Industrial robot arms, parallel grippers, force-detecting sensors and 3-D cameras successfully assembled an IKEA chair in 20 minutes. / Suárez-Ruiz et al. Sci. Robot. 3, eaat6385 (2018)
Commercial off-the-shelf robotic hardware can assemble an IKEA chair outside of a factory setting, scientists in Singapore have shown. The robots used independent and coordinated movements to complete a complex, multi-step task — qualities that have so far only been demonstrated for simpler tasks like picking and placing objects, welding or painting.
The robots, described in the April 18 issue of Science Robotics, also showed human-like dexterity to construct the chair, suggesting that these manufacturing machines may soon be ready for use in a wider range of applications, such as aircraft manufacturing, without needing special mechanical modifications or well-organized surroundings.
"The capabilities developed in this study — vision, planning, control, bimanual manipulation — all in unstructured environments [where the setting is not pre-planned], can be of use in virtually all manufacturing tasks, such as handling, drilling, glue dispensing, assembly, inspection and more," said Quang-Cuong Pham, assistant professor of mechanical and aerospace engineering at Nanyang Technological University in Singapore and co-author of the study.
The more skilled manufacturing robots become, the more benefits they may be able to offer humans, Pham noted. "Robots can potentially relieve humans from many tedious or dangerous tasks in manufacturing."
Though often second nature to humans, dexterity involves the mastery of various skills, including hand-eye coordination, the detection of forces and fine control of multiple movements.
"Fine manipulation is one of those key human skills," said Pham. "In general, robotics researchers are interested in reproducing these human skills in their robots."
Enter the IKEA chair. As anyone who has bought IKEA furniture knows, the company's products are made for do-it-yourself assembly and require a range of movement and agility — what Pham and his colleagues recognized as "a typical task that solicits all manipulation skills." Because the chair assembly was specifically designed for humans, without any consideration for robots, it would represent a truly random, uncontrolled challenge for any robots on the market.
The researchers scattered parts of a STEFAN-style IKEA chair around industrial robot arms, parallel grippers and force-detecting sensors — using two of each to mimic the hands and arms humans would use for such a task. A 3-D camera was placed in the middle of the robotic assembly as the robots' all-seeing eye. The researchers purposefully chose off-the-shelf hardware for their test to ensure that their methods could be easily deployed across industries.
After some trial-and-error, the robots successfully assembled the chair in approximately 20 minutes. Three major operations, involving fine coordination of sight and "touch," allowed them to complete the task. First, the robots were fed snapshots of chair parts with the 3-D camera, from which they could quickly and reliably identify correct parts in a randomly cluttered environment. The robots were even able to pinpoint and align objects like pins and holes, requiring precision at a scale of less than two millimeters.
The robots also coordinated fast, collision-free motions when grasping and piecing together all the chair parts. The force sensors mounted on their "wrists" allowed for detection of force changes as they gripped the chair pieces, to verify that pins slid into the correct holes, for example.
Their 20-minute achievement can be broken down into three seconds for object identification, around 11 minutes for motion planning and about nine minutes to construct the chair.
Is it possible that these robots assembled the STEFAN chair faster and better than we could? "Probably not," explained Pham. "Human 'hardware' is much better than industrial robot hardware. If human arms were rigid, such as our robot arms, and if human hands only had two fingers, such as our grippers, then humans would probably have a harder time than our robots assembling the chair."
While Pham and his team used a substantial amount of coding to program the robots' movements, they hope in the future that the robots will be advanced enough to construct the chair after a quick glimpse at a picture, for example.
"We are planning to integrate more AI [artificial intelligence] into this task so that the robot can automatically learn how to sequence the different subtasks of the assembly," said Pham. "Manufacturing was the first field of application of robotics, but the pace of development is still very high, with many new opportunities," he added.