KUKA Robot Builds Ikea Furniture Via Kinesthetic Teaching
Demonstration method triggers stiff and compliant behavior from the robot for on-line collaboration
By RoboticsTrends' News Sources - Filed Apr 23, 2013

Researchers from the Department of Advanced Robotics at the Italian Institute of Technology and the Institute of Robotics and IndustrialInformatics (News - Alert) at the Spanish National Research Council/Polytechnical University of Catalonia have developed a robot capable of learning by imitation, and have demonstrated this capability by teaching it to help build an IKEA table.

In a video released by Leonel Rozo and Sylvian Calinon, two users are shown demonstrating an assembly skill requiring different levels of compliance (in this case, building IKEA furniture). Each piece has unique construction characteristics that need to be translated to the robot. Rather than manually programming the robot for each one, the robot can instead learn the requisite construction skills by being hand guided through a demonstration. 

Through a process called kinesthetic teaching, a user grasps the robot and moves it by hand to demonstrate how it should work alongside human users. A single force sensor combined with a marker-based vision tracking system allows the robot to record the precise positioning, orientation, and force information needed to cooperate with its handlers at each step. The ultimate goal of this technology is to eventually develop robots that not only replicate tasks on their own, but also interact with humans in safe, natural ways.

According the project’s website, “After demonstration the robot learns that it should first be compliant to let the user re-orient the table top in a comfortable pose to screw the corresponding table leg. Once the user starts to screw the leg, the robot becomes stiff to facilitate the task. This behavior is not pre-programmed, but is instead learn[sic] by the robot by extracting the regularities of the task from multiple demonstrations.”

Click here to reach the full paper, published in Proceedings of the AAAI Conference on Artificial Intelligence.

<< Return to story