An open robotic value-foundation-model family from Alibaba DAMO Academy and Hupan Lab. It learns temporal distance to a language-specified goal from more than 7,000 hours and roughly 3.09M clips, without preference labels. RynnValue reaches Kendall's tau-a 0.675 and raises real-robot success from 52.5% to 72.5% online and 63.8% to 82.5% offline.

Model Details

License Apache 2.0

Variants

Name Parameters Notes
RynnValue-4B 4B
RynnValue-8B 8B

Paper

roboticsembodiedreward-modelreinforcement-learningopen-weight