Interferobot: aligning an optical interferometer by a reinforcement learning agent
This addresses the challenge of applying deep RL to real-world robotics with limited training data, specifically for optical experiments, though it is incremental as it builds on existing simulation-to-real transfer methods.
The researchers tackled the problem of aligning an optical interferometer by training a reinforcement learning agent in simulation and transferring it to a real-world setup, achieving expert-level performance without fine-tuning.
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.