Robots. We make them pick up our trash, beat Captcha, solve Rubik’s cubes, and pass the butter. We probably shouldn’t teach them how to wield weapons, lest our uppance come, and yet humans are still teaching robots how to stab us, and now roboticists at the New Jersey Institute of Technology have taught one the martial art of nunchaku. Oh. Good. Great. We needed that.
There actually is a method to this madness. As lead researcher Cong Wang explained to Inverse, this robot could also help steal your job. (Good. Great. We needed that.) It often takes years to program a robot to do a single dexterous task, but Wang’s team sought to quickly teach a robot a skill by having it use motion-reading sensors to learn from a human practicing the same skill. Wang spent two months learning how to flip nunchucks, then let the robot watch, and the robot was performing the same skill within hours.
[protected-iframe id=”dec5654559a42872670d3a3e448c3d64-60970621-4867901″ info=”//content.jwplatform.com/players/2Vf1SAdK-NjrKGkKb.html” width=”650″ height=”365″ frameborder=”0″]
A robot like this could someday replace jobs like picking apples or assembling the interior of cars (the exterior assembly is already done by robots). “The tasks require a lot of hand manipulation, a lot of fine motor skills, and a lot of handling of composite objects that are partly soft and partly rigid,” Wang told Inverse. “So our vision is with our technology in the future, those tasks can also be robotized.” Regarding the task of picking apples — which would normally require a team of researchers programming every movement — Wang said, “If we can figure out how to teach a robot how to do this level of skills intuitively, just like teaching another human being, it just takes another apple picker to teach the robot.”
One thing not accounted for by this research is the likelihood us humans would go full Luddite and smash the robots if our boss told us, “Let this robot watch you work so I can fire you tomorrow.”
The research paper, titled Robot Composite Learning and the Nunchaku Flipping Challenge, is available on arXiv.