Mykhailo Fedorov, Ukraine’s former Defence Minister, raised eyebrows recently when he launched an ambitious scheme for an “Army of Robots” and said he wanted humanoid robots fighting on the battlefield within six months.

It is an extremely ambitious target. But this type of machine is no longer science fiction, and the U.S. Army is already developing requirements and testing Phantom Mk1 humanoids developed by Foundation Future Industries . These have also been demonstrated in Ukraine. While they will start out with reconnaissance and other missions, there is no doubt about the robots’ ultimate mission.

“We have done some weapons handling and eventually will do some weapon use cases,” Sankaet Pathak, co-founder and CEO of Foundation, told me. “We are not shying away from combat.”

Not The Terminator You Were Expecting

Pathak hears a lot about Terminators. Arnold Schwarzenegger’s iconic killing machine has left a lasting false impression of what humanoid military robots will be like. This is something Pathak has to struggle with constantly, though he does it with good humor.

“Terminators have time travel and are self-healing. I don’t think we’re there yet,” says Pathak.

Unlike the science-fiction version, Foundation’s robots are very much under direct human control. And their main purpose is not killing but keeping people out of danger, the company slogan being “ Don’t send a Marine where you can send a robot first .” (One of the founders is an ex-Marine).

Pathak does not believe that robots, like drones, should be more alarming than other military hardware.

“It’s an idiotic idea that it’s going to cause more damage than a bomb,” says Pathak, calling objections to robots that happen to be humanoid “a very emotional subconscious thing.”

Other military robots, known as UGVs or Uncrewed Ground Vehicles, are a familiar sight on the battlefield in Ukraine. These wheeled and tracked robots transport supplies and evacuate the wounded. They may carry explosive charges for kamikaze attacks, or be fitted with a machine-gun turret controlled by a remote operator. So why do we need humanoids?

“It is a more universal design than wheels or tracks,” says Pathak.

A bipedal humanoid can go virtually anywhere. While UGVs stay on paths or risk getting stuck or overturned, bipeds can cross rough ground and traverse mud and snow. And although Phantoms sometimes fall over, unlike UGVs they can get up again and keep going. They cannot go everywhere yet, but are progressively becoming more mobile.

“We’ll soon be demonstrating robots on some serious inclines, like mountain climbing,” says Pathak.

The U.S. military is interested in the humanoids initially for reconnaissance and materials handling. Reconnaissance includes navigating through an urban environment, negotiating staircases and doors, tasks virtually impossible for existing UGVs. Pathak says his machines have mastered many of the required skills, and should be fully capable in 18 months.

Elsewhere we see how fast humanoid mobility is developing. Recent milestones include running faster than any human over 100 meters, and coming first in a half-marathon in China . The half-marathon win is a sign of how battery life has improved since the early 2010s, when humanoids typically operated for an hour or less.

“We can go for four to six hours, depending on how strenuous the job is,” says Pathak. “That’s usually sufficient for any operation.”

Phantoms take 30-45 minutes to recharge, so only a few members of a robot squad need to be recharging at any given time.

The materials handling task includes jobs like loading and unloading trucks, arduous and time-consuming for soldiers. Outsourcing to machines frees people for jobs that require more intelligence. Materials handling could also include hazardous activities like placing explosive charges or disarming them. Both require a high degree of dexterity, and Pathak says Phantom will be equipped with a hand capable of matching human dexterity in early 2027.

Arguably the biggest challenge is autonomy. Autonomous robots will only need to be supervised, not directly control like current UGVs. Pathak says that better sensors, improved visual perception and the power of AI transformer models, the architecture that has boosted LLMs, mean that the robots are now able to learn fast and cope with unfamiliar situations.

“Zero shot learning,” where a machine can deal with a situation it has never encountered before, is a Holy Grail of robotics. Pathak says that, because military tasks tend to be structured, complete unknowns are comparatively rare and “few shot learning” is sufficient. This enables rapid learning so gaps can be filled in quickly. Before long the robots should, for example, be able to handle almost any type of object and traverse (or know to avoid) any type of obstacle.

Taking Bullets For The Team

Pathak expects his company’s machines to end up in combat. This prospect makes some developers uncomfortable; Boston Dynamics, for example, strictly prohibits arming any of its robots. But putting robots in the firing line rather than humans prevents casualties. In Ukraine, one unit plans to replace a third of front-line personnel with UGVs by the end of 2026. Humanoids can stand in for soldiers in far more situations, especially in urban combat, which is notoriously bloody.

Humanoids may be used as bullet sponges to draw enemy fire. Pathak says Phantom can withstand bullet hits to several parts of its body and still keep going. Unlike a human, it will not bleed out and does not have a torso full of vital organs. And, even if put out of action, a robot can be repaired.

Phantom is also resistant to blast, and can withstand a momentary shock of 100g, equivalent to an artillery shell landing close by. This would likely cause traumatic brain injury and other harm to a human.

Nobody is likely to send an ultra-expensive piece of hardware to check whether a path is mined or whether there is an ambush ahead. That is a job for an expendable unit. Rather than just a few robots for special missions, Pathak wants to see whole squads of low-cost machines. The current cost is around $50k and Pathak sees that going down as economies of scale kick in.

“We can see a path to $25-$30k. As the number of units go up, the cost approaches the cost of the materials,” says Pathak.

Ultimately, Foundation may be able to produce large numbers of basic humanoid machines for less than $10k each, with more capable units at higher cost. Taking a dozen human casualties in an assault would be a terrible loss. The same number of humanoids could be expended like rounds of ammunition. In fact they would be cheaper than many types of ammunition; a Javelin anti-tank missile costs $200k, a single MLRS rocket $140k.

The U.S. currently fields a handful of UGVs, mainly special-purpose units for bomb disposal. Pathak wants to see the number of machines expand to tens of thousands, and perhaps hundreds of thousands in ten years. In time robots could outnumber humans in the military.

Early humanoid soldiers may be basic recruits only able to follow simple instructions. But they will make great bullet sponges, going ahead to set off booby-traps, draw fire and locate enemy positions. They can stand guard, and make first contact with the enemy.

Future versions will be far more capable. Chinese developers talk about a “ChatGPT moment” for humanoid robots producing a dramatic leap in capability sometime in the next two years, and the Chinese military is certainly taking an interest. But deploying even simple versions could help save lives.

Fedorov’s six-month timeline for humanoids in combat roles is deliberately ambitious and provocative. But humanoid robots may already be able to take over some combat roles. And if technology exists to put a robot in the front line, it makes far more sense than risking a human life.