Anchor-Free Multi-Target Azimuth Auto-Alignment with UWB and IMU

NOTE

📄 IEEE Paper: Anchor-Free Multi-Target Azimuth Auto-Alignment with UWB and IMU (IEEE Conference, 2026)

We built a multi-target cooperative positioning system that enables a group of users to localize themselves relative to each other in GNSS-denied environments (indoor, underground, dense urban) without any pre-deployed base stations, anchor nodes, or absolute heading references.

Each user wears two small sensor modules: a foot-mounted IMU (BMI088 on ESP32-C6) for pedestrian dead-reckoning, and an upper-arm-mounted UWB node (DecaWave DW1000) for inter-node ranging. The system fuses these two data sources to jointly estimate every node’s initial position and relative azimuth, mapping all local trajectories into a unified global coordinate frame.

The system operates under completely anchor-free conditions — no base stations, no magnetometers, no prior position information. At startup ($t=0$), Multidimensional Scaling (MDS) analytically recovers the relative spatial topology from UWB distance measurements alone. As users begin to move ($t>0$), a spatiotemporal optimization jointly solves for global rotation and individual node azimuths while a dual-branch parallel strategy eliminates mirror ambiguity — the correct configuration yields residuals several orders of magnitude smaller than the incorrect mirror solution.

Field tests were conducted on a real hardware platform. The UWB module achieves a typical ranging accuracy of $5\text{--}20\,\text{cm}$ via Double-Sided Two-Way Ranging. The foot-mounted IMU provides robust local displacement increments through ZUPT (Zero Velocity Update) to suppress cumulative drift.

The experiment took place on a $28\,\text{m}\times15\,\text{m}$ indoor basketball court with 5 subjects performing straight-line walking, right-angle turns, circular loops, and irregular maneuvers. The five nodes started from scattered positions with random initial orientations — no base stations or preset position information was available.

Throughout data collection, each node’s foot-mounted module independently ran INS algorithms. Displacement data and UWB ranging were aggregated at the master node, and after removing NLOS outliers, a multi-epoch observation sequence was constructed for reconstruction.

The results demonstrate that the algorithm successfully reconstructed continuous multi-node trajectories under real sensor noise, synchronization errors, and complex motion patterns — providing an autonomous initialization scheme for multi-device cooperative navigation in GNSS-denied environments.

NOTE

📄 Paper: Anchor-Free Multi-Target Azimuth Auto-Alignment with UWB and IMU — IEEE Conference, 2026

Invalid admonition directive. (Admonition directives must be of block type ":::note{name="name"} <content> :::")