Two robots from Nanyang Technological University in Singapore recently completed a task that most people are familiar with - the assembly of an IKEA furniture piece. The robotic project was developed by Assistant Professor Pham Quang Cuong and his group of engineers from the School of Mechanical and Aerospace Engineering.
The project used IKEA's Stefan chair, apparently constructed from pine and with an assembly instruction document that's eight PDF pages long. Francisco Suarez-Ruiz, Xian Zhou and Pham Quang-Coung published an article in Science Robotics last month that detailed the 'limits of robotics manipulation' through the automated assembly of the component. One goal was to take a project design only for human performance and use a robot to perform the task without special accommodations made for the robot.
Denso robots were used along with commercially available grippers, force sensors and cameras. The robot's methodology was to first look at all of the pieces, then plan the motions, and then perform the motions. Force control was used for instances where the robots' actual motions had to be changed from planned motions. The assembly took 20 minutes and 19 seconds - 11 minutes 21 seconds were used for motion planning, 3 seconds for localization and 8 minutes 55 seconds for the execution of the build. Most of the planning was done using Bidirectional Rapidly-exploring Random Tree algorithms in conjunction with the Robot Operating System.
I saw this build projects in a few places in April and early May but this week the brilliant writers at Deadspin released their take with an article called "The Deadspin Idiots Try to Build a Chair Faster Than a Chair-Building Robot". As a small spoiler to this article, the humans were not able to beat the robotic assembly champions. The Science Robotics article is a fascinating discussion of the methods and learning mechanisms that roboticists can use to teach more complicated human tasks. I look forward to seeing more robot assembly projects from Nanyang in the future.