Evo-1: Lightweight Vision-Language-Action Model with Preserved Semantic Alignment
Summary
Evo-1 studies how to keep semantic alignment strong while making vision-language-action models lighter and more practical for embodied tasks.
My Contributions
- Supported simulation data processing
- Adapted the model and benchmark environments
- Ran most of the simulation experiments
- Participated in result organization and analysis
Status
This work has been accepted by CVPR.