Until now, many humanoid robots were more of a technological showcase: they appeared in videos performing tricks, running, or even dancing, but their contribution in a real-world work environment was limited. That's precisely what they want to change. Google by bringing the Gemini artificial intelligence model directly to factories, in a bid to turn these spectacular prototypes into useful industrial tools.
During CES in Las Vegas, Google DeepMind and Boston Dynamics announced a partnership to integrate Gemini Robotics into humanoid and quadruped robots., with the aim that they can understand their environment, recognize parts and manipulate them with their hands on real production lines, starting with the automotive industry.
From spectacular demonstrations to real industrial work
Atlas, Boston Dynamics' most famous humanoid, has already demonstrated that it can run, jump and perform acrobatics with amazing agilityBut that's not enough to turn it into a factory worker. Until now, these robots operated following very specific sequences, without really understanding what was in front of them or how they should react when something changed in their environment.
The integration of Gemini aims to address precisely that gap. It is an AI model. Designed from the outset as multimodal, capable of processing visual, spatial, and contextual information from the physical worldnot just isolated text or images. With Gemini Robotics as the "brain," the goal is for the robot to be able to interpret data from its sensors, identify objects, and decide how to manipulate them.
The first tests will not be done in closed laboratories, but in Hyundai factories, the group that controls Boston DynamicsAutomotive plants are considered an ideal environment for this leap: they are relatively controlled spaces, with repetitive processes but with enough variation to test the adaptability of robots.
The goal is not for Atlas or Spot to repeat the same programmed move over and over again, but rather be able to see a new piece, analyze its shape, decide how to hold it and what to do with it without needing to reprogram the system every time a change is introduced in the production line.
In this initial phase, Atlas will focus on performing various manufacturing tasks within Hyundai's plantsMeanwhile, Spot, the quadruped robot, can perform inspection and monitoring tasks in difficult or repetitive areas, collecting valuable data to continue training the Gemini model in real-world scenarios.
An AI platform for a whole generation of robots
Google DeepMind does not intend to launch "Google's robot", but something more ambitious: make Gemini the standard intelligence platform for robots from different manufacturersThis is similar to what Android achieved in the mobile phone market. The idea is that any robotics company can integrate this model into its hardware.
Demis Hassabis, CEO of DeepMind, has made it clear that The vision involves feeding general-purpose robots with a single AI brain.This allows very different machines—humanoids, robotic arms, quadrupeds, or mobile systems—to benefit from the same advance in physical intelligence. The more robots that use Gemini, the more data the model will collect and the better it will understand the physical world.
For Boston Dynamics, this alliance fits perfectly with its current strategy. The company, a pioneer in advanced robotics and known for its experiments with quadruped robots and humanoids, It has been evolving for years from military and research projects to more commercial applications, such as industrial inspection or logistics in complex environments.
Robert Playter, CEO of Boston Dynamics, emphasizes that the real leap forward is no longer in robots maintaining their balance or moving spectacularly, but in them Be aware of the context and use your hands to manipulate virtually any object in a factoryAccording to the executive, Hyundai's production environments offer the ideal testing ground to verify whether this new generation of systems is ready for daily work.
The agreement includes a joint research effort that will be developed both in the Boston Dynamics teams and in the Google DeepMind teamsEngineers, AI experts, and robotic behavior specialists will collaborate to adjust visual, linguistic, and action models to the demands of the manufacturing industry.
Gemini Robotics: Multimodal intelligence for the physical world
Gemini Robotics is the adaptation of the Gemini model to the field of robotics. It leverages the system's multimodal capabilities—text, image, video, and other data types—to Robots not only react to orders, but also reason about what is happening around them.This includes interpreting data from cameras, depth sensors, joint forces, or changes in the environment, relying on hardware such as the compact brain for physics AI.
Carolina Parada, senior director of Robotics at Google DeepMind, explains that Gemini's architecture was conceived from the beginning to learn from the physical worldThis makes it particularly suitable for controlling general-purpose robots. The automotive industry, in their words, is a good starting point, but the intention is to extend this technology to other sectors over time.
In contrast to classic industrial systems, where robots are designed to a very specific task in an almost unchanging environmentGemini aims to offer flexibility. If the position of a part changes, a new component is introduced, or there are slight variations in the line, the goal is for the robot to be able to adapt without a team of engineers having to redesign the base programming.
This approach also opens the door to More versatile robots in European and Spanish plantsIn this context, adapting to short production runs, model changes, and product customization is becoming increasingly common. Instead of completely reconfiguring a production cell, it might be enough to update the model or adjust the robot's behavior through further learning.
According to Parada, Google DeepMind's robotics focuses on creating Advanced AI that can power general-purpose robotsnot mere automatons. The medium-term ambition is that this type of system can be transferred to other areas, from logistics to infrastructure maintenance, including key sectors in Europe such as automotive, electronics, and the pharmaceutical industry.
A global race for the functional humanoid
The move by Google and Boston Dynamics is part of a Increasingly intense international competition to develop truly useful humanoid robotsIn the United States, companies like Tesla, Figure AI, Agility Robotics, Apptronik, 1X and other startups are working on prototypes capable of operating in factories, warehouses and, in the future, even in homes.
In China, the situation is even more active: It is estimated that around 200 companies are working on humanoid systemsWith the support of industry associations and public programs seeking to accelerate the adoption of this technology in logistics, manufacturing, and services, this competitive pressure is pushing major global players to move quickly.
The difference in the case of Google DeepMind lies in the approach. While certain manufacturers focus their efforts on building the “complete robot”—hardware, software, and services—, Gemini's goal is to become the layer of intelligence that others can incorporate into their own machines.If this strategy takes hold, it could create an ecosystem where multiple brands share a single AI brain.
For Europe, where the automotive industry and advanced manufacturing have considerable weight, this race for the functional humanoid robot is crucial. It can influence the competitiveness of plants located in Spain, Germany, France or Eastern EuropeThe adoption of smarter and more flexible robots could be a key factor in maintaining production on the continent compared to other regions.
However, the mass deployment of humanoids in European factories will depend not only on technology, but also on regulatory, labor and safety issuesIt will be necessary to fit these new machines into regulatory frameworks such as those promoted by the European Commission regarding AI and industrial safety.
From theory to the workshop: tests at Hyundai factories
Gemini's initial deployment in the physical world will take place in various Hyundai automotive plantswhere Boston Dynamics robots will have to tackle real-world manufacturing tasks. These aren't one-off demonstrations, but rather tests to see if these machines can add value in demanding production environments.
In this context, Atlas will attempt to execute operations that go beyond the simple repetition of movements. The challenge lies in flexible handling: picking up pieces with varied shapes, working with different tools, and adapting to changes in the workflow.Each variation will be a test for the Gemini model and its ability to generalize what it has learned.
Spot, for its part, can take care of tour the plant to inspect hard-to-reach areas, monitor equipment, or detect anomaliescollecting visual and other sensor data. All this information will be used to further improve Gemini Robotics' perception and decision-making algorithms.
This type of testing will allow for evaluation whether AI is capable of maintaining reliable behavior during long working hoursThis is essential for considering a wider implementation in European factories, including those in the automotive sector in Spain, where the pressure to reduce costs and increase production flexibility is growing.
The collaboration also involves a constant flow of data from the machines to Google DeepMind's teams, so that Each interaction of the robot with its environment serves to refine the modelThis continuous improvement approach is one of the key points for AI to move from being a laboratory experiment to a robust industrial tool.
Safety and risks: when AI moves things, not just pixels
Allowing AI to make decisions about physical systems is a qualitative leap compared to purely digital applications. A mistake in AI-generated text can be annoying; a malfunction in a robot sharing space with operators can be dangerous.That's why Google and Boston Dynamics insist that safety is a central pillar of the project.
Boston Dynamics systems already incorporate security and control mechanisms to prevent unstable behaviorOn that basis, Gemini will add additional layers of reasoning, designed to anticipate potentially risky behaviors and limit actions that could compromise the integrity of people or facilities.
Carolina Parada emphasizes that, in addition to physical and programming limitations, Gemini will apply a type of artificial reasoning aimed at detecting and curbing dangerous behaviors. before they occur. The idea is that the AI continuously evaluates the context and stays within safe operating limits.
Robert Playter, for his part, points out that even moderately sized robots They can become dangerous if not managed properly.Therefore, at least in the initial phases, these systems will operate under close human supervision, with very defined usage scenarios and clear shutdown and emergency protocols.
For the European context, where legislation on workplace safety and liability is particularly strict, This approach of prudence and control will be key if you want to introduce humanoid and quadruped robots into assembly lines shared with human workers.
One more step towards physical intelligence in industry
After decades in which industrial robots focused on repetitive movements in highly controlled environmentsThe arrival of models like Gemini opens the door to a new era. Artificial intelligence is no longer limited to answering questions or generating digital content; it is now taking over space in factories, workshops, and warehouses.
The alliance between Google DeepMind and Boston Dynamics aims to transform Atlas, Spot and other robots into systems capable of performing a variety of tasksto adapt to changing situations and collaborate with people in the day-to-day operations of production. The automotive industry, including that developed in Europe and Spain, will be the initial testing ground, but it will hardly be the only one.
If the technology lives up to its promises, Factories could move from rigid, highly specialized robots to more versatile machines.that can be reconfigured using software and data instead of requiring costly physical modifications. This would have a direct impact on costs, adaptation times, and the ability to produce short or customized runs.
At the same time, the project forces us to confront debates on employment, regulation and responsibility These factors will be especially relevant in the European context. The way these systems are integrated and the role workers play in this transition will largely shape the public perception of this new wave of automation.
With Gemini entering factories through Boston Dynamics' robots, industrial robotics is taking a turn towards a physical intelligence closer to human intelligence: The coming years will tell to what extent these prototypes of today will eventually become commonplace on assembly lines around the world, including European plants seeking to remain competitive in an increasingly automated market.