VISTA-UMI-5K: 5,000 Verified Robotic Manipulation Trajectories

VISTA-UMI-5K is a verified robotic manipulation dataset containing 5,000 UMI trajectories across 30 tabletop manipulation tasks. The dataset was created by TeleEmbodied and last updated on 2026-06-04. All trajectories are validated through a physical-validation pipeline and verified for replay on three robot embodiments.

Use Cases

• Training robot manipulation policies based on high-quality, physically validated demonstrations.
• Benchmarking robot learning algorithms across different embodiments based on the verified replay data.
• Developing vision-grounded action models based on the multimodal trajectory data.
• Studying the transferability of manipulation skills across different robot platforms.

Strengths

• Contains 5,000 high-quality UMI trajectories.
• Covers 30 distinct tabletop manipulation tasks.
• Trajectories are validated for replay on three specific robot embodiments: RealMan rm7500, ARX AC one, and Galaxea R1 Pro.

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

TeleEmbodied via Hugging Face.

Collection Method

Contains UMI demonstrations validated through a physical-validation pipeline.

Freshness

Last updated 2026-06-04 06:14:12; freshness should be verified.