Author(s): Hu, K (Hu, Ke); Xu, B (Xu, Bing); Xu, ZX (Xu, Zhenxing); Wen, LH (Wen, Lianghua); Yang, P (Yang, Ping); Wang, S (Wang, Shuai); Dong, LZ (Dong, Lizhi) 

Source: OPTIK  Volume: 178  Pages: 785-793  DOI: 10.1016/j.ijleo.2018.09.160  Published: 2019   

Abstract: An aberration correction algorithm for wavefront sensorless adaptive optics (WFSless AO) systems based on deep reinforcement learning is presented. An actor critic structure is designed to evaluate a control policy through the deep deterministic policy gradient (DDPG) algorithm. The algorithm performance is verified with a set-up simulation environment. According to the correction results, the aberration correction process can be expressed as a Markov decision process (MDP). The method exhibits excellent performances in correction capacity and speed. Similar correction effects are obtained with the stochastic parallel-gradient descent (SPGD) algorithm and WFSless AO based on the general-modes (AOG) algorithm. Moreover, the correction speed is improved by approximately 9 and 2.5 times, respectively. 

ISSN: 0030-4026