Inertial Odometry

Inertial odometry uses accelerometers and gyroscopes to estimate how a person or robot moves, providing localization when GPS is unavailable. Classical systems integrate IMU measurements with bias and gravity estimation, often using filters or factor graphs; newer learned methods predict velocity or displacement from motion sequences, sometimes with physical constraints or uncertainty estimates. Cameras, LiDAR, radar, wheel measurements, and leg kinematics can anchor inertial estimates, with systems evaluated in settings from pedestrian datasets such as OxIOD and RoNIN to aerial and robotic benchmarks such as EuRoC. The central tension is that inertial-only estimates accumulate drift and may generalize poorly across devices and motion, while sensor fusion improves accuracy but can falter when other sensors are degraded, motion is weakly observable, or real-time resources are limited.