2557 exported frame pairs derived from 20,918 source replay frames of a Unitree Go2 robot. The dataset is designed for the WorldForge score contract, linking current robot-view images and candidate action deltas to predicted future visual latents. It was created by user 'espejelomar' and last updated on Hugging Face in May 2026.
Use Cases
- Training world models to predict future visual states based on current robot-view images and egomotion actions.
- Benchmarking scoring functions that compare predicted future visual latents against goal latents.
- Developing and testing visual foresight algorithms for robotic control systems.
Strengths
- Derived from a specific, publicly available source of 20,918 replay frames from the dimensionalOS/dimos Unitree Go2 assets.
- Provides 2,557 structured frame pairs designed for a specific prediction task (WorldForge score contract).
- Includes a defined split with 1,791 training and 383 test examples.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count for the full dataset is unknown, which may limit suitability assessment.
- Description metadata is limited; actual data quality and content require manual inspection after download.
Provenance
- Source
- Public dimensionalOS/dimos Unitree Go2 replay assets.
- Collection Method
- Derived and processed for the WorldForge score contract.
- Freshness
- Last updated 2026-05-28 11:38:14; freshness should be verified.