The same week the robotics industry was celebrating factory throughput, a quieter and stranger milestone surfaced in military circles: humanoid robots are now being tested for ground combat, and the role envisioned for them is not the killer android of cinema. It is something closer to a body double. The U.S. Army is evaluating Foundation's Phantom MK1, a nearly six-foot, 176-pound machine with a glass visor and five-fingered hands, for "breaching" operations, the work of entering mined buildings, climbing booby-trapped stairs and walking into kill zones. The point of sending the robot is that it can absorb what would otherwise hit a soldier.
The experts quoted are blunt about why this is the near-term use case rather than autonomous robot soldiers. "The near-term value of humanoids is less about replacing infantry and more about substituting humans in activities where casualties are otherwise highly likely," says Oleksandra Molloy of the University of New South Wales. Foundation, founded by former Marine Mike LeBlanc, sells the idea in a single slogan: "Don't send a Marine where you can send a robot first." Ukraine has already acquired two Phantoms, the natural next step after its 13th Brigade carried out the first robot-only assault back in 2024 using wheeled and tracked machines.
What makes humanoids interesting on a battlefield is the very thing that makes them awkward everywhere else: legs. Wheeled and tracked ground robots, the workhorses of Ukraine's logistics and casualty-evacuation missions, get stuck on rubble and stairs. A machine shaped like a person can climb, open doors and pick its way through forested or urban terrain, then hide in a trench or under debris until an enemy reveals itself. Nobody expects backflips. As military analyst Robert Bunker puts it, today's humanoids are slow and clumsy in new environments, but they can go where the wheels cannot, and the capability is improving by the month.
The grimmer logic is arithmetic. A human unit typically breaks after taking around a quarter of its members as casualties; a robot position has to be destroyed to the last machine before it can be taken. "Robot casualties don't count," in Bunker's phrase, which turns expendability itself into a force multiplier. He even reaches for an unsettling comparison, likening explosive-laden humanoids charging a trench to fighters who detonate themselves in enemy lines. Shooting at the robots, meanwhile, forces a defender to give away their own position. The operator watches all this through a narrow video feed, coolly, from somewhere safer.
The constraints are still real. Battery life runs to a few hours, autonomy remains unreliable, and the early machines will be remote-controlled rather than thinking for themselves. The deeper driver is cost, and here the military story folds back into the industrial one. A Phantom runs an estimated $150,000, roughly the price of recruiting and training a soldier, and it is expensive mainly because so few are built. Elon Musk wants to make 50,000 Optimus units in 2026 and eventually sell them below $20,000; Unitree has similar ambitions. The pattern is clear and a little chilling: the assembly lines being built for warehouses and showrooms are the same ones that will, almost as an afterthought, make expendable soldiers cheap.