Reproduce “GRACE: Generalizing Robot-Assisted Caregiving with User Functionality Embeddings” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ziang Liu, Yuanchen Ju, Yu Da, Tom Silver, Pranav N. Thakkar, Jenna Li, Justin Guo, Katherine Dimitropoulou and 1 more
GRACE predicts personalized functional range of motion (fROM) for caregiving robots using clinical assessment scores, eliminating the need for motion capture for new users. The approach demonstrated fROM prediction accuracy comparable to ground truth models for unseen users and significantly enhanced task success, user agency, physical effort, and safety in real-robot experiments.
Key insights
- Standardized occupational therapy assessment scores can serve as a scalable and accessible proxy for predicting an individual's detailed functional range of motion, overcoming motion capture limitations.
- Learning a latent representation of user functionality from clinical assessments enables a robot to generalize personalized behaviors across unseen users and diverse mobility conditions.
- Tailoring robot assistance based on predicted fROM leads to a controllable trade-off between maximizing user agency and ensuring task success, significantly improving user experience in human-robot interaction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.