Robots have long been helping us with various tasks, from washing the dishes to frying chips. But while they’ve come a long way since their debut, there is still work that needs to be done on getting them to handle soft items without smashing them – something which could soon change with advances like Google DeepMind technology. This week’s video on YouTube showcases an advanced machine-learning robot capable of peeling a banana–something even a monkey would likely find easy. Reuters reports that Heecheol Kim, Yoshiyuki Ohmura and Yasuo Kuniyoshi’s team at University of Tokyo has taught their dual-armed robot to remove banana skin in under three minutes using both “fingers.” Researchers used “deep imitation learning,” in which they demonstrated something repeatedly for a robot to learn and replicate it. Over 13 hours of training on peeling bananas successfully managed by this method, 57% were peeled successfully during its sessions.
Researchers designed their robot with two arms that each contain four fingers or “fingers,” each attached to a claw-like end that allows it to grasp objects similar to human hands. Furthermore, the robot can rotate bananas for peel removal as well as grasp its peel at various points on the fruit – essential because not all sections of bananas have equal sizes or ripenness levels.
Kuniyoshi hopes that using their machine-learning approach to teach robots basic human tasks could ease Japan’s labor shortage issues in industries like bento lunch box manufacturing plants or food processing factories where manual work is done extensively.
As seen in the video, there are some flaws with this system, yet scientists have already published their results on arXiv preprint server. For example, when trying to peel bananas for removal of peel section one it has an affinity for repeatedly pulling at one spot which could damage or bruise its flesh beneath – research team working towards solving this problem.
Even with its limitations, this robot is impressive, and could potentially be employed in restaurants or other settings where food preparation requires delicate handling. Aitken would like the bot to learn to handle more misshapen fruit such as apples and other soft produce and be better at identifying which parts of a banana it is peeling. Even though the current system only succeeds at peeling bananas 57% of the time, that represents an improvement from earlier robots which were only capable of doing this approximately 20% of the time. Universal Robots of Boston has been marketing a robotic system designed to peel bananas since 2012; sales should begin sometime around 2019. A Korean team also released their version in 2015.