LiDAR-inertial odometry

Combine fast inertial motion with scan geometry#

An inertial measurement unit measures angular velocity and specific force at a high rate. A lidar measures ranges across a scan. LIO estimates motion by combining these streams instead of treating either as a complete pose sensor.

Inertial estimates can predict motion between range measurements and help account for movement while a scan is collected. Geometric consistency between new points and a local map can then correct pose, velocity, gravity direction, and sensor-bias estimates, depending on the system.

Direct registration is one implementation choice#

FAST-LIO2 tightly couples LiDAR and inertial data in an iterated Kalman-filter framework. It registers raw points directly to an incrementally maintained map rather than requiring a hand-designed feature-extraction stage. The authors evaluated that system across public sequences and experiments with different LiDAR scanning patterns.

That design is an example of LIO, not its definition. Other systems can use extracted features, factor graphs, or batch optimisation. LIO also differs from point-cloud registration alone because inertial states and measurements participate in the motion estimate.

Odometry still drifts#

Long corridors, open spaces, or repeated surfaces may provide too little geometric constraint in some directions. Incorrect time synchronisation, extrinsic calibration, IMU bias modelling, moving objects, and vibration can all corrupt the estimate.

LIO remains odometry: it provides locally consistent incremental motion but does not necessarily provide a permanent global frame. A larger SLAM system may add place recognition, loop closure, and global optimisation.