Beijing, China – Researchers at Tsinghua University have developed a method allowing robots to learn grasping skills directly from humans, tackling one of robotics’ toughest challenges: adaptive handling of objects with different textures, weights, and fragility.

“Humans use tactile sensation to recognize grasping states in real time and instantly fine-tune grip according to grasping states,” said Prof. Rong Zhu, corresponding author of the study. “We want to give robots that similar ability — to sense, cognize, and act on tactile feeling in real time.”

The team created a tactile glove equipped with sensors to capture contact, slip, and pressure during human grasping. Using a sensory-control synergy framework inspired by human neurocognition, the system encodes these tactile signals into semantic states like “stable,” “slightly unstable,” or “highly unstable.”

This abstraction allows robots to generalize grasping skills across diverse objects without thousands of precise measurements. A fuzzy logic controller mimics human decision-making, adjusting grip strength according to real-time feedback.

Once transferred to robotic hands with sensors, the approach enables robots to grasp slippery umbrellas, fragile eggs, and heavy bottles with an average success rate of 95.2%. In dynamic tests, robots resisted external pulls and prevented slips autonomously.

The researchers also demonstrated a robot hand-brewing coffee, completing each step — from scooping coffee powder to stirring and serving — by applying tactile-based control to manage uncertainties.

“Robots learn universal grasping through understanding sensory and control logic behind sensing data rather than imitating human motion trails,” Prof. Zhu explained. “We teach robots the ability to draw inferences from one instance.”

This human-taught sensory-control synergy represents a breakthrough toward adaptive, intelligent robots capable of handling real-world tasks.

Podcast: Play in new window | Download