UP-VLA: A Unified Understanding and Prediction Model for Embodied Agent

Open paper

Benchmark	Model	Rank	Results
robot-manipulation-on-calvin	UP-VLA	–	avg. sequence length (D to D): 4.08