Adaptive and Robust Control of Diamond Quantum Sensors via Meta‐Learning

ABSTRACT

High‐fidelity quantum sensing with nitrogen‐vacancy (NV) ensembles is critically dependent on the precise optimization of microwave control pulses. A significant challenge for the industrial‐scale deployment of these sensors is the poor transferability of optimized control protocols, as performance degrades substantially when a protocol for one specific device is applied to another, necessitating a costly and time‐intensive recalibration for each new scenario. This problem stems from unavoidable device‐to‐device variations such as material inhomogeneities, manufacturing tolerances, differences between diamond samples, as well as from dynamic environmental factors like temperature fluctuations and stray magnetic fields. In this work, data‐driven optimization techniques are applied to meta‐learn quantum‐control protocols for NV‐ensembles that generalize across hardware and environmental variations. It is demonstrated experimentally that the best meta‐learned optimizers can learn nearly optimal protocols and adapt to unseen conditions in as few as iterations, representing an improvement of several orders of magnitude relative to state‐of‐the‐art. The resulting transferable surrogate models can rapidly adapt to new, unseen device characteristics, representing a critical shift from single‐device optimization to a more robust and scalable strategy for high‐fidelity NV‐based quantum sensing in practical applications.