“For the first time ever in public, please welcome Atlas to the stage,” Boston Dynamics general manager Zachary Jackowski told a CES audience as a life-sized biped lay on the floor, then pushed itself upright.
The occasion was not so much spectacle as index. Boston Dynamics has taken years to allow its robots to shine in well-crafted videos; it is not as indulgent on a stage. Then Atlas walked across the stage in a steadiness which betrayed the attempt to get the talk out of viral athleticism and into industrial routine–work, which occurred again and again, which grew larger and larger, and which had to be checked.
Hyundai, which has a majority stake in the Massachusetts-based robotics company, put that routine in terms that were unusual. The firms indicated that a product variant of Atlas that would aid the assembly of the vehicles is already being made and will be deployed in 2028 at an electric-vehicle production plant of Hyundai near Savannah, Georgia. The tone was set with the same presentation through the use of choreography: four quadruped Spot robots danced to K-pop in unison, then the humanoid entered, the audience being reminded that the publicity of Boston Dynamics has always focused on performance. However, the real tale is now behind the curtain–supply chains, maintenance cycles, safety cases, and the ugly math of uptime.
That change is one of the reasons why humanoids are uncommon on live stages. This is because falls and stumbles become the headline and the engineering story has to control, sense, actuate and recoveries flattened into a meme. Boston Dynamics escaped this pitfall by making the walk a demonstration, and an engineer operated Atlas at a distance remotely. The company set autonomy as the ultimate position of the actual deployments, whereas teleoperation is a reasonable intermediate of a publicly debut.
The center of gravity was however elsewhere: the software stack under construction around the machine. Hyundai declared a collaboration with Google DeepMind that will offer a type of AI technology to the robots of the Boston Dynamics, which is intended to provide an already tested body with a quick adapting AI “brain.” Carolina Parada, DeepMind, explained the ambition as follows: “Rather than having a set of predefined, loaded tasks onto the robot, we think robots should understand the physical world the same way we do.” In such a framing, Atlas is not a single product, but rather a stage on which embodied models – perception, reasoning, and action as bound together by industrial constraints – are displayed.
The specifications of the production version of the same reveal those limitations. The materials provided by Hyundai talk of the Atlas with 56 degrees of freedom and human-scale hands that have tactile sense and it can lift 110 pounds Other facts, omitted in the consumer-facing hype of robotics, indicate factory reality: -4°F to 104°F operation, an IP67 rating against water and dust, and a design that is washable down. The four hour battery life of the machine is low by shift standards, though Boston Dynamics has indicated a workaround involving hot-swapping its batteries in three minutes or so which is instantly obvious to the manufacturing engineers as a throughput gimmick.
The durability, sensing and serviceability of that combination is an indicator of the distinction between an artifact and a humanoid, a line-side tool. It also alludes to the reason why Hyundai is constructing a special robot-training plant in the United States this year, a so-called Robot Metaplant Application Center (RMAC), which will aim to train the robots in movement patterns and gather training information. Simultaneously, Hyundai has outlined a bigger circle: beginning with activities such as sequencing of parts and eventually broadening the circle as the processing is confirmed to be safe and having good quality.
The subtext cannot be separated with the competition. Humanoids have become a marquee bet by Tesla, and other platforms have been publicly tested by automakers such as Mercedes-Benz and BMW. However, what stands out about the Hyundai strategy is not so much the stage performance as a concerted pledge: a giant manufacturer allied with a robot manufacturer, an artificial intelligence laboratory, and an entire factory presence, but also proclaiming the ability to make tens of thousands of vehicles a year. Hyundai has indicated that it hopes to have a plant able to manufacture 30,000 Atlas robots per year starting in 2028, which, had it happened so, would compel the industry to regard humanoids as items of purchase and not as prototypes.
The most difficult aspects are the more prosaic ones, even with scale plans; safety in shared spaces, predictable behaviour, the actual cost of deployment. Boston Dynamics has prioritized designs that are supposed to be operated with people, such as onboard designs which are used to identify nearby workers and stop, and physical design considerations to minimize pinch points and to provide padding. The applicability question remains long term in commercial consideration. According to McKinsey partner Alex Panas, “I think the question comes back to what are the use cases and where is the applicability of the technology.”
To date, the CES launch by Atlas put a limitation on the near term: industrial work, human-scale setting, and a roadmap where autonomy is delivered, rather than a stage show. The applause in the ball room was actual, but the more important audience is in other places–the process engineers, measuring cycle times, the safety managers, reading standards, the plant operators, deciding whether the humanoid should be on the floor or in the lab.
