Weak-field magnetometry is an essential branch of quantum sensing, playing a critical role in areas such as mineral exploration, geomagnetic monitoring, navigation and positioning, and anti-submarine warfare. However, state-of-the-art magnetometers are frequently constrained by trade-offs among sensor size, sensitivity, and dynamic robustness, severely limiting their application on high-speed mobile platforms and in magnetically disturbed environments. To address these limitations, this work presents a highly robust, miniaturized ^87\textRb atomic magnetometer based on optical-magnetic resonance. A key feature of our magnetometer is the integration of a vertical-cavity surface-emitting laser, micro-optical assemblies, a sub-centimeter atomic vapor cell, micro-heating modules, flexible thin-film radio-frequency coils, and a non-magnetic photodiode. This highly integrated architecture reduces the probe volume to only 7\,\textcm^3 with a total power consumption below 5\,\textW. The radio-frequency amplitude is optimized to maximize the peak-to-width ratio of the Lorentzian resonance absorption profile, yielding the steepest frequency discrimination slope. The feedback gain is maximized within the constraints of intrinsic hardware delays without introducing additional baseline noise. As a result, the magnetometer achieves a baseline sensitivity of 10\,\textpT/\sqrt\textHz and a maximum tracking slew rate of 25200\,\textnT/s. A lock-loss detection scheme based on the amplitude of the second-harmonic phase-sensitive demodulation signal of the transmitted light is introduced. By monitoring the second derivative of the absorption profile in real time to quantify resonance degradation, the system identifies lock loss and automatically performs a full-band frequency sweep to relock the magnetic field on a sub-second time scale. Field tests demonstrate the robustness and practical utility of the magnetometer. In long-term continuous outdoor geomagnetic monitoring, it records the complete evolution of a severe geomagnetic storm without data interruption. In a surface-based remote sensing experiment above a subway station, it accurately maps the complex dynamic magnetic signatures of underground trains despite transient disturbances exceeding 2000\,\textnT/s. In a handheld non-invasive localization experiment, it identifies the dipole signature of a concealed underground magnetic target while withstanding mechanical vibrations and attitude disturbances. The approach presented in this work offers a viable pathway to bridge the gap between high-performance quantum sensing and field applications, thereby promoting the deployment of miniaturized quantum magnetometers in real-world scenarios.