clock menu more-arrow no yes

Filed under:

Watch a robot have an awful time building IKEA furniture

It's easy to fret over the impending robotic invasion, what with reports outlining the effects of robots driving our cars and taking our factory jobs and Elon Musk telling us all to prepare for annihilation. But it's good to remember that robots in their current form are nowhere near as sophisticated as people. In fact, they sometimes suck at even the simplest of tasks, like building IKEA furniture.

Researchers at Nanyang Technological University in Singapore decided to test a robot's capabilities in messy environments by having it try and construct an IKEA chair through trial and error, something current bots can't do. It's well known that modern robots are very good at doing one thing and doing it well, like assembling one portion of an automobile. It's a product of the same type of factory programming that lets robots package IKEA furniture, but it fails when those same machines are asked to build it. In the researchers' published paper, "A Framework for Fine Robotic Assembly," they detail how they helped a robot equipped with two grippers, force sensors to measure grip strength, and a six-camera vision setup overcome IKEA's hallmark peg-in-a-hole task.

Essentially, the team had the robot map out the furniture's surface by trying and failing to plug in the peg until it figured out the proper direction to move. So while it can't construct an entire chair, the robot is capable of performing one of the process's most complex tasks that humans happen to find easy.

What makes the Swedish furniture company's standard building process such a good test case is that it's a surprisingly good model of the human brain's sophistication. IKEA instructions use only pictures instead of words, relying instead on the human brain's knack for intuition and prevailing at a task even after you've lost one of those important-looking tall screws and installed the leg of the chair upside down. Roboticists are only just now beginning to marry robotic precision and task programming with vision and motion sensing capabilities. So we have a while to go before robots can build items as well as they can package them.