Deep Reinforcement Learning Based on Optical Neural Networks in Path Planning

We propose the optical deep Q network (ODQN) algorithm based on optical neural networks (ONNs) to accelerate calculation in 2D grid path planning task. The calculated results demonstrate that the innovative algorithm is competitive with the original deep Q network (DQN) algorithm based on artificial neural networks (ANNs). Further, we analyze the noise errors of ONNs and prove the relatively good robustness of the ODQN algorithm in the 2D grid world. Compared to previous works, we introduce an efficient learning strategy for ONNs and demonstrate its broad application prospects in reinforcement learning and so on.