Research paper
SAVED · Constraining robot exploration with demonstrations
SAVED uses a small set of imperfect demonstrations and task-completion feedback to guide model-based reinforcement learning. The authors evaluate constraint satisfaction in six navigation/manipulation simulations and a knot-tying task on a da Vinci robot.
Reviewed 2026-11-05 · Source and bibliography checked
- Original title
- Safety Augmented Value Estimation From Demonstrations (SAVED): Safe Deep Model-Based RL for Sparse Cost Robotic Tasks
- Publisher
- Institute of Electrical and Electronics Engineers (IEEE)
- Authors
- Brijen Thananjeyan, Ashwin Balakrishna, Ugo Rosolia, Felix Li, Rowan McAllister, Joseph E. Gonzalez, Sergey Levine, Francesco Borrelli, Ken Goldberg
- DOI
- 10.1109/lra.2020.2976272
- Publication date
- 2020-04
- Source language
- en
- Edition
- IEEE Robotics and Automation Letters; 2020; DOI 10.1109/lra.2020.2976272
- Access
- Publisher metadata and open author manuscript
- Publication status
- Published
Related resources and engineering notes
Source overview
SAVED uses a small set of imperfect demonstrations and task-completion feedback to guide model-based reinforcement learning. The authors evaluate constraint satisfaction in six navigation/manipulation simulations and a knot-tying task on a da Vinci robot.
Use and review
Checked bibliography and author-archive abstract only. Reported task success does not establish clinical safety or universal constraint satisfaction; demonstrations, model uncertainty and probabilistic constraints require application review.
Sources & editions
Safety Augmented Value Estimation From Demonstrations (SAVED): Safe Deep Model-Based RL for Sparse Cost Robotic Tasks ↗
Institute of Electrical and Electronics Engineers (IEEE) · IEEE Robotics and Automation Letters; 2020; DOI 10.1109/lra.2020.2976272
IEEE RA-L bibliography; author archive abstract https://arxiv.org/abs/1905.13402v8. Journal date 2020-04 retained, distinct from archive posting.