The global site of the UK's leading magazine for automation, motion engineering and power transmission
29 March, 2024

LinkedIn
Twitter
Twitter link

Robots learn how to grasp unfamiliar objects reliably

23 May, 2014

Researchers at the University of Birmingham in the UK have developed a way of teaching robots to pick up unfamiliar objects without dropping or breaking them.

The research paves the way for robots to be used in more flexible ways and in more complex environments. These could include manufacturing and packaging industries where a wide variety of different tasks need to be undertaken, and where humans and robots have to be able to work together.

It is already fairly common to program robots to pick up particular objects and move them around. But when the objects vary in size or shape, robots tend to get clumsy.

Researchers in the University’s School of Computer Science have designed a way of programming a robotic hand to pick up an object and then using information learned in that first grip to grasp and move a range of similar objects.

The researchers taught the robot a specific grasp type – for example, a power grip, which uses the whole hand to curve around an object, or a pinch grip, which uses two or three fingers. The robot was then able to generalise the grip and adapt it to other objects.

“Current robot manipulation relies on the robot knowing the exact shape of the object,” explains Jeremy Wyatt, Professor of Robotics and Artificial Intelligence at the University. “If you put that robot into an unstructured environment – for example, if it is trying to pick up an object amongst clutter, or an object for which it doesn’t already have an exact model – it will struggle.

“The programming we have developed allows the robot to assess the object and generate around 1,000 different grasp options in about five seconds," he adds. "That means the robot is able to make choices in real time about the best grasp for the object it has been told to pick up and it doesn’t need to be continually retrained each time the object changes.”

The robotic hands used by the team look similar to human hands, with five jointed fingers. However, the programming technique would also work with robots that had other types of hand, such as pincer grips.

Alta Innovations, the University of Birmingham’s technology commercialisation office, is now looking for partners interested in licensing the technology. The University is already working with several companies that are keen to incorporate the technology into their processes.

The University of Birmingham research was carried out within the PaCMan (Probabilistic and Compositional Representations for Object Manipulation) consortium, funded by the European Union. The University is leading the consortium, which also includes the Università di Pisa in Italy and the Universität Innsbruck in Austria.




Magazine
  • To view a digital copy of the latest issue of Drives & Controls, click here.

    To visit the digital library of past issues, click here

    To subscribe to the magazine, click here

     

Poll

"Do you think that robots create or destroy jobs?"

Newsletter
Newsletter

Events

Most Read Articles