HAGS: Hand and Glove Segmentation for DOE Glovebox Human-Robot Collaboration

A dataset of 2876 annotated frames and 16 videos per participant, collected in a Department of Energy glovebox environment. It features ten participants performing two collaborative tasks with a robot arm under four conditions, recorded from two camera angles. The dataset was created by Shivansh Sharma and last updated on 2026-04-27.

Use Cases

• Training segmentation models to detect hands and gloves based on annotated frames.
• Developing safety systems for human-robot collaboration based on video footage of interaction tasks.
• Evaluating model robustness using in-distribution and out-of-distribution frame splits.
• Studying the impact of environmental variables like glove use and green screens on perception algorithms.

Strengths

• Contains 2876 annotated frames, providing a substantial basis for model training.
• Includes 8 hours of video footage capturing diverse interaction scenarios.
• Features data from ten participants across two distinct experiments, enhancing variability.
• Each experiment was recorded from two camera angles (top and side view), offering multiple perspectives.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The dataset's scale and specific file formats are not detailed in the provided metadata.

Provenance

Source

Texas Data Repository Harvested Dataverse

Collection Method

Video data collected in a standard DOE glovebox using a GoPro and an Intel RealSense camera, with a pre-programmed UR3e robot arm. Frames were sampled and annotated.

Freshness

Last updated 2026-04-27 04:14:41; freshness should be verified.