1 paper · 1 filter
Zicheng Kong, Zhenbo Xu, Haoyan Zhang +7
Video world models enable candidate actions to be evaluated through imagined outcomes, but high predicted rewards can favor trajectories with poor execution quality. We introduce T…