sungkyunner

UniVTAC SmolVLA Baseline Evaluation

Closed-loop evaluation of a tactile-free SmolVLA baseline trained on 800 UniVTAC Isaac45 demonstrations, where tactile observations are simulated but never sent to the policy.

Downloads4
Episodes800

Why This Matters for Physical AI

This evaluation demonstrates the effectiveness of vision-language-action models for robotic manipulation in simulation, establishing baseline performance metrics for multi-task closed-loop control without tactile feedback.

Technical Profile

Modalities
rgbproprioceptionlanguage
Action Space
end_effector_delta
Environment
simulation
Task Types
manipulationgraspingpick_and_placeinsertion
Episodes
800
Annotation Types
language_instructionsreward_labels
Part of the UniVTAC family

Access

Need custom rgb data?

Claru builds purpose-built datasets for simulation applications with dense human annotations and quality assurance.

Request a Sample Pack

Related Datasets