Weak-Signal Interpretation for Robotics Anomaly Sensing

Robotic systems often encounter weak, noisy, partial, and locally ambiguous signals before a clearer anomaly becomes legible. A drive system may show slight current asymmetry, a manipulator may exhibit repeated micro-slip under otherwise plausible contact, or an environmental obstruction may appear inconsistently across perceptual routes. In such cases, the architectural problem is not only detection, but promotion control: how weak findings should influence attention before they are allowed to become stronger operational consequence. This paper argues that robotics anomaly sensing is a suitable application domain for a stratified weak-signal architecture that separates bounded observation, provisional interpretation, short-horizon attentional escalation, and governed promotion. The contribution is narrow and operational. The paper defines a minimal transfer contract for embodied anomaly handling: a light state ladder, a bounded event object, and illustrative promotion logic showing how weak signals may influence inspection before they justify stronger commitment. It does not propose a full robotics stack, semantic SLAM system, safety-certified controller, or universal diagnosis engine. Its narrower claim is that embodied systems need a disciplined place for weak anomaly signals to matter before they harden into motion change, maintenance significance, or ignored noise.