A team of researchers at the University of Tokyo managed to teach a dual-armed robot how to peel a banana without turning it to mush in the process.
Teaching a robot to handle delicate objects and perform intricate tasks on them is extremely difficult, but as Reuters reports, the key to getting a robot to peel a banana looks to be imitation.
Researchers Heecheol Kim, Yoshiyuki Ohmura, and Yasuo Kuniyoshi from the Intelligent Systems and Informatics Laboratory (ISI Laboratory) at the university used a "goal-conditioned dual-action deep imitation learning (DIL)" method with the robot. It uses human demonstration data to learn the actions for peeling, which required the research team to repeat the process themselves hundreds of times to create the learning data set.
After 13 hours of training, the robot can only successfully peel a banana 57% of the time, but that's pretty good considering no two bananas are the same size or shape. The success rate should improve in future due to the dual-action process preventing compounding errors typically associated with previous DIL methods. Hopefully practice makes perfect.
By continuing to improve the success rate, the research team believes robots can eventually help with labour shortages in Japan for sectors that require handling delicate objects repeatedly. For example, the work carried out in food processing factories.
Sign up for What's New Now to get our top stories delivered to your inbox every morning.
This newsletter may contain advertising, deals, or affiliate links. Subscribing to a newsletter indicates your consent to our Terms of Use and Privacy Policy. You may unsubscribe from the newsletters at any time.
Your subscription has been confirmed. Keep an eye on your inbox!
Read more on pcmag.com