There are many skills in which robots cannot match humans, and one of them is cooking. Now, researchers at the University of Cambridge have trained a robot “chef” to make dishes after watching videos and learning how to cook.


Training robots in certain tasks has been a challenge for a long time. For example, if a user asks how to cook a certain dish, a robot may be able to give clear instructions such as: get the vegetables out of the fridge, chop them, put them into a pan, add these other ingredients, etc. But when the robot is asked to do these tasks, it usually fumbles.


In the new experiment, researchers programmed the robot with eight simple recipes, all of them salads. Then it was made to watch a video in which a human demonstrates one of the recipes. Having been programmed, the robot was able to identify which of the eight recipes was being demonstrated, and eventually made the recipe itself. In fact, after successfully completing the eight dishes, the robot came up with a ninth recipe on its own, Cambridge University said in a statement.


It said the results of the experiment shows that video content could lead to easier and cheaper deployment of robot chefs. The results have been described in the journal IEEE Access.


ALSO READ | Science For Everyone: How The Periodic Table Arranges The Elements, And Its Significance


Over the years, many science-fiction films have shown robots preparing food. In real life, commercial companies have built prototype robot chefs. However, the Cambridge statement said, “none of these are currently commercially available, and they lag well behind their human counterparts in terms of skill”.


 “We wanted to see whether we could train a robot chef to learn in the same incremental way that humans can – by identifying the ingredients and how they go together in the dish,” the statement quoted the paper’s first author, Grzegorz Sochacki, as saying.


The robot was trained with an existing neural network, which had already been programmed to identify several fruits and vegetables (broccoli, carrot, apple, banana and orange). Using computer vision techniques, the robot watched the different objects shown in the video, as well as the movements of the human demonstrator. It watched 16 videos and recognised the correct recipe 93% of the time, and the human chef’s actions 83% of the time, the statement said.


 “It’s amazing how much nuance the robot was able to detect,” Sochacki was quoted as saying.