Robots can learn from a single example but with notable limits
Teaching a robot from one example can start a reusable program and built-in checks before use.
- Publication
- arXiv
- Stage
- Preprint
- What we read
- Summary of the paper
- Authors
- Yuyao Liu, Jiayuan Mao, David Hsu, Leslie Pack Kaelbling, Tomás Lozano-Pérez
- Universities and research institutions
- Not yet supplied in verified metadata; the Brief does not guess.
What they did and found
From one human demonstration, the system creates simple robot actions and then fuses them into a plan that can adapt to new work layouts. It was tested in simulations and on real hardware with eight tasks, showing some generalization but also failures.
Why it matters
The process uses a loop of make, test, and adjust. It relies on a shown task and checks to avoid fitting only to one scene.
From one demo, the system creates simple actions and combines them into a plan that fits new layouts while keeping the shown steps intact.
In practice, you still need careful checks in different scenes; busy or unexpected environments may need more demonstrations or tweaks for reliability.
Original sources · 1
- RAPID: Robot Agentic Programming from Demonstrations ↗arXiv · 2026-09-24
Check the original paper for its authors, methods, version and access terms.
What remains uncertain
Generalization varies by task; some scene changes are still unreliable and real-world edge cases may not be captured.
What is your take?
Ask a question, add useful context or share a different perspective. Keep the conversation respectful and grounded.