Laser systems utilized in scientific research and high-precision applications, such as those for coherent beam combining (CBC), demand exceptionally stable and accurate beam alignment for optimal performance. While manual alignment establishes an initial operational state, these systems inherently misalign over time due to environmental factors. Classical automated stabilization systems often perform poorly when applied to low-pulse repetition frequency (PRF) and burst-mode laser systems. These systems are essential for DESY’s X-ray Free-Electron Lasers. The poor performance is due to the long dark time between pulses and the accompanying environmental disturbances. The goal of this project is to establish machine learning methods trained on additional environmental sensor data to improve this malbehavior. The PhD candidate will train a reinforcement learning (RL) agent to directly control mirror actuators by optimizing key real-time performance metrics, such as measured laser power or specific beam profile characteristics. The training methodology will feature a robust pipeline, beginning with policy pre-training in a physics-based simulation environment, followed by real-world fine-tuning using data-efficient off-policy RL algorithms. Once trained, the resulting robust policy can be deployed to rapidly re-optimize the system (e.g., to counteract power degradation) or run continuously to maintain peak operational performance, offering a significantly more adaptive and performance-driven solution than traditional methods.

Requirements:

• Academic Background
• Master's degree in physics, electrical engineering, mechatronics, computer science, or a closely related field
• Technical / Methodical Skills
• demonstrated hands-on experience setting up and aligning optical systems or similar complex experimental set-ups
• demonstrated skills in interfacing physical hardware ( sensors and actuators) to software
• strong programming skills in Python
• demonstrated track record on research publications
• strong interdisciplinary research and team-work skills
• Advantageous Qualifications
• practical experience with machine learning, particularly Reinforcement Learning. Familiarity with frameworks like PyTorch or JAX is a significant plus
• knowledge in control theory
• practical experience with control systems
• experience in developing physics-based simulations for the purpose of modeling or training control agents
• practical experience with ultrafast lasers

Position:

• Deutsches Elektronen-Synchrotron DESY
• 75% EGR. 13 (TVöD) position for three years

At DESY, gender equality is an important aspect. To support work-life balance, we offer flexible working hours, variable part-time, job-sharing models, and participation in mobile work (up to 50%).

DESY promotes equal opportunities and diversity. The professional development of women is very important to us. Therefore, we strongly encourage women to apply for the vacant position.

Applications from severely disabled persons will be given preference if they are equally qualified (sbv.desy.de).