High-vibration platforms: when the sensor is shaking while it scans

A rotating LiDAR doesn't capture a scene all at once. Each revolution of the sensor head takes a measurable amount of time, during which the platform and the sensor are moving. In stable conditions, this inter-scan motion is small and predictable, and a simple constant-velocity model corrects for it well enough. On platforms with significant vibration, this assumption breaks down.

Heavy machinery, rough-terrain vehicles, and industrial equipment produce oscillation profiles that distort the scan in ways a linear motion model cannot compensate. A point cloud captured during a vibration cycle is not a snapshot of the world at a single moment in time. It is a composite of observations taken at slightly different platform positions across the rotation, each displaced from where it would have been without the oscillation. The result is a smeared point cloud where surfaces that should be flat appear jagged and edges that should be sharp appear blurred.

This distortion propagates directly into localization. Scan matching algorithms attempt to align the incoming point cloud with the reference map. If the incoming cloud is distorted, the match is degraded. The resulting pose estimate carries systematic error that is not detectable from the match score alone, because the match quality metric reflects the quality of the (distorted) scan, not the true pose accuracy.