The week's most disruptive science, explained for humans.

Rapid Learning of Dexterous In-Hand Pen Writing through Real-Time Jacobian Estimation

An anthropomorphic robot hand writes with a grasped pen after about 18 seconds of on-robot Jacobian estimation — no analytic contact model, simulation training, or collected demonstrations — and hits sub-millimeter in-plane precision.

• ▸What: The authors control in-hand pen writing by estimating a task Jacobian of the combined hand–object system in real time on the physical robot, then using that estimate to write single-stroke letters and shapes in air and on paper.
• ▸Why it matters: Abundance angle: reliable dexterous physical work is still scarce, expensive labor or capital. A laptop-CPU controller that adapts online without sim or demo datasets is a step toward cheaper shared dexterity — not a claim that factory hands become a default on a calendar.
• ▸Who should care: Robotists working on contact-rich manipulation, people who collect or simulate dexterous demos, and operators watching whether simple online estimators can beat data-heavy RL/IL for some in-hand skills.