Li Yiming’s team at Tsinghua University and Liqing Intelligence released PrismWAM, a world action model that separates semantic planning, physical-state rollout, and executable action generation, according to 机器之心. PrismWAM reportedly ranked first on RoboTwin 2.0 with an 83.67% average success rate across two scenario types and a 75.5% Clean-to-Random success rate, while RoboDoctor v1.0 scored 57.10 out of 100 across 54 RoboDojo tasks and led the compared methods. The report says contact and force prediction still require task- and experiment-specific verification.