided UAV a collision-free trajectory in unknown environment. Through several simulations, the proposed algorithm could gain better performance and gain higher success rate than classic Q-learning (CQL). Besides, this method was extended for deep reinforcement learning, such as DQN, which is more suitable for practical applications.