Neural-NF learns a mapping from intrinsic Laplacian features to local PDE coefficients whose solution yields a collision-free, monotonically descending navigation function with global goal minimum by construction, achieving up to 5x better zero-shot transfer than direct value-function predictors.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.RO 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Neural Navigation Functions for Zero-Shot Generalizable Motion Planning
Neural-NF learns a mapping from intrinsic Laplacian features to local PDE coefficients whose solution yields a collision-free, monotonically descending navigation function with global goal minimum by construction, achieving up to 5x better zero-shot transfer than direct value-function predictors.