Imagine A Robot With An Imagination Of Its Own

Machines now think in very constrained ways. But scientists are trying to instill our robot-friends with a little more abstract thought. Imagine the implications, before they can, too.

Imagine A Robot With An Imagination Of Its Own
[Image via Shutterstock]

Robots can think and reason increasingly well, but can a robot have an imagination? Once the answer would have been: Hey, even a machine can dream.


That sci-fi vision is now moving closer to reality in research labs.

First, imagination needs to be defined. It’s a slippery problem. One company claimed that it created an “imaginative robot,” but this was really just facial detection software applied to a situation that seemed like human imagination: Gazing at clouds in the sky, the robot could pick out the ones that look like faces.

A group of Spanish roboticists have created something a little closer to the real thing in their lab, presenting their findings in a paper at a major robotics conference in Japan this month. “How are robots going to transform the world if they can’t even imagine how?,” is the question asked at the end of the talk given by the researchers from the Universidad Carlos III de Madrid.

They created robotic software that goes beyond recognizing patterns towards actually envisioning and creating. The “robot imagination system” allows a robot to visualize and draw an object that researchers described to it out loud–and, here’s the key–it was an object the robot had never seen before.

Of course, it wasn’t able to make up a shape out of thin air. Rather, it could create something new based on similar objects it had seen in the past. The researchers trained their system with 300 computer generated images of shapes of different colors and positions on a grid, and gave each image a seven word description. Based on this training, the robot could construct a “mental model” of an object that was described but not yet seen and then sketch it out. The results were reasonably accurate, especially for objects with fewer features specified.

“You could actually show the algorithm something on the top, something on the top right, something on the left, something on the bottom left, and the system would be able to imagine something that’s on the top left,” says Juan G. Victores, the PhD researcher who is the lead author of the paper.


The researchers imagine (in the real, human way) a number of useful applications for their robot’s form of imagination. It could make household robots smarter by giving them the ability to recognize and retrieve items based on more natural language interactions. A person could ask a robot to fetch her old black jacket, and the robot might be able to find it without having ever seen that or any other black jacket in the past. These are basic tasks a child with an imagination could do, but that a commercial robot would struggle with today.

A next step for the researchers is incorporating more types of data from different kinds of sensors and actually testing their robots’ imagination against young children–for example, asking a child to touch a small green circle and a robot to do the same. In this way, the “mental age” of the robot could even be calculated. They are also looking at expanding the imagination system beyond object visualization to robot actions and decision-making.

“The authors consider the research on developing robot imagination systems a natural step in robotic research and cognitive science,” they write.

About the author

Jessica Leber is a staff editor and writer for Fast Company's Co.Exist. Previously, she was a business reporter for MIT’s Technology Review and an environmental reporter at ClimateWire.