XGQ12345apache-2.0
Grounded in Time (GiT)
A benchmark and dataset for temporal grounding under perceptual ambiguity in robotic dual-arm manipulation, where object references cannot be resolved from individual frames alone but require tracking temporal structure across episodes.
Downloads18
Episodes4680
Why This Matters for Physical AI
This dataset addresses temporal reasoning in robotic manipulation by requiring models to ground language instructions through episode history rather than single-frame perception, advancing embodied AI's ability to handle perceptual ambiguity.
Technical Profile
- Modalities
- rgbdepthproprioception
- Robot Embodiments
- AgileX dual-arm
- Environment
- labsimulation
- Task Types
- manipulationtemporal_grounding
- Episodes
- 4680
- Data Format
- HDF5
- Annotation Types
- language_instructionstemporal_grounding_annotationssubtask_labels
- License
- apache-2.0
Access
Need custom rgb data?
Claru builds purpose-built datasets for lab applications with dense human annotations and quality assurance.
Request a Sample Pack