RoboJudge-Bench for Evaluating Embodied Video Generation
October 19, 2026
RoboJudge-Bench is a new benchmark assessing whether MLLMs can accurately judge the physical adherence and instruction alignment of embodied video generation. It covers 800 human-annotated videos involving grippers and dexterous hands.
HOW THIS AFFECTS YOU
●
builderYou can use this to benchmark how well your multimodal judges can oversee video generation pipelines.
●
researcherThis provides a much-needed evaluation framework for the physical validity of generative world models.