RoboAlign-R1 raises six-dimension scores on robot video generation by 10.1% via reward-aligned RL post-training and sliding-window inference stabilization.
Stage 1 consumes only the pixels{Is+k}T−1 k=0; Stage 2 additionally consumes the paired actions to construct the joint token sequence of §B.3
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
-
RoboAlign-R1: Distilled Multimodal Reward Alignment for Robot Video World Models
RoboAlign-R1 raises six-dimension scores on robot video generation by 10.1% via reward-aligned RL post-training and sliding-window inference stabilization.