AMR-Pose uses LED markers to keep underwater robots coordinated

AMR-Pose, introduced by Zeyu Sha, Xiaorui Wang, Mingyang Yang, and Feitian Zhang, tackles a persistent problem in underwater robotics: maintaining reliable relative positioning between cooperative AUVs without GPS. The team designed a compact physical marker module—one red central LED surrounded by three blue peripheral LEDs—that provides high-contrast visual features even when water is turbid, lighting shifts, or reflections interfere. A probabilistic switching Perspective-n-Point (PSwPnP) estimator then fuses these observations with Lie-group pose propagation on SE(3), meaning it uses the robot's known motion model alongside pixel data to predict and correct position and orientation.

What stands out is the estimator's ability to switch between different measurement modes based on how many LEDs are visible. When one or more markers are briefly occluded, the system automatically down-weights unreliable observations and relies more on propagated pose, maintaining geometric consistency and temporal stability. The authors validated AMR-Pose in water-tank experiments using motion-capture ground truth, showing accurate, smooth 6-DOF tracking in degraded conditions. They also ran closed-loop leader-follower tests, where the follower used AMR-Pose's relative pose estimates to track a moving leader in real time—demonstrating it works not just in simulation, but as part of an actual control loop.