To address poor scanning consistency, unstable probe coupling, insufficient adaptability of fixed excitation parameters, and difficulty in three-dimensional representation of detection results in array eddy current testing of conductive metal components, this paper proposes an embodied sensing and control method for array eddy current defect detection based on physics-informed reinforcement learning. An automatic inspection system composed of dual manipulators, red-green-blue depth (RGB-D) vision, an array eddy current testing unit, and a digital twin platform is constructed. Visual three-dimensional reconstruction, inspection surface segmentation, coverage path planning, and digital twin verification are combined to realize safe automatic scanning of irregular metal components. To address coupling instability caused by lift-off and posture fluctuations under vision-guided coarse scanning, an electromagnetic-feedback pose correction method based on a soft-gated mixture-of-experts model is proposed to realize closed-loop compensation of the probe pose. Furthermore, the optimization of array excitation parameters is formulated as a sequential decision-making problem, and a physics-informed advantage actor-critic (Physics-A2C) method is proposed. Electromagnetic-twin physical priors are embedded to achieve adaptive regulation of multi-channel excitation amplitudes and phases. Vision-electromagnetic fusion three-dimensional imaging and localization are further realized by combining the scanning pose sequence with the point-cloud surface mapping relationship. Experimental results show that the proposed method achieves coverage rates above 96% and overlap rates below 5% on three types of metal specimens. After closed-loop correction, the baseline standard deviation of detection signals is reduced by more than 80%, and the defect signal-to-noise ratio is improved by 0.5~5.2 dB. After adaptive excitation regulation, the average sensitivity improvement for multiple types of defects reaches 72.4%, enabling the detection of defects with a width of 0.1 mm and a depth of 1 mm. The three-dimensional localization errors of defects on flat and curved specimens are controlled at the millimeter level. The results verify the effectiveness of the proposed embodied sensing and control method in stable scanning, sensitivity enhancement, and three-dimensional spatial representation.