MIT has introduced a universal interface to teach robots without code

MIT researchers have developed a universal interface to teach robots unnecessary. The solution supports three basic methods of demonstration training: Remote control, direct physical manipulation and observation of human actions.

The team called it a “universal demonstration interface” (VDI). The device is attached to the hand of a collaborative work. It is equipped with a camera and pressure sensors that fix the movement and effort in real time. The device allows you to control the robot through the joystick, manually move its limb or perform the action yourself – the robot will remember and reproduce what you see.

The development was protested together with production professionals. Volunteers trained the work on two typical factory tasks: installation of pins in the holes (press-fitting) and the formation of material around the rod (molding). In each case, they used all three teaching methods.

According to the results, most users liked the “natural demonstration” – when a person simply performs VDI tasks in their hands and the robot watches. But each method has advantages in certain scenarios: telemanipulation – for working with toxic substances, kinesthetics – for heavy objects, natural learning – for accurate actions.

“We believe that flexible demonstration tools can help far beyond production, in other areas where we hope to see an increase in the introduction of robots, such as households or care facilities,” said MIK Gageno development co -author.

The presentation will be held in October at the IEEE IROS conference. The work is partially funded by the MIT Postdoctor Fellowship Program for Engineering Excellence and the Valenberg Foundation.

Recall that in June, Google Deepmind presented the optimized version of the GEMINI ROBOTICS Shi model, which can work directly at work without internet connection.

The gun