Reinforcement Learning for Aligning Laser Optics with Kinematic Mounts

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Aligning laser beams in optical systems is typically challenging, demanding both substantial time and expertise. This paper introduces a novel approach utilizing reinforcement learning (RL) architectures, specifically Deep Q-Learning (DQL), to align laser beams through controlling laser spot positions in optical setups. Our method employs RL for mirror position control, enabling the proposed solution to tune the mirror angles seamlessly. Real-time camera feedback informs the machine learning model's decisions. We propose a twostep laser alignment procedure: initially aligning the laser spot on the front aperture using the first RL model, and then once the laser is within the first aperture area, aligning the laser through both apertures using the second RL model. We verify the success of the RL model for the first step through experiments with a physical laser alignment setup of various configurations. We verify the RL model for the second step through simulations.
Original languageEnglish
Title of host publication20th IEEE International Conference on Automation Science and Engineering 2024
PublisherIEEE
Publication statusAccepted/In press - 3 Jun 2024
Event20th IEEE International Conference on Automation Science and Engineering 2024 - The Nicolaus Hotel, Bari, Puglia, Italy
Duration: 28 Aug 20241 Sept 2024
Conference number: 20
https://2024.ieeecase.org/

Conference

Conference20th IEEE International Conference on Automation Science and Engineering 2024
Abbreviated titleCASE 2024
Country/TerritoryItaly
CityBari, Puglia
Period28/08/241/09/24
Internet address

Fingerprint

Dive into the research topics of 'Reinforcement Learning for Aligning Laser Optics with Kinematic Mounts'. Together they form a unique fingerprint.

Cite this