Tag
VAORA proposes a novel reward design for vision-language models that aligns reasoning with visual contexts and action outcomes, improving physical reasoning and generalization across unseen tasks and environments.