Experts at the Toyota Research Institute see potential applications in removals and care
Researchers at the Toyota Research Institute (https://www.tri.global/) have trained the ‘Punyo’ robot provided by Boston Dynamics (https://bostondynamics.com/) to handle large objects with its entire body, similar to a human being. It is easy for the robot to lift something heavy onto its shoulder or a large box onto its back, provided it has the strength to do so. It uses not only its hands, but also other parts of its body such as its upper arms, hips and thighs as supports and levers.
Jug on the shoulder
Punyo can carry a large water jug on its shoulder or lift and hold a large box. To control its movements, the robot relies on feedback from its soft, pressure-sensitive skin and joint sensors.
The key to success is the softness of its body and the flexibility of its joints. ‘Adding any kind of adaptability is beneficial and leads to an average increase in the success rate of 206 per cent,’ say the developers. They present Punyo in the journal ‘Science Robotics’.
A lot of practice required
Furthermore, only minimal learning processes are necessary to teach Punyo its unusual abilities. The researchers used what is known as ‘example-guided reinforcement learning’. In a virtual environment, the team showed it what to do with just one demonstration. The robot then practised independently until it figured out how it worked.
‘A single remote-controlled demonstration recorded in a simulation is sufficient to train strategies for stylised, contact-rich movements,’ the team adds. This technology is an important step towards developing robots that make everyday life easier for many people.
For example, they can move large, unwieldy objects such as furniture into a flat during a house move or transport heavy packages safely and reliably in a warehouse. They are also helpful in the care sector, lifting people out of bed and washing them. And the robots do not need to be meticulously programmed, as they learn human-like skills from a single demonstration, the team concludes.
