Fiducial marking is indispensable in mobile robots, including their pose calibration, contextual perception, and navigation. However, existing fiducial markers rely solely on vision-based perception which suffers such limitations as occlusion, energy overhead, and privacy leakage.
We present Polaris, the first vision-free fiducial marking system, based on a novel, full-stack magnetic sensing design. Polaris can achieve reliable and accurate pose estimation and contextual perception, even in NLOS scenarios. Its core design includes: (1) a novel digital modulation scheme, Magnetic Orientation-shift Keying (MOSK) that can encode key information like waypoints and coordinates with passive magnets; (2) a robust and lightweight magnetic sensing framework to decode and localize the magnetic tags. Our design also equips Polaris with three key features: sufficient encoding capacity, robust detection accuracy, and low energy consumption. We have built an end-to-end system of Polaris and tested it extensively in real-world scenarios. The testing results have shown Polaris to achieve an accuracy of up to 0.58 mm and 1◦ in posture estimation with a power consumption of only 25.08 mW.