This egocentric video dataset captures open-and-close actions performed on a range of everyday objects in ordinary living environments, recorded for Physical AI model training.
Potential Use Cases
- Manipulation Policy Training: One action class repeated across many object types gives robotic policies the variation needed to generalize opening and closing behaviour to objects not seen in training.
- Affordance Learning: Because the same intent is executed on differently shaped handles, lids, and doors, the footage supports learning which part of an object affords the action and how it must be approached.