Google DeepMind is teaming up with Boston Dynamics to give its humanoid robots the intelligence required to navigate unfamiliar environments and identify and manipulate objects—precisely the kinds of capabilities needed to perform manual labor.

The collaboration, announced at CES in Las Vegas, will see Google’s Gemini Robotics model deployed on various Boston Dynamics’ robots, including a humanoid called Atlas and a robot dog called Spot.

The companies plan to test Gemini-powered Atlas robots at auto factories belonging to Hyundai, Boston Dynamics’ parent company, in the coming months. The move is an early look at a future where humanoids are able to quickly master a wide range of tasks.

A humanoid robot called Atlas from Boston Dynamics.

Courtesy of Boston Dynamics

Atlas can already dance and perform acrobatics but, like other humanoids, lacks the intelligence needed to understand its environment, make complex decisions, and manipulate unfamiliar objects with its hands. That could start to change with the addition of an advanced AI model like Gemini, though it’s unclear how robots will match the adaptability and subtlety of human manual dexterity.

“The real value going forward is for our robots to be contextually aware of their environment and able to use their hands to manipulate any object,” Robert Playter, CEO of Boston Dynamics, told WIRED ahead of today’s announcement. “And we think manufacturing environments, like in Hyundai factories, are a perfect place to deploy that today.”

For several decades, Boston Dynamics has pioneered the development of advanced robots, including humanoids and four-legged systems tested by the military as a way to carry supplies over rough terrain. The company was sold to Google in 2013 and was bought by SoftBank in 2017. In 2021, Hyundai acquired a controlling stake.

In recent years, advancements in motors, batteries, sensors, and other components have made it easier for startups and other companies to compete in the robotics space. More than a dozen firms in the US, including Agility Robotics, Figure AI, Apptronik, 1X, and Tesla, are working on humanoid systems. Overseas, the competition is even fiercer, as roughly 200 Chinese firms are developing humanoid systems, according to CMRA, a Chinese industry association.

Many AI firms are turning their attention to robotics, believing that AI models need a deeper understanding of the physical world to achieve humanlike intelligence. OpenAI is said to be developing humanoids. Tesla has shown off a humanoid called Optimus. Several startups are also working on models designed to let robots quickly master various physical tasks. Some hope to deploy humanlike robots as home helpers, despite their current limitations.

Google DeepMind hired the former CTO of Boston Dynamics in November. Rather than building its own robots, Google DeepMind CEO Demis Hassabis said he envisions Gemini being used by many different robot makers, similar to how Android runs on a wide range of smartphones.

While nobody knows the best way to give machines the kind of physical intelligence that human beings take for granted, adding artificial intelligence to industrial systems could greatly expand their usefulness.

Likewise, the data collected by Boston Dynamics’ machines will help improve Gemini’s ability to operate in the physical world, according to Playter.

Carolina Parada, senior director of robotics at Google DeepMind, notes that Gemini was designed to be multimodal, making it well suited to understanding and learning about the physical world. “Google DeepMind’s robotics focus is building advanced AI that can power general-purpose robots,” Parada said in an interview with WIRED. “Automotive is a great place to start, but certainly our intention is to continue expanding to more and more applications.”

Having AI take control of physical systems could also introduce new risks. Parada says that on top of the safety controls already built into Boston Dynamics’ systems, Gemini will perform an artificial kind of reasoning in order to preempt and prevent potentially dangerous behavior.

Playter says that being able to guarantee human safety will be crucial if humanoid robots are going to take off. “Even the little ones can be dangerous,” he admits.