Adaptive High-Precision Measurement for Optical Encoders at Various Speeds Based on Deep Reinforcement Learning

To improve the measurement precision of the optical encoder working at various speeds, an adaptive high-precision measurement framework is designed based on deep reinforcement learning (DRL), cascading a compensation module, and a positioning module and a decoding module. To the best of our knowledge, this is the first attempt to introduce the DRL into the measurement via the optical encoder working at various speeds. To this end, an improved deep Q-learning network (DQN) is designed to construct the compensation module. Specifically, to construct the state space, a mapping rule is designed to convert high-dimensional grating patterns into 1-D angle variables. Adaptive actions and adaptive reward are designed to improve the compensation ability of the DQN model. The compensated grating patterns successively pass through the positioning module and the decoding module for high-precision measurement. Experimental results indicate that the designed measurement method is superior to exiting methods for precise measurement of the optical encoder at various speeds, especially for high-speed measurement.