uuu-QuantMixed (see source-specific terms: CC-BY-4.0, Apache-2.0, MIT, project-authored)
Embodied Agent Arena — task data
A collection of 1,000 uniquely identified evaluation cases for embodied AI agents, spanning multiple benchmark tracks (W1-W5) with both fixed and interactive inputs from diverse source datasets and simulators.
Downloads313
Episodes1000
Why This Matters for Physical AI
This benchmark dataset evaluates embodied AI agents across diverse vision and manipulation tasks from multiple source datasets and simulators, providing standardized evaluation infrastructure for physical AI systems.
Technical Profile
- Modalities
- rgbdepthvideo
- Environment
- simulation
- Task Types
- perceptionnavigationmanipulationreasoning
- Episodes
- 1000
- Data Format
- JSONL
- Annotation Types
- language_instructionsbounding_boxessegmentationdepth_ground_truth
- License
- Mixed (see source-specific terms: CC-BY-4.0, Apache-2.0, MIT, project-authored)
Access
Need custom rgb data?
Claru builds purpose-built datasets for simulation applications with dense human annotations and quality assurance.
Request a Sample Pack