By subtracting a vision-only policy's action from the full policy, CofactVLA reduces vision-override and improves VLA robustness and out-of-distribution generalization in simulation and real robots.
Unleashing large-scale video generative pre-training for visual robot manipulation
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
CofactVLA: Deconfounding Vision-Language-Action Models via Counterfactual Intervention
By subtracting a vision-only policy's action from the full policy, CofactVLA reduces vision-override and improves VLA robustness and out-of-distribution generalization in simulation and real robots.