Mingyu Ding

By decomposing long-horizon manipulation into transport and object-centric interaction, LiLo-VLA achieves state-of-the-art zero-shot generalization and robustness, outperforming end-to-end VLA models by a large margin.

VLAs often ignore your instructions and just do what they've seen before, but a simple "counterfactual comparison" trick can fix it.

Unlock zero-shot generalization to new robot hand designs by representing diverse dexterous hands in a unified, canonical space.