Speaker
Description
Speeding up the locking procedure and establishing the resonance condition of optical cavities are crucial aspects to improve the duty cycle of gravitational wave detectors, enhancing the time within we are able to detect new significant signals. However, the process is highly challenging due to several non-linear effects, such as cavity ringing and resonance drift caused by thermal effect and radiation pressure. These effects spoil the main optical signals as the optical power and the Pound Drever Hall error signal, crucial cavity state witnesses for control purposes. To address these challenges, we propose a deep reinforcement learning based solution capable of adapting to the dynamic and non-linear nature of the cavity’s behavior. A fast time-domain simulator was developed to model the optical response of a FP cavity, taking into account the ring down effect. Subsequently, the simulator was used to develop a custom Gymnasium environment with which the DRL agent could interact and learn the best action policy. Leveraging HPC resources, different configurations between architectures and reward signals were investigated to understand better which would have gave the best performances. Currently, we are implementing the physics of the mirrors within the simulated environment to start filling the reality gap and address in the future SimToReal transfer.