Image-Based Reinforcement Learning for Robust Mode Hop Recovery in External-Cavity Diode Lasers

Abstract

Overview This is the initial public release of the code associated with the paper "Image-Based Reinforcement Learning for Robust Mode Hop Recovery in External-Cavity Diode Lasers." The repository provides a simulation and reinforcement-learning framework for recovering an external-cavity diode laser from mode hops using image-based observations and piezo control. Highlights Rust/PyO3 simulation environment for efficient reinforcement-learning experiments Image-based Rainbow DQN training and evaluation pipeline Simulation-first execution with explicit opt-in hardware control Runtime configuration for site-specific laser controllers, wavemeters, and oscilloscopes Fail-safe evaluation behavior and bounded hardware communication waits Training and evaluation defaults aligned with the accompanying paper Public-release documentation, citation metadata, automated checks, and tests Repository contents simulator/: laser mode-hop simulator and Python bindings train.py: reinforcement-learning training entry point eval.py: simulation and hardware evaluation entry point server.py: TCP interface for the laser controller wavemeter.py: wavemeter interface oscilloscope.py: oscilloscope interface scripts/audit_public_tree.py: check for machine-specific paths and accidental private artifacts Data and hardware notes Empirical laboratory datasets, recovery logs, trained checkpoints, and hardware-specific configuration values are not included in this release. Hardware deployment requires the user to provide the relevant controller address, wavemeter address, oscilloscope resource, calibration data, and safety limits for their own setup. Review all limits and safety behavior before connecting the software to laboratory equipment. Installation and usage Installation, training, simulation evaluation, and hardware configuration instructions are provided in the repository README. Citation If you use this code in academic work, please cite the accompanying paper. Citation metadata is provided in CITATION.cff.