OpenRoboCare
EmPRISE Lab at Cornell University
315 sessions of occupational therapists doing caregiving tasks, with video, touch, and gaze.
A multimodal dataset of expert demonstrations: 21 occupational therapists performing 15 daily caregiving tasks across 315 sessions, totaling 19.8 hours, with synchronized RGB-D video, tactile sensing, eye-gaze tracking, pose annotations, and action labels. For teaching robots to assist with caregiving.
Access
- Cost
- FreeNo charge to download; hosted in an AWS S3 bucket.
- License
- Open, non-commercial onlyBSD-3-Clause; academic and non-commercial use permitted
- Sign-up
- No sign-up
- How you get it
- API and Bulk downloadsee documentation
What's in it
- One row is
- One caregiving session with synchronized sensor streams
- Covers
- Not geographic
- Years
- Not stated
- Updated
- One-time release
- Size
- 315 sessions, 19.8 hours
Quality
Valuable and detailed for robotics, but small and narrow in participants and tasks.
- Small participant pool
- Non-commercial terms
Connects to
- Yale-CMU-Berkeley (YCB) Object and Model Set: Object models for manipulation research
- RACECAR Dataset: Another robotics-focused dataset on the registry
Questions it could answer
- Which tasks need the most tactile information?
- Where do experts look before they act?