Deep Reinforcement Learning-Assisted Triaxial Magnetic Moment Vector Measurement Using an Atomic Magnetometer

Accurately measuring the triaxial magnetic moment vector of paleomagnetic samples is crucial for geological research. Atomic magnetometers possess extremely high sensitivity for magnetic field measurements; however, achieving high-precision triaxial magnetic moment vector measurements remains challenging. This difficulty primarily arises from the strong dependence of measurement accuracy on the relative positioning between the sensor and the sample, which is difficult to determine in advance. To address this issue, we propose a measurement approach based on deep reinforcement learning (DRL) to dynamically optimize the measurement distance. Specifically, we introduce a dueling deep transformer Q-network (Dueling DTQN) algorithm, which treats the atomic magnetometer as an agent interacting with the magnetic field environment to autonomously determine optimal measurement positions. We constructed a measurement system using a spin-exchange relaxation-free (SERF) atomic magnetometer and validated the proposed method on paleomagnetic samples. The experimental results demonstrated high measurement sensitivity of $7.8\times 10^{-6}$ , $8.6\times 10^{-6}$ , and $4.2\times 10^{-6}$ emu along the x-, y-, and z-axes, respectively, and strong consistency with results obtained from a superconducting rock magnetometer (SRM). This work highlights the novel application of DRL in magnetic moment measurement, offering an efficient and intelligent measurement in paleomagnetism and materials analysis.