RTR Robot Sys: Robot Manipulation Training Datasets for Latent-Space Policies

Training datasets for the paper "Learning High-Frequency Continuous Action Chunks in Latent Space". The datasets are used to train latent-space high-frequency policies and the Reuse-then-Refine chunk-refinement strategy. They were created by author sadpiggy and last updated on 2026-05-27.

Use Cases

• Train latent-space high-frequency policies for robotics based on the described imitation learning approach.
• Develop and evaluate the Reuse-then-Refine chunk-refinement strategy for robot action sequences.
• Benchmark robot manipulation algorithms using the provided training data.
• Research in diffusion policy applications for continuous control tasks in robotics.

Strengths

• Dataset size is indicated to be between 1,000 and 10,000 records.
• Associated with a specific, peer-reviewed research paper and methodology.
• Last update timestamp is provided (2026-05-27).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Author sadpiggy on Hugging Face.

Collection Method

Created for training models in the associated research paper.

Freshness

Last updated 2026-05-27 04:36:53.