Simulation of ray behavior in biconvex converging lenses using machine learning algorithms

Abstract

This study used machine learning (ML) algorithms to investigate the simulation of light ray behavior in biconvex converging lenses. While earlier studies have focused on lens image formation and ray tracing, they have not applied reinforcement learning (RL) algorithms like proximal policy optimization (PPO) and soft actor-critic (SAC), to model light refraction through 3D lens models. This study addresses that gap by assessing and contrasting the performance of these two algorithms in an optical simulation context. The findings of this study suggest that the PPO algorithm achieves superior ray convergence, surpassing SAC in terms of stability and accuracy in optical simulation. Consequently, PPO offers a promising avenue for optimizing optical ray simulators. It allows for a representation that closely aligns with the behavior in biconvex converging lenses, which holds significant potential for application in more complex optical scenarios.