How to Train Your Gyro: Reinforcement Learning for Rotation Sensing with a Shaken Optical Lattice

Liang-Ying Chih, Dana Z. Anderson, Murray Holland

Published 2022-12-29Version 1

As the complexity of the next generation of quantum sensors increases, it becomes more and more intriguing to consider a new paradigm in which the design and control of metrological devices is supported by machine learning approaches. In a demonstration of such a design philosophy, we apply reinforcement learning to engineer a shaken-lattice matter-wave gyroscope involving minimal human intuition. In fact, the machine is given no instructions as to how to construct the splitting, reflecting, and recombining components intrinsic to conventional interferometry. Instead, we assign the machine the task of optimizing the sensitivity of a gyroscope to rotational signals and ask it to create the lattice-shaking protocol in an end-to-end fashion. What results is a machine-learned solution to the design task that is completely distinct from the familiar sequence of a typical Mach-Zehnder-type matter-wave interferometer, and with significant improvements in sensitivity.