Speed changes the geometry of localization in a straightforward way that is easy to underestimate. At low speeds, consecutive LiDAR scans share a large overlap area. The vehicle has moved only a short distance between observations, so matching the current scan to the previous one, or to the map, is a well-constrained problem with many available correspondences. As speed increases, that overlap shrinks. The vehicle covers more ground between scans, fewer features are shared between consecutive observations, and the matching problem becomes harder.
A rotating LiDAR that completes one revolution per scan takes a fixed amount of time to produce each frame. At 20 km/h that time corresponds to a small displacement. At 120 km/h it corresponds to a distance where two consecutive scans may share very little structure. At 200 km/h and above, some scan pairs will share essentially no common coverage, depending on the environment geometry and the sensor's horizontal field of view.
The second constraint is latency. A navigation system that updates pose at 5 Hz on a 2 m/s forklift is marginally acceptable. At 70 m/s (250 km/h), a 200 ms pose update cycle means the vehicle travels 14 meters between position estimates. Vehicle control at that speed requires pose information at 10 Hz or higher, with low and bounded latency, not average latency.
The third constraint is error propagation speed. Localization errors at high velocity don't stay small for long. A 10 cm error in position at 2 m/s gives the system time to detect and correct the discrepancy before consequences become severe. The same error at 70 m/s may produce a collision before the next scan has been processed.
If your platform operates at speeds where standard localization architectures degrade, we can tell you precisely where the failure threshold is and what addressing it requires.
Book a demo →