Quantum Sensing Protocols from Reinforcement Learning

Nitrogen-vacancy (NV) spin ensembles in diamond provide an advanced magnetic sensing platform at ambient conditions. Improvements in fabrication techniques provide access to diamond material with moderate (∼1 ppm) to high (>10 ppm) NV densities while maintaining isotopic purity and low (∼10 kHz) crystal strain gradients. Combining dynamical decoupling techniques and suppression of spin bath induced dephasing, NV coherence magnetometry can operate within the interaction-limited regime, where spin coherences are restricted by interactions between NV sensor spins. By considering the effective average Hamiltonian during a quantum control sequence, previous influential work in Hamiltonian engineering successfully extends NV coherence by suppressing the effects of such dipolar interactions, resulting in sensitivity improvements to AC magnetic fields. Here, we instead utilize a reinforcement learning algorithm to design pulse sequences for sensitivity improvement. This requires the definition of a robust reward structure for quantum sensing, which includes considerations for noise sources such as magnetic disorder, interactions and pulse errors.