Google DeepMind unveiled Deep Loop Shaping, an AI‑driven control strategy that slashes vibration noise in gravitational‑wave detectors. The technique was tested at LIGO’s Livingston, Louisiana site, where it cut the most unstable feedback loop’s noise by 30‑100 times. By keeping interferometer mirrors steadier, the observatory can now capture fainter ripples from distant mergers, with unprecedented precision, boosting overall data quality substantially overall.

Deep Loop Shaping replaces traditional linear controllers by training a reinforcement‑learning agent that optimizes frequency‑domain rewards. During simulations, the agent learns to suppress control‑induced noise while maintaining mirror stability. When deployed on hardware, the new controller matched simulation performance, proving that AI can outperform hand‑tuned designs in one of the most sensitive feedback systems ever built for astronomical observations today.

By reducing control noise, Deep Loop Shaping expands LIGO’s reach, potentially increasing detectable events by hundreds each year and sharpening tests of general relativity. The approach also offers a blueprint for vibration‑sensitive instruments in aerospace, robotics, and structural engineering. Ultimately, the method pushes the boundary of what can be measured in the fabric of space‑time for physicists worldwide and researchers.