Camera pose and spatial arrangement are the most important dimensions for both collecting diverse robot data and retrieving useful subsets, and aligned retrieval from DROID beats full-dataset co-training on real tasks.
The different co-training spatial distributions are concentric boxes that increase in size until they cover the target spatial distribution completely
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.RO 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
What Matters in Learning from Large-Scale Datasets for Robot Manipulation
Camera pose and spatial arrangement are the most important dimensions for both collecting diverse robot data and retrieving useful subsets, and aligned retrieval from DROID beats full-dataset co-training on real tasks.