Sign in to view source links and access this dataset
Description
230,000 annotated episodes for robot manipulation, built on top of the DROID and RH20T datasets. The annotations include subtasks, primitive skills, segmentation, gripper/object bounding boxes, placement proposals, affordance boxes, grasp poses, traces, and contact points. This dataset was developed by InternRobotics as part of the RoboInter project and was last updated in February 2026.
Use Cases
Training robot manipulation models based on dense per-frame annotations like bounding boxes and segmentation.
Developing skill segmentation algorithms based on annotated primitive skills and subtasks.
Researching robotic grasping and placement using provided grasp poses and placement proposals.
Analyzing object affordances and contact dynamics based on affordance boxes and contact point annotations.
Strengths
Annotations cover 230,000 episodes, providing a substantial scale for model training.
Includes a diverse set of annotation types such as bounding boxes, segmentation, grasp poses, and contact points.
Built on established robot manipulation datasets (DROID and RH20T), suggesting a foundation in prior work.
Limitations
Column-level documentation is absent; field semantics must be inferred after download.
Row count, file formats, and license information are unknown, which may limit suitability assessment.
Provenance
Source
InternRobotics, built on DROID and RH20T datasets.
Collection Method
Developed as part of the RoboInter project; specific annotation methodology not detailed.
Freshness
Last updated 2026-02-20 02:53:56; freshness should be verified.
License is unknown; users must verify terms of use before application.