Biscue5/egoscaler-v2
Robotics Dataset
EgoScalerV2 is a large-scale egocentric video dataset for training vision-language-action (VLA) models, providing 6DoF object trajectories, egocentric visual observations, and natural-language action descriptions in LeRobot v2.0 format, with over 45k episodes of human activities.
Catalog metadata
- Repository owner
- Biscue5
- Source
- Hugging Face / LeRobot
- Robot
- human activity
- Episodes
- 45157
- Frames
- 1409418
- Cameras
- 1
- Frames per second
- 20
- License
- apache-2.0
Example tasks
- pull out the test pack from the covid test pack.
- pick the hammer from the table.
- pass the ladle in her right hand to her left hand.
- pick the chopstick on the bowl.
- drop the needle nose plier on the workbench.#unsure.
- pick the test swab packet from the table.
- pick the spoon from the frying pan.
- close the cabinet door.
- close the low part of the fridge.
- dry off the plate.
Browse the catalog to compare robots, tasks, sensors, and licenses. Sign in to open protected download and model links.