localization-hero
Localization

Start away from the dock,
hold position as the site changes

Knowing where the robot is right now is where every task begins.
We couple the prior map directly to live sensing so that starting point is never lost.

Restart

Power it up anywhere,
not only at the dock

You do not have to drive the robot back to a fixed station every time you start it. We match the prior map against what the sensors see right now to recover the robot's initial position and heading, then carry straight on into continuous tracking.

  • Start anywhere Restart the robot somewhere else in the work area and it finds where it stands before anything else, ready to run.
  • Where it applies Within the prior map and the configured search area, wherever the observations carry enough information to tell one place from another.
  • What it builds on Initial pose search is built on BBS3D. Trajectory error after a normal initialisation and the success rate of restarting from an arbitrary pose are two different figures.
Continuity

When the map shifts
and the sensor is blocked

Stock moves, equipment is rearranged and the sensor's view is blocked for a while. That is what a real floor looks like. We combine the prior map, continuous LiDAR observation and motion from the IMU to keep tracking through it.

Driving through unmapped ground
A changing map - where less and less matches the prior map, frame-to-frame motion and inertial data carry the estimate.
Sensor covered with a paper cup
Partial occlusion - when part of the view is blocked, whatever the remaining sensors can see keeps the estimate alive.

This is not a guarantee of drift-free operation under every condition where the prior map and sensing disappear for a long stretch.

Performance

Five kilometres in,
and nothing has slipped

Map registration is coupled directly into the pose estimate, so error against the map does not build up — it keeps being pulled back.

Position error over runs of 5km and longeraround 2cm
MethodTightly coupled

From an IsaacSim stress test.

Internal experiment Trajectory error against open-source 3D localization over 48m Three 8m round trips, about 48m in all, driven in 54 seconds. The metric is aligned ATE RMSE, and lower is better.
LiOps localization2.081cmBaseline
FAST-LIO24.519cmLiOps 53.9% lower
SAGE8.720cmLiOps 76.1% lower
Point-LIO11.529cmLiOps 81.9% lower
MULLS28.486cmLiOps 92.7% lower
From an in-house run. The evaluation sequence, version and parameters for each method compared, and what initial pose and prior map each was given, are covered by a non-disclosure agreement and are not published.
Data flow background

Find out whether it holds
on your floor

Send us the site conditions and a map, and we will come back with whether it applies and what performance to expect.