Interferobot: aligning an optical interferometer by a reinforcement learning agent

TL;DR

This paper presents a reinforcement learning agent trained in simulation to align a Mach-Zehnder interferometer using only images, successfully transferring to real hardware without fine-tuning, matching expert performance.

Contribution

It introduces a simulation-trained RL approach for optical interferometer alignment that generalizes to real-world hardware without additional tuning.

Findings

The RL agent achieves human expert-level alignment performance.

Domain randomization enables successful transfer from simulation to physical interferometer.

The method requires no hand-crafted features or physics-based prior knowledge.

Abstract

Limitations in acquiring training data restrict potential applications of deep reinforcement learning (RL) methods to the training of real-world robots. Here we train an RL agent to align a Mach-Zehnder interferometer, which is an essential part of many optical experiments, based on images of interference fringes acquired by a monocular camera. The agent is trained in a simulated environment, without any hand-coded features or a priori information about the physics, and subsequently transferred to a physical interferometer. Thanks to a set of domain randomizations simulating uncertainties in physical measurements, the agent successfully aligns this interferometer without any fine tuning, achieving a performance level of a human expert.