This Slightly Haunting Childlike Robot Has Helped Scientists Crowdsource Research for Over a Decade
It’s called iCub and is able to do some pretty amazing feats.
Meet the iCub. Image: D.Farina / IIT
Giorgio Metta recently marveled as a 13-year-old successfully cleared a table. This may not sound like a particularly impressive feat, but Metta wasn’t observing any teen, he was watching his 13-year-old humanoid robot, iCub.
The original iCub was developed back in 2004 as a way to create a standardized, open-source software and hardware system for a humanoid robot. Metta, the vice scientific director at IIT-Istituto Italiano di Tecnologia and director of the iCub Facility, and his peers wanted to have a platform where researchers from all over the world could build on one another’s findings. He published a review of the iCub’s evolution in the journal Science Robotics Wednesday.
“We were researching robotics related to A.I. and needed a robot complex enough to be able to do vision, manipulation, and interact with humans,” Metta told me over the phone. “At that time, there was nothing available off-the-shelf. We decided to start designing a robot entirely by ourselves.”
Aside from being an impressive innovation that’s facilitated reams of research in A.I. over the last decade, the iCub is also a triumph for the open-source movement, showing how beneficial it can be to collaborate and share resources in the scientific community.
The robot they created is about the size of a five-year-old child, and able to crawl on all fours, sit up, walk, manipulate objects with its hands, see, hear, and even feel—it’s one of the only robots in the world with full-body “skin” sensors. Since the first iteration, the iCub has been updated twice, and enabled researchers across Europe, the US, Korea, Singapore, and Japan to study everything from language acquisition to fine motor skills.
“Having many users was useful because people were finding problems and improvements on the initial design,” Metta said. “We have people working on different topics like computer vision, and tactile learning.”
Over the years, researchers were able to share software changes and updates online, and build off one another’s findings, allowing for a more rapid development of both the iCub itself and the types of things researchers can learn from it. In that last year, 160 developers have added new code to the online database for iCub’s software, according to the paper. Metta wrote that all of the incremental improvements are pushing the project closer to one of its original goals: to create a robotic child. Going forward, each new development will help them inch closer to that dream.
Metta said it’s satisfying to see all the ways researchers are pushing the hardware and software to its limits. He’s seen the iCub move all elements of its upper body—head, neck, arms, shoulders—at the same time in a natural, fluid motion, something that would have been extremely difficult for the robot on Day 1. And then there’s the table clearing.
“We put together the work of many students into a complete application where the robot can clear the table: whatever you put on the table, the robot will pick it up and put it in a box,” Metta said. “It’s a nice moment to see the robot interacting and working with people.”
Get six of our favorite Motherboard stories every day by signing up for our newsletter .