MosaicIMU: Composing Carrier Experts for Generalizable Neural Inertial Odometry
MosaicIMU: Composing Carrier Experts for Generalizable Neural Inertial Odometry
This paper presents MosaicIMU, a carrier-conditioned Mixture-of-Experts framework designed to improve neural inertial odometry across heterogeneous platforms. By using a prototype-based router to compose carrier-specific expert features and integrating them with a history-aware Extended Kalman Filter (EKF), the method reduces average Absolute Trajectory Error (ATE) and Relative Translation Error over 10 seconds (RTE-10s) by 40% and 34%, respectively, compared to learning-based baselines. The approach supports domain adaptation through freezing a pretrained base model while learning lightweight expert residual branches, and enables efficient edge deployment via online sample selection for incremental updates.
The study introduces MosaicIMU, a carrier-conditioned Mixture-of-Experts framework that reduces average ATE and RTE-10s by 40% and 34% respectively in experiments, demonstrating improved generalization across heterogeneous inertial measurement units.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
he router to select informative online samples for efficient incremental updates. Experiments show that MosaicIMU consistently outperforms learning-based baselines, reducing average ATE and RTE-10s by 40% and 34%, respectively. These results highlight that MosaicIMU provides a scalable pretraining-to-deployment paradigm for generalizable and adaptive neural inertial odometry.
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org