Does VLA Even Know the Basics? Measuring Commonsense and World Knowledge Retention in Vision-Language-Action Models
Nikita Kachaev, Andrey Moskalenko, Matvey Skripkin, Nikita Kurlaev, Daria Pugacheva, Albina Burlova, Mikhail Kolosov, Denis Shepelev, Andrey Kuznetsov, Elena Tutubalina, Aleksandr I. Panov, Alexey K. Kovalev, Vlad Shakhuro
THE PROBLEM
This paper focuses on Modern Robot LearningVision-Language-Action model (VLA)A model that takes images and language as input and outputs robot actions.. This paper reveals a critical gap: Modern Robot LearningVision-Language-Action model (VLA)A model that takes images and language as input and outputs robot actions. models lose semantic and world knowledge when fine-tuned on Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. data compared to their source VLMs. Act2Answer lets you measure this knowledge degradation by grounding knowledge tests in physical Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. actions (object placement to answer questions), helping developers understand whether failures stem from bad knowledge or bad Control & PlanningControlThe method used to make the robot move the way you want.. Read the paper by tracking the Core ConceptsTaskThe job the robot is supposed to complete, such as pick-and-place, navigation, or drawer opening. definition, the Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. or data assumptions, and the evidence that supports the claimed improvement.
HOW IT WORKS
Task framing
Core method
Data and supervision
Evaluation evidence
KEY RESULTS
This paper reveals a critical gap: Modern Robot LearningVision-Language-Action model (VLA)A model that takes images and language as input and outputs robot actions. models lose semantic and world knowledge when fine-tuned on Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. data compared to their source VLMs. Act2Answer lets you measure this knowledge degradation by grounding knowledge tests in physical Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. actions (object placement to answer questions), helping developers understand whether failures stem from bad knowledge or bad Control & PlanningControlThe method used to make the robot move the way you want..
WHY DEVELOPERS SHOULD CARE
This paper reveals a critical gap: Modern Robot LearningVision-Language-Action model (VLA)A model that takes images and language as input and outputs robot actions. models lose semantic and world knowledge when fine-tuned on Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. data compared to their source VLMs. Act2Answer lets you measure this knowledge degradation by grounding knowledge tests in physical Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. actions (object placement to answer questions), helping developers understand whether failures stem from bad knowledge or bad Control & PlanningControlThe method used to make the robot move the way you want..
LIMITATIONS
The main limitation to check is whether the claimed behavior holds outside the paper's reported setup. That means testing across different Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. embodiments, scenes, objects, and data distributions.
WHAT COMES NEXT
The practical next step is independent reproduction with clear baselines, ablations, and stress tests. For a developer, the useful follow-up is to map the paper's Modern Robot LearningVision-Language-Action model (VLA)A model that takes images and language as input and outputs robot actions. assumptions onto a concrete Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. stack, then test the smallest version of the method that could run end to end.