Asharq Al-awsat English Middle-east and International News and Opinion from Asharq Al-awsat Newspaper

Scientists Teach Robots Curiosity

Scientists Teach Robots Curiosity

Tuesday, 4 June, 2019 - 05:15
Humanoid intelligent robot Alpha developed by UBTECH. (photo credit: REUTERS)
Vienna - London - Asharq Al-Awsat
When a robot finds itself in front of something it cannot recognize, it stands still. But researchers from Vienna want to change that by integrating "curiosity" into robots.

Traditional robots can perform detailed duties in a known environment with remarkable precision and quickness.

But many of these systems behave literally like a cow when it stands in front of a new gate, when it is not used to deal with something, or more precisely, when it is not programmed to deal with this thing, according to German news agency.

But when it becomes necessary to give robots more complex duties in different circles, these machines need to be able to operate autonomously along with what we call curiosity.

Researchers led by Markus Vincze from the Vienna University of Technology (TU Wien), are working on a project funded by the Austrian Science Fund (FWF), aiming at enabling robots to expand their own horizons.

But researchers say it's primordial that robots first recognize that a certain thing is an object, "which can be really difficult when several objects cannot be separated from one another because they overlap in part," Vincze said.

When the system gets a picture of a certain body, it needs to form a three-dimensional model of it, as it is the only way the robot can touch and lift the body. A statement by the Austrian Science Fund said that while children have an innate ability to get to recognize objects spatially, doing such a thing with robots requires a tremendous effort.

The Vienna Research Institute particularly focuses on what happens when the device does not recognize an object, or when the robot compares a picture of the object it sees with its database, but does not find a match. This means that this machine has to learn what it does not know.

If this happens, "the robot has to form a picture of this thing and start looking for it on the Internet," says the main idea of the researchers.

Researchers have been able to improve search algorithms step-by-step by relying on the available images and definitions of images in Google, for instance. The researchers, in collaboration with partners from Italy, France and the UK, experimented with the HOBBIT robot, a machine designed for use in old people's homes for chores such as finding things that have been mislaid.

The robot was practically tested by tackling 10 desktop gadgets, such as keyboard, mouse, and a hole punch. The researchers then deleted one of these objects from the robot's database, and HOBBIT had to look for information about it.

The robot was more successful when the object being asked about was close to other things that fit the same subject.

When the researchers used an example of table utensils, and the object the robot was looking for was also a kitchen tool, the system did not struggle to find and classify it. "Such contextual information can be analyzed and used in a targeted manner and thus narrow down the search," Vincze said.

Editor Picks