Neural sensing and control in a kilometer-scale gravitational-wave observatory

Suspended optics in gravitational-wave (GW) observatories are susceptible to alignment perturbations, particularly slow drifts over time, due to variations in temperature and seismic levels. Such misalignments affect the coupling of the incident laser beam into the optical cavities, degrade both the circulating power and optomechanical photon squeezing, and thus decrease the astrophysical sensitivity to merging binaries. Traditional alignment techniques involve differential wave-front sensing using multiple quadrant photodiodes but are often bandwidth restricted and limited by the sensing noise. We present a successful implementation of neural-network-based sensing and control at a GW observatory and demonstrate low-frequency control of the signal-recycling mirror at the GEO 600 detector. Alignment information for three critical optics is simultaneously extracted from the interferometric dark-port camera images via a convolutional neural net-long short-term memory network architecture and is then used for multiple-input-multiple-output control using soft actor-critic-based deep reinforcement learning. The overall sensitivity improvement achieved using our scheme demonstrates the capabilities of deep learning as a viable tool for real-time sensing and control for current and next-generation GW interferometers.