Teach the robot to open a door and it should open a lifetime of opportunities. Not so for one of Alphabet’s youngest subsidiaries, Everyday Robots. A little more than a year after Alphabet’s X Moonshot lab ended, the team that trained more than a hundred wheeled, one-armed robots to clean dining room tables, separate trash and recycling, and yes, open doors, is shutting down as part of budget cuts via parent Google, a press release confirmed. secretary.

“Everyday Robots will no longer be a standalone project at Alphabet,” says Denise Gamboa, director of marketing and communications for Everyday Robots. “Some of the technology and some of the team will be integrated into existing robotics efforts at Google Research.”

The robotics venture is the latest failed bet for X, which also launched internet-enabled balloons (Loon) and kites (Makani) in the past decade before deeming them too commercially unviable to stay afloat. Other former X projects, such as Waymo (developing autonomous vehicles) and Wing (testing unmanned drones with grocery deliveries), operate as companies within Alphabet, although their financial prospects remain shaky with regulatory and technology challenges. Like Everyday Robots, these businesses used new technologies that showed impressive promise during testing, but not fragile reliability.

Everyday Robots emerged from the ruins of at least eight robotics acquisitions by Google a decade ago. Google co-founders Larry Page and Sergey Brin expected machine learning to revolutionize robotics, and Page in particular wanted to develop a consumer-oriented robot, says a former employee involved at the time, speaking on condition of anonymity to discuss internal discussions. By 2016, they commissioned software entrepreneur Hans-Peter Brøndma to lead the project, then called Help (and later, for a time, Moxie), to use machine learning to develop robots that could perform routine tasks and adapt to different environments, the source said. . .

The team created weapon farms and arenas where a group of robots repeated the same task for months, such as sorting garbage. It was a crude attempt to generate data to train a machine learning model that could then imbue robots with the know-how needed to use their cameras, hands, wheels, and finger pens to interact with the world around them. The innovation freed engineers from the traditional approach in robotics, which involves the need to code specific instructions that machines must follow for every small potential scenario. The idea mostly worked for the initial tasks. At Google, a fleet of Everyday Robots helped clean the search giant’s cafeterias and check that conference rooms weren’t cleaned in the middle of a pandemic.

Provided by Google

Last year, Everyday Robots demonstrated further progress with Google AI researchers. The project integrated a large language model similar to the one underlying ChatGPT into a robotics system, allowing a mechanical assistant to, for example, respond to someone saying they’re hungry by bringing them a bag of chips. But Google and Everyday Robots emphasized at the time that the roving butler at their request remained distant from consumers. Variations that seem trivial to people, such as the type of lighting in a room or the shape of a bag of chips, can cause malfunctions.

A former employee says that from its earliest days, Everyday Robots struggled with whether its mission was to conduct cutting-edge research or bring a product to market. It has more than 200 employees, including people who oversee customer work, teach robots to dance, and create perfect designs. Robotics experts estimate that each of his robots probably costs tens of thousands of dollars.

Those costs were too much for Alphabet, whose more speculative “other bets” like Everyday Robots and Waymo lost about $6.1 billion last year. Alphabet’s total revenue fell 21 percent to $60 billion last year as ad spending at Google slowed and activist investors demanded cuts from the company. On January 20, Alphabet announced that it would lay off about 12,000 workers, or 6 percent of its total workforce. Everyday Robots was one of the few disbanded projects.

Source by [author_name]

Previous articleWhy it’s so hard to add electrical transmission lines in the U.S.
Next articleNvidia (NVDA) Q4 2023 Earnings