Using AI to perceive the universe in greater depth

This post details the application of Deep Loop Shaping, a novel AI method leveraging reinforcement learning with frequency domain rewards, to reduce control noise in LIGO's gravitational wave observatory. The method achieved noise reduction of 30-100 times in the most unstable feedback loop, surpassing traditional linear control methods and eliminating controller-induced noise in the observation band. The performance was validated on both simulated and real hardware at the LIGO Livingston observatory.