This Robot Learned to Make a Salad by Watching YouTube

3 minute read

Julia Child taught a generation of Americans how to cook gourmet French cuisine by breaking it down into simple steps that anyone could follow. A robot named for her at the University of Maryland took a similar approach when it taught itself to make a salad.

Using pattern recognition software designed by the interdisciplinary robotics team at the College Park campus, Julia the robot watched YouTube videos of people making salads to learn the steps, from cutting vegetables to tossing the ingredients and even pouring the salad dressing at the end.

Surprisingly, it was that last step that proved the most difficult, since dressing doesn’t always come out evenly and the robot does not yet have good feedback mechanisms to know when too little or too much is coming out. Even taking the cap off the bottle proved challenging.

Computer science professor Yiannis Aloimonos said the team chose cooking because it is something that everyone understands how to do but which is actually challenging for a robot to learn. But the lessons they’ve learned in programming Julia could be applied to just about any human activity, from stocking shelves to working on a factory floor.

“If you can work in the kitchen with your hands and do things, basically you can do almost anything else,” he said.

So far, Julia has made tomato salads, fruit salads and even a tuna salad that required some tricky work spooning out the mayonnaise. Aloimonos, who moved to the U.S. from Greece in 1982, said they’re now working on a Greek salad, but not just because he likes them.

“The objects that are involved—tomatoes and cucumbers—are not easy to cut and manipulate, and then you have soft things like feta cheese and oily things like olives,” he said. “It’s a challenge to put them all together at the same time.”

The robot was purchased with an educational discount from Rethink Robotics for about the price of a small car. Cornelia Fermüller, a research scientist at the university, developed the pattern recognition software to allow it to learn from watching YouTube—a process she compared to learning how to speak a language.

For now, it’s a language that Julie speaks at only a basic level. But there’s always more YouTube videos to watch.

“I don’t think that we’ll get to that gourmet level soon,” Fermüller said.

See the Most Iconic Artificial Intelligence in Film

Brigitte Helm In 'Metropolis'
False Maria – Metropolis, 1927Michael Ochs Archives/Getty Images
The movie "2001: A Space Odyssey", directed by Stanley Kubrick. Screenplay by Stanley Kubrick and Arthur C. Clarke. The camera eye of the HAL 9000 computer on Discovery One spaceship. Initial theatrical release April 6, 1968. Screen capture. © 1968 Metro-Goldwyn-Mayer Studios. Credit: © 1968 MGM / Flickr / Courtesy Pikturz.Image intended only for use to help promote the film, in an editorial, non-commercial context.
HAL 9000 – 2001: A Space Odyssey, 1968MGM
R2D2 – Star Wars, 1977Lucasfilm
Star Wars (1977)Directed by George LucasShown: Anthony Daniels (as C-3PO)
Lucasfilm
Tron – Tron, 1982Buena Vista
WarGames
WOPR/Joshua – WarGames, 1983Hulton Archive/Getty Images
Dot Matrix – Spaceballs, 1987MGM
MCDSHCI EC003
Johnny 5 – Short Circuit 2, 1988TriStar
The Terminator – Terminator 2: Judgement Day, 1991TriStar
T-1000 – Terminator 2: Judgement Day, 1991TriStar
Agent Smith in The Matrix, 1999Warner Bros.
Andrew Martin – Bicentennial Man, 1999Touchstone Pictures
MSDAAII EC022
David – A.I. Artificial Intelligence, 2001Warner Bros.
MSDAAII EC001
Gigolo Joe – A.I. Artificial Intelligence, 2001Warner Bros.
MCDIROB FE010
Sonny – I, Robot, 200420th Century Fox
MCDHIGU EC003
Marvin – The Hitchhiker’s Guide To The Galaxy, 2005Touchstone Pictures
Optimus Prime – Transformers, 2007DreamWorks
WALL•E
Wall-E – Wall-E, 2008Pixar/Disney
HER
Samantha – Her, 2013Warner Bros.

More Must-Reads from TIME

Contact us at letters@time.com