R3D is a dataset of pre-processed trajectories for challenging 3D manipulation benchmarks, including RoboTwin and ManiSkill. The data supports the R3D framework for robust and scalable 3D imitation learning, developed by eddie-cui and last updated on Hugging Face in April 2026. The dataset is intended for research into training stable 3D policies using a transformer-based 3D encoder and a diffusion decoder.
Use Cases
- Training 3D imitation learning models based on provided manipulation trajectories.
- Benchmarking policy learning algorithms on the RoboTwin and ManiSkill environments.
- Developing robust 3D encoders and diffusion decoders based on the described R3D architecture.
- Investigating causes of training instability in 3D policy learning based on the dataset's trajectories.
Strengths
- Data is pre-processed for immediate use in training.
- Focuses on challenging manipulation benchmarks like RoboTwin and ManiSkill.
- Supports a specific research framework (R3D) for scalable 3D imitation learning.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count, file formats, and license are unknown, which may limit suitability assessment.
- Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
- Source
- Hugging Face user eddie-cui.
- Collection Method
- Pre-processed trajectories from manipulation benchmarks.
- Time Range
- null
- Freshness
- Last updated 2026-04-18 03:47:36.
- Geography
- null