glory-hyeokother

synth100

100 generated video clips for validating Pre-Contact Level Filtering, derived from Fourier Intelligence's ActionNet dataset with counterfactual instructions where the generator is asked to fetch different objects than those in the real episodes.

Downloads33
Episodes100

Why This Matters for Physical AI

This dataset enables validation of vision-based contact prediction and occlusion detection models for robotic manipulation with counterfactual instruction grounding, providing a challenging negative control for evaluating pre-contact filtering in embodied AI systems.

Technical Profile

Modalities
rgbdepthproprioceptionlanguage
Robot Embodiments
Fourier GR1-T1
Action Space
joint_positions
Environment
lab
Task Types
manipulationgraspingpick_and_place
Episodes
100
Data Format
npz
Annotation Types
language_instructionssegmentationbounding_boxesaction_labels
License
other
Part of the ActionNet family

Access

Need custom rgb data?

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

Request a Sample Pack

Related Datasets