New tool enables anyone to train a robot — no programming skills required

So far, training robots has required experts with specialised programming skills. We recently reported on a training centre for humanoid robots in China, which offered a glimpse into the role of a robot coach. However, engineers at MIT have now developed a new training method for robots. This enables users to teach robots in three intuitive ways. The special feature is that training can be carried out not just by programmers, but by anyone.

This approach, called 'Learning from Demonstration' (LfD), aims to enable anyone to train a robot. It uses a single sensor-equipped tool that makes training simpler and more flexible. The basic principle of this teaching method is as old as humanity itself: The robot observes an action performed by a human and must then execute it itself. Previous LfD methods generally fell into one of three categories:

• Teleoperation: The robot is controlled remotely (e.g. via a joystick). This approach is advantageous when handling toxic substances, for example.
• Kinesthetic training: The user physically moves the robot arm, which can be helpful when adjusting the positioning of heavy loads, for instance.
• Natural teaching: the human performs the task themselves while the robot observes and mimics. This is ideal for delicate, precise manoeuvres, such as drawing a logo.

The innovation: The Versatile Demonstration Interface (VDI)

Mike Hagenow's team at MIT, led by Professor Julie Shah, developed a three-in-one tool called the VDI. This makes it possible to apply any of the learning methods described above and could increase the types of both users and 'teachers' who interact with robots. It is a handheld, sensor-equipped control element that can be mounted onto common collaborative robot arms.

This attachment is equipped with a camera, position-tracking markers, and force sensors to measure pressure. The new tool was tested by volunteers performing tasks such as press-fitting (pushing pegs into holes) and moulding (shaping a dough-like substance around a rod). The researchers found that volunteers with manufacturing expertise generally preferred the natural teaching method.

The new tool can be used to train industrial robots such as the Walker S2. A video of this robot recently went viral on social media, causing concern. But the developers also see great potential for the VDI in areas such as home care and robots as household aids.

A good example of a household robot of this type is the 1X Neo, which is now available to pre-order in the USA and is scheduled to be delivered in 2026. Mike Hagenow, a postdoctoral researcher in the Department of Aeronautics and Astronautics at MIT, commented on the application areas of the teaching method:

We are trying to create highly intelligent and skilled teammates that can effectively work with humans to get complex work done. We believe flexible demonstration tools can help far beyond the manufacturing floor, in other domains where we hope to see increased robot adoption, such as home or caregiving settings.