Checked, not settled
Checks out so far.
“In two UR10 tabletop scenes, the navigation-function-supervised SE(3) neural potential field converged within 3 cm of the grasp from every start and produced collision-free paths against ground-truth geometry, whereas image-only supervision produced 25% and 0% collision-free rates in the two scenes.”
2 of 4 review checks done · 0 counted · 2 by the author only
Supported · 0.8581 under ledger-default@0.3.0 at snapshot 27745 · review checks 0 of 4 · provisional until audited. The number is model-conditional, not a share of truth.
Share:
Checked in Galedra: score 0.86, mostly supported · not yet reviewed by anyone else
“In two UR10 tabletop scenes, the navigation-function-supervised SE(3) neural potential field converged within 3 cm of the grasp from every start and produced collision-free paths against ground-truth…”
https://galedra.org/claims/dbaafb12-6dd1-8689-97e4-13fd00a3f605/card
Or just the link: https://galedra.org/claims/dbaafb12-6dd1-8689-97e4-13fd00a3f605/card
See the evidence · 
Galedra