This week, the computer chip company Nvidia opened a new robotics lab in Seattle. The lab has a lot of what you might expect–loads of electronics and robotic parts, work spaces, and computers everywhere–but the centerpiece is a bit unorthodox: It’s an Ikea kitchen.
That’s because University of Washington robotics professor Dieter Fox, who is on leave from the university to head up this Nvidia lab, believes that the kitchen is the perfect test-bed for the robots of the future.
Right now, most commercialized robots are either navigation robots that can deliver items from point A to point B in warehouses, hospitals, hotels, and maybe even urban areas, or they’re “manipulation” robots, which are almost entirely found in factories. These robots move objects around, and they’re optimized to do so over and over again in a very precise way. That means they’re easily programmable and don’t require any intelligence to do their work. According to Fox, the next big challenge for robotics is to develop manipulation robots that can work in an open-ended, changing environment–in other words, anywhere outside of a robot-only factory. These robots will need to be smart enough to perceive how their surroundings change over time. Importantly, they’ll need to interact with humans.
It’s this type of robot that Nvidia’s robotics lab will focus on. Fox plans to do the basic research necessary to make these robots possible, so consumer products are off the table for now (though Fox also hopes to discover some cool AI tricks that might wind up in one of Nvidia’s offerings). But longer-term, such robots could be used in manufacturing to work alongside human workers or in healthcare facilities to handle the menial tasks of elderly care, for instance. That’s likely why Nvidia is investing in making them a reality without expecting any products to come out of the lab in the next three to five years. After the company’s stock flagged at the end of 2018, opening a robotics lab could be a move toward a new area of business for the company–supplying the chips that can power future domestic and commercial robots, or maybe even building the robots themselves. It’s something Nvidia has already started branching into with its Jetson AGX deep learning chip.
The kitchen, The ultimate test-bed
Fox is convinced that the best place to test out these robots is the kitchen. Robotic chefs have been a fixation among technologists for decades–in the last few years alone, we’ve seen robots designed to flip burgers, toss salads, and brew coffee. But the robots of the future will do much more than work in a kitchen.
“We didn’t choose the kitchen because it’s going to be the killer app in the home or it will be a product in the new few years, but because it can represent any of these other application domains,” Fox says. That means that all the actions that a robot might do in a kitchen, like opening drawers and cabinets, retrieving specific items like the salt shaker, performing fine motor skills like chopping onions, or taking instructions from humans and working alongside them to prepare a meal, are the same types of actions that will be useful in other situations in factories, healthcare centers, and elsewhere. Fundamentally, the underlying algorithms will work the same.
Plus, because of the broad nature of tasks that a robot could perform in the kitchen, it’s easy to come up with increasingly difficult tasks as the algorithms get better and better at the basics. Right now, the Nvidia research team is researching different methods for opening doors and drawers, recognizing objects in the kitchen, and moving those objects around. In the future, they might test out how a robot can retrieve the mustard from the fridge when a human asks it to do so. Today, the robot uses a 3D model of the kitchen that the team built when they brought it back from Ikea; another challenge might be to ask it to navigate and retrieve objects without any kind of model, forcing the bot to rely on its own powers of perception.
An additional benefit to testing in a kitchen: It allows the researchers to work in a real-world environment. Fox says that many robotics labs tend to dream up arbitrary tasks for their robots to complete, but Nvidia’s robots will be focused on tasks derived directly from an everyday space that resembles something robots may some day occupy. “By forcing ourselves to look at this kitchen challenge, we want to see what it takes to solve these tasks in the real world,” Fox says.
Fox plans to publish the Nvidia team’s kitchen-based software system so that other researchers around the world can download it and begin building more specific applications on top of it. He dreams of the kitchen as a global standard for robot testing, where lots of labs that are working on robot manipulation, human-robot interaction, and robot design can all do testing in their local kitchen, which would make it easier for different research groups to see how one method for chopping onions compares to another. The kitchen doesn’t have to be the same off-the-shelf Ikea kitchen that Nvidia’s built, nor does it have to be an Ikea kitchen at all. As long as each kitchen has its own 3D model, researchers should be able to test out various methods developed in other kitchens in the comfort of their own space.
“The research community is lacking testing scenarios that labs can use to compare different research approaches,” Fox says. “If you want to write a research paper and say ‘My approach is better than yours,’ you should be able to run your approach in my kitchen.”
Why this obsession with kitchens in the first place? Fox says he does like to cook, so that might have something to do with it. But ultimately, Fox’s own personal kitchen might prove to be something of a challenge for robots. He has modern-style drawers without any handles at all, just a lip at the top. They’re easy enough for a human hand to use, but for now, they’re robot-challenged.