While the U.S. trains humanoid robots to do backflips and shine on demo stages, the real race for global manufacturing leadership will be won by whoever makes machines more adaptable — not more powerful. That’s the argument Robert Ambrose, who spent decades designing robots for space, makes in a commentary for Fortune.
Earlier this year, China lined up a whole troupe of humanoid robots to dance in front of the German Chancellor. Many saw an impressive show of technological muscle. Ambrose, by his own account, saw something else — bragging. And, more importantly, a symptom of a problem the United States itself is in danger of falling into: the gap between spectacle and strategy.
Robert Ambrose currently chairs robotics and artificial intelligence at the firm alliant, and previously ran NASA’s Software, Robotics and Simulation Division. In other words, this is someone who knows firsthand how robots behave where the cost of failure is a failed mission.
Impressive ≠ useful
Ambrose’s core point is simple: American robots look great, but they’re measuring the wrong thing. At any major U.S. demo you’ll see fluid movements, precise manipulation, maybe even a backflip. The most advanced Boston Dynamics machines lift and carry loads heavy enough to injure a human worker. By “showcase” metrics, the U.S. looks competitive.
The problem is that those metrics are captured in ideal, controlled conditions. Ambrose points to a recent Stanford study: robots that succeed at nearly 90% of tasks in simulations manage just 12% of real-world household tasks. That’s not a rounding error — it’s essentially the whole problem. The U.S. is optimizing its robots for a sprint and calling it a marathon strategy.
The story of one robot at BMW’s plant
A telling example is the 02 model from Figure AI. The robot logged 1,250 hours at BMW’s Spartanburg plant and moved more than 90,000 parts. By current standards, a success. But look closer and, for ten straight months, the machine performed exactly one operation: picking up sheet-metal parts and placing them on a welding fixture.
And that, Ambrose argues, raises an awkward question. A major corporation like BMW can afford a ten-month single-task “pilot” and write it off as research and development. A mid-sized manufacturer — the backbone of U.S. industry — won’t sink thousands of dollars into a machine that does just one thing. Successful one-off deployments mask the real question: does this pay off at scale?
What NASA taught the author: brittleness kills
At NASA, Ambrose says, decades of work revealed a pattern: the machines that failed were precisely the ones built for a single scenario. The ones that survived could switch between tasks and be reprogrammed.
His favorite example is the Space Shuttle’s robotic arm. It was built to position an astronaut who would catch and later release a satellite. As it turned out, the robot was better at making the catch itself — and its ability to position things precisely proved useful for other jobs, such as repairing the Hubble Space Telescope. Flexibility beat narrow specialization.
Humans’ main advantage, the author reminds us, isn’t strength or speed but adaptability. In half a day, a single warehouse worker can pick orders, restock shelves, spot a safety hazard, and step around a spill. That fluid task-switching is what makes human labor valuable. To replace people on factory floors, robots will have to become more flexible than we are — and for now, in most plants, a few humans deliver more return than a single humanoid.
The policy isn’t ready either
For anything to change, Ambrose says, the rules of the game have to be reset first. Mid-tier American manufacturers have almost no clear path to adopting these robots at scale. Current federal tax credits reward inventing robots, not deploying them: a company that spends $800,000 integrating a humanoid system gets roughly the same credit as one that buys a new forklift.
Venture investors have already poured some $2.5 billion into robotics, but private money alone isn’t enough. The author proposes several fixes: a dedicated tax incentive specifically for deploying robots in real production (one that offsets integration, retraining, and process-redesign costs); an expansion of the federal program that advises small and mid-sized businesses; and shared interoperability standards so robots from different makers can be safely combined — a task he suggests NIST could take on together with NASA.
What the “right” deployment looks like
Factories themselves will also have to change, Ambrose writes. Most production processes are built around human improvisation. Robots need something different: managing a “fleet” of machines the way ride-hailing services dispatch cars; clear safety rules for mixed zones where people work alongside robots; and protocols for robots to interact with single-purpose equipment.
Both the U.S. and China, in the author’s view, get the main thing wrong — the assumption that robots will replace people wholesale. Their real value lies elsewhere: filling the “in-between” work that’s too variable for a conveyor and too repetitive to justify a skilled employee. Moving parts between stations, restocking warehouses, tending machines, inspecting dangerous and confined spaces — unglamorous tasks, but exactly the ones adaptable robots can start handling this decade.
The stakes: leadership for decades
Ambrose’s conclusion is blunt: America has the talent, the capital, and the industrial base to lead this transition, yet it’s optimizing for the wrong outcomes and ignoring the policies that would make real deployment possible.
The country that first defines what “good enough to deploy at scale” means will set the rules of global manufacturing for decades. Right now, the author concludes, that country is not the United States. But it doesn’t have to stay that way.
Source: Robert Ambrose, “Former NASA Robotics Chief: America is building the wrong kind of robots — and China knows it,” Fortune, May 23, 2026.