sw/
← projects
Q-Bot·2019–20

Camera-LiDAR extrinsic calibration

Extrinsic calibration for the sensor turrets on the MK6.3 and MK7 spray robots and the Home Health Monitor, driven by a cost function that maximises point-cloud crispness straight from the 3D scans.

Problem

Every Q-Bot sensor turret carries colour cameras, thermal cameras and a LiDAR. Everything downstream (maps, foam thickness, thermal overlays, spray targeting) assumes they agree on where things are. When the camera-LiDAR extrinsics drift even slightly, scans from different turret poses land in slightly different places, joists double up and every measurement built on top becomes unreliable. And calibration had to be done by a production team, not by the person who wrote it.

What I did

I developed the camera-LiDAR extrinsic calibration system. It optimises a cost function that maximises the crispness of the point cloud, computed directly from the 3D scans: when the extrinsics are right, overlapping sweeps collapse onto thin, sharp surfaces; when they are wrong, surfaces smear. We validated it on real and simulated data and showed it was robust.

The same routine served three products: the MK6.3 and MK7 spray robots and the Home Health Monitor survey device. On the Home Health Monitor I also built the floorplan-to-3D-model pipeline that turns a home’s plan into a model for the survey. I helped design the physical calibration rig, and built an interactive UI so the production team could calibrate each unit quickly before it was commissioned for an on-site job.

Result

One calibration routine across two robot generations and a separate survey device, run by production rather than engineering. Sharp, consistent scans that the mapping, foam-thickness and spray features depend on. The Home Health Monitor is covered by the “Sensor apparatus” patent application, on which I am a co-inventor.

footage

muted, looping

MK6.3 turret scan of the same void, before calibration (left) and after (right). Beams and joists that were doubled collapse onto one surface.
Calibrated scans of a whole floor void, joists and all, seen from above.

figures

click to enlarge

fig. 01Home Health Monitor sensor unit: rangefinder, thermal imager and camera on a rotating head.US 2022/0003542 A1
fig. 02The survey unit on a tripod, talking to a tablet and the cloud.US 2022/0003542 A1, fig. 16
fig. 03Alternative mount: the sensor unit on a mast over a tracked base.US 2022/0003542 A1, fig. 9
esc