Frontier & Open Problems
The reliability gap that keeps demos out of production, and the competing theses about closing it.
This domain overview lists 6 articles, ordered as the reliability gap first, because the entries after it are competing arguments about how to close that gap.
80% is a demo, 99.9% is a product: what deployment numbers actually show.
Contact-rich manipulation, the tactile sensing gap, in-hand reorientation, and deformables.
What the pi0.5 and pi0.7 results demonstrate, and what they do not: the open-world gap.
End-to-end scaling versus hierarchy versus world models versus RL fine-tuning, with falsification criteria.
Why this could be another robotics winter, and the milestones that would prove it wrong.
Industrial robotics can certify a control system but not a learned policy, so what ships is a verifiable safety layer wrapped around an unverifiable one.