LoMAP projects each guided diffusion sample onto a PCA subspace of nearby offline trajectories, reducing infeasible plans and improving returns in Maze2D, MuJoCo locomotion, and AntMaze.
Although these methods provide asymptotic exactness, their practical efficiency under limited sampling budgets remains a significant challenge
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Local Manifold Approximation and Projection for Manifold-Aware Diffusion Planning
LoMAP projects each guided diffusion sample onto a PCA subspace of nearby offline trajectories, reducing infeasible plans and improving returns in Maze2D, MuJoCo locomotion, and AntMaze.