IndustrialVLA-Bench: A Traceable Multi-Axis Evaluation of Open Robot Policy Models
arXiv:2609.25562v1 Announce Type: cross Abstract: Open robot policies increasingly follow two paradigms: vision-language-action models (VLAs) directly map observations and instructions to actions, whereas world-action models (WAMs) incorporate learned video or world dynamics into policy learning or…