ScaleVLA Uses a 552B VLM for Robot Policies and Reports Improved Manipulation Performance with Fewer Demonstrations

Written by

in

The ScaleVLA paper combines a 552B DeepSeek-V4.1-Flash VLM with a trainable action DiT to evaluate whether a larger vision-language backbone can reduce the demonstration requirements of robot policies.

Source:explainx.ai · Read Original

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *