arXiv Robotics — research abstracts· Bochen Yang, Lianlei Shan·· 1 days agoEditorial score22
PearlVLA: Progressive Latent Plan Refinement for Embodied Action
PearlVLA: Progressive Embodied Action-Plan Refinement in Latent Space
Summary
PearlVLA refines latent plans using feedback from predicted outcomes. It improves action generation while balancing latency and computational cost. Experiments show competitive performance on LIBERO and RoboCasa.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
Source：arXiv Robotics — research abstracts · arxiv.org