EgoSteer/EgoSteer-RealWorld

Robotics Dataset

EgoSteer-RealWorld is a large-scale real-robot bimanual dexterous teleoperation dataset with 54,454 episodes (192 hours, 20.75M frames) across 193 tasks, collected on a RealMan dual-arm robot with two Ruiyan dexterous hands and head/chest RGB-D cameras, released in LeRobot v3.0 format.

Catalog metadata

Repository owner
EgoSteer
Source
Hugging Face / LeRobot
Robot
egosteer_realman_bimanual
Episodes
54454
Frames
20750677
Cameras
4
Frames per second
30
License
apache-2.0

Browse the catalog to compare robots, tasks, sensors, and licenses. Sign in to open protected download and model links.

Explore more robotics datasets and models