AGNOSTOS

AGNOSTOS is a novel simulation benchmark that contains 23 unseen manipulation tasks designed to rigorously evaluate the cross-task zero-shot generalization capabilities of vision-language-action models in robotic manipulation.

Papers using AGNOSTOS (2)

Decompose and Recompose: Reasoning New Skills from Existing Abilities for Cross-Task Robotic Manipulation2026Exploring the Limits of Vision-Language-Action Manipulations in Cross-task Generalization2025