Google is reportedly “retooling” its troubled robotics efforts, according to a report from The New York Times. The new operation is simply named Robotics at Google and will be led by AI scientist Vincent Vanhoucke. According to the Times, Google is focusing on using machine learning to teach robots how to grasp objects and navigate environments, but it’s far from clear where the company’s ambitions in this area lie.
Although Google is a pioneer in AI research, its efforts in robotics have produced no commercial successes to date. The company’s last significant foray into the field started in 2013 with a program named “Replicant” led by Android co-founder Andy Rubin. An initial flurry of activity led to the purchase of six up-and-coming robotics companies, including Boston Dynamics and DARPA challenge winner Schaft.
But these efforts stuttered, likely because the ambitious machines Google purchased were far away from commercialization. These firms were sold off or shut down in the years following, and Rubin left Google in 2014 after allegations of sexual harassment. (He was reportedly paid a $90 million exit package after Google investigated claims made against him, including that he coerced a fellow employee into performing oral sex on him in 2013.)
Since then, Google’s researchers have continued to produce new papers on robotics control problems, but none that have been presented with the same ambitious framing.
This week’s Times story doesn’t exactly change that. There’s no mention of significant new products, and Google has offered no long-term roadmap for its robotics research.
Instead, the company simply seems to be pouring more resources into applying machine learning — a technology it knows better than any other firm in the world — to the sort of robotics problems that would be useful in factories and warehouses. These include manipulating objects and navigating crowded spaces. This week, for example, the company published new research showing how AI can teach robot arms to toss objects.
Google is not alone in exploring this avenue of research. A number of other startups are pursuing similar work, including the research lab OpenAI, warehouse robotics designer Kindred, and startup Embodied Intelligence. Conquering these sorts of control problems could lead to a huge influx of cheap robots in various areas of industry, and would have a significant effect on the labor market as a result. But where Google would want to take such research is, for the moment, anyone’s guess.