FastUMI-100K: Over 100,000 Robotic Manipulation Demonstrations for 54 Tasks

FastUMI-100K provides over 100,000 demonstration trajectories for robotic manipulation learning. The dataset covers 54 diverse tasks and includes multi-view wrist-mounted fisheye images and high-frequency end-effector states. It was created by IPEC-COMMUNITY and last updated in April 2026.

Use Cases

• Training imitation learning models based on the 100K+ demonstration trajectories.
• Developing vision-based robotic control policies based on the multi-view wrist-mounted fisheye images.
• Benchmarking robotic manipulation algorithms across the 54 diverse tasks.
• Researching multimodal learning for robotics based on the combination of image and state data.

Strengths

• Contains over 100,000 demonstration trajectories.
• Covers 54 diverse manipulation tasks.
• Includes multimodal data with images and high-frequency end-effector states.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect bias inherent to the specific demonstration collection method.

Provenance

Source

IPEC-COMMUNITY

Collection Method

Likely contains demonstration trajectories collected via robotic simulation or teleoperation.

Time Range

null

Freshness

Last updated 2026-04-16 15:07:15; freshness should be verified.

Geography

null