From Power Concentration to Structured Light Generation: Scalable, Efficient and Generalizable Coherent Beam Combination With Model‐Based Reinforcement Learning
TL;DR
This work proposes a model‐based reinforcement learning (MBRL) framework that enables intelligent phase control by effectively emulating CBC system dynamics via a physics‐informed environment model, and establishes a powerful paradigm for integrating MBRL into high‐power optical field modulation.
Abstract
High‐power lasers are indispensable in numerous scientific and technological frontiers. Coherent beam combination (CBC) offers a compelling path to surpass the fundamental power limit of a single laser. However, achieving scalable and efficient phase control across different sizes of laser arrays remains a formidable challenge. We propose a model‐based reinforcement learning (MBRL) framework that enables intelligent phase control by effectively emulating CBC system dynamics via a physics‐informed environment model. Our MBRL‐CBC exhibits scalable and robust performance across simulated CBC systems with up to 91 channels through label‐free training. The success is also extended to a 19‐channel experimental platform, with a potential of further scaling up, if given improved hardware. Furthermore, MBRL‐CBC extends to structured light generation, as demonstrated through both simulations and experiments on orbital angular momentum beams. This work establishes a powerful paradigm for integrating MBRL into high‐power optical field modulation.