MIT researchers channel AI to turn hand gestures into robot training data

Humanoid robots struggling with tasks similar grasping a cupful person a caller teacher — a idiosyncratic wearing an ultrasound wristband that captures the question of muscles, tendons and ligaments beneath the skin

CAMBRIDGE, Mass. -- Humanoid robots struggling with tasks similar grasping a cupful person a caller teacher — a idiosyncratic wearing an ultrasound wristband that captures the question of muscles, tendons and ligaments beneath the skin.

Researchers astatine the Massachusetts Institute of Technology developed the instrumentality to cod information of quality manus question that could yet assistance robots execute the dexterity that has been hard for machines to master.

“Imagine radical doing housework,” said Xuanhe Zhao, an MIT prof of mechanical engineering. “We tin usage the information obtained by our strategy to bid a robot to bash precisely (that) housework with this dexterous manus motion.”

As overmuch of the tech satellite is inactive captivated with artificial quality assistants that are taking connected computer-based tasks, Zhao is among the scientists trying to imbue AI with much sensory information from the carnal world.

Beyond housework, the exertion could assistance with different tasks that necessitate flexing fingers and hands, specified arsenic surgery.

The wristband uses high-frequency dependable waves to “see” done its wearer's skin. It relays images of the musculus and tendon movements to a machine that uses AI to alteration a adjacent robotic manus to mimic the gestures.

An AI algorithm is trained to decode images generated by the instrumentality into what engineers telephone degrees of state – circumstantial ways a associated tin crook oregon rotate. The quality manus has 22 of them.

In earlier systems, tracking adjacent a fraction of those movements was a important challenge.

In laboratory demonstrations with 8 volunteers, developers showed the wristband could precisely reflector manus gestures – including each 26 letters successful American Sign Language – wrong 120 milliseconds.

The wristband tin run wirelessly, meaning the controlling idiosyncratic and the receiving robot request not beryllium successful the aforesaid room.

Beyond distant control, the squad sees a way toward utilizing the wristband to physique immense datasets of quality question that could yet alteration humanoids to larn dexterous tasks without quality guidance.

___

AP Technology Writer Matt O'Brien contributed to this report.

Sponsored Content by Taboola