Satsop plant study

Challenge

Two campaigns surveyed the plant for the DARPA Subterranean Challenge, each delivered as its own registered point cloud. As delivered, the second sat about 20 cm from the first, and after the best rigid fit the two still bent apart by up to about 4 cm in patches across the plant: each campaign's own registration had bent a little.

Approach

Both read straight from the delivered archives: 1.05 billion points. The second campaign registered onto the first, rigidly and then with a smooth elastic correction of at most 46 mm, checked on held-out points. Then both fused on a 2 cm grid, with a per-point record of which campaign measured each surface and how closely the two agree.

Outcome

One model of the plant's main level: 150 million points on a 2 cm grid and about 68,500 m² of surface. The campaigns mostly covered different rooms and floors, so fusion completes the plant. Where both measured the same surface, their fused surfaces agree to a median of about 10 mm, 86 % within 3 cm; on points held out of the fit, the median mismatch is 13.2 mm.

Confidence classes

The model's Confidence view, point by point.

• Both agree within 10 mmBoth survey campaigns measured the surface, and their surfaces lie within 10 mm of each other.
• Both agree within 30 mmBoth campaigns measured it, 10 to 30 mm apart.
• One campaignMeasured by one campaign only: the two mostly covered different rooms and floors.
• Campaigns disagreeBoth campaigns measured it, more than 30 mm apart: something moved between them, or the campaigns still differ there.