Proximity sensing detects object presence without contact, but it's rarely explored in granular materials (GM) due to complex granule properties and visionless scenarios. This paper proposes a granular-material-embedded autonomous proximity sensing system (GRAINS), which operates on basic granular phenomena (fluidization, jamming, and failure wedge zone) and utilizes mechanical signals to perceive proximity to buried objects. Specifically, using Gaussian process regression, the GRAINS captures force anomalies caused by granular jamming resulting from the proximity to buried objects by real-time learning force patterns obtained from the probe raking along the proposed spiral trajectory within GM. The results demonstrate that it can adaptively select its system parameters according to different granules and achieve a near-field perception of approximately 0.5 to 7 centimeters ahead, thereby avoiding direct contact with subterranean obstacles. Applied with Bayesian optimization, GRAINS successfully localizes and outlines buried objects, offering a contactless method for underground perception without human intervention.