Foundation Robotics' Phantom is Learning Basic Tasks
In an industrial space located in a technology-focused area of San Francisco, expectations of witnessing a menacing humanoid robot soldier performing combat-related activities were met with a different reality. Instead, the black, shiny, faceless Phantom robot was observed engaging in "free play," manipulating a collection of colored children's blocks.
"We need data from it just interacting with its environment…[and] this is today's menu,"explained Sankaet Pathak, co-founder and CEO of Foundation Robotics, a two-year-old startup developing Phantom for both military and civilian uses.
Later, Pathak demonstrated the robot's stability by pushing its 80kg steel-covered body around the room and showcased its walking capabilities.
While numerous companies are developing autonomous humanoid robots for industrial, domestic, or companion purposes, Foundation Robotics asserts it is the only U.S. company focusing on a wide range of defense applications. These applications include support roles such as supply pickup, reconnaissance, equipment or casualty recovery, and hazard inspection. More controversially, the company also aims for warfighting capabilities to engage and neutralize threats, a concept Pathak refers to as "frontline weaponisation."
"Arming robots could keep human soldiers out of harm's way,"Pathak argues. He suggests that robots could enter and search buildings where chokepoints pose lethal risks and potentially reduce collateral damage. According to him, land-based autonomous systems can be more precise than autonomous aerial strikes.
However, these capabilities remain futuristic for Foundation's Phantom. The first-generation model, Phantom MK-1, currently lacks a battery, is not dust or waterproof, and cannot recover if it falls. Its hands, a significant robotics challenge, lack sufficient strength and dexterity, and it does not yet have functional wrists.
A second-generation model is under construction in a restricted area of the facility. Pathak states that Phantom MK-2 will be weatherproof, equipped with a large battery providing approximately six hours of runtime, capable of self-recovery after falls, and able to withstand greater physical force.
Improved hands are critical for the robot's advancement. The upcoming set will have increased mobility and wrists designed to facilitate weapon handling, according to Pathak.
Foundation Robotics aims to produce at least 40,000 units annually by the end of 2027, targeting a long-term cost below $20,000 (£15,000) per unit.
Strategic Vision and Military Interest
Pathak emphasizes that China is actively pursuing similar technologies and that Western nations must keep pace. He envisions hundreds of thousands of AI-driven humanoid robots forming a ground force, complementing the increasing use of autonomous drones in aerial combat. Such a fleet could serve as a significant deterrent to conflict.
Foundation Robotics currently holds $24 million (£18 million) in research contracts to pilot its technology with the U.S. military and has two units undergoing testing with the Ukrainian military. The U.S. military pilot focuses on handling rather than firing weapons, while weaponisation is part of the testing in Ukraine.
The company gained additional attention earlier this year when Eric Trump, son of the U.S. President, became an investor and advisor.
For Pathak, Foundation Robotics also represents an opportunity to demonstrate his capabilities following the bankruptcy filing of Synapse, a financial services firm he co-founded and led in 2024.
Challenges and Ethical Considerations
The military's interest in humanoid robots is evident. Dean Fankhauser of Robozaps, a humanoid robotics advisory firm operating a commercial systems marketplace, notes a current U.S. Army contest for humanoid robots that could eventually support soldiers in diverse tasks.
"It is completely inevitable,"Fankhauser states, that companies will explore weaponising this technology.
Currently, simpler robots such as drones and some ground systems are employed to carry explosives, missiles, and other payloads, with battlefield use notably visible in Ukraine. Some companies have also worked on weaponising dog-like quadruped robots, although these have not been widely deployed in active warfare yet, according to Fankhauser.
Other legged robot manufacturers have opposed weaponisation, citing risks of harm and ethical concerns. Pathak disagrees, asserting it is dangerous that more firms are not following Foundation's example.
He argues that humanoid robot soldiers are logical because the world is designed for humans; existing tools such as screwdrivers and weapons do not need reinvention. Pathak emphasizes that humans should remain "in the loop," approving any lethal force use before the system acts, although he acknowledges exceptions where autonomous firing might be necessary to prevent catastrophic outcomes and envisions scenarios where human authorization is less critical.
Artificial Intelligence and Operational Capabilities
A significant challenge for all humanoid robot developers is creating artificial intelligence capable of operating in real-world, unpredictable, and complex environments.
Phantom is controlled by an AI system named Cortex, with a new version under development. The concept involves assigning Phantom a goal—such as transporting supplies or mapping a building—based on tasks it has been trained to perform through demonstrations using videos, images, and text.
The robot navigates its surroundings using cameras embedded in its helmet that provide 360-degree vision, enabling the AI to assess the environment and adapt its movements accordingly.
Pathak explains that Cortex integrates two types of AI models:
- A "reasoning model," trained on task-specific examples, which interprets the goal and formulates Phantom's action plan.
- A broader "world model," trained on internet videos and data collected from the robot's interactions with the physical world—including its "free play" with blocks—which predicts environmental responses to help Phantom move safely and execute actions.
Alternative Robot Forms and Technological Progress
Not all experts agree that humanoid robots are the most effective form factor for military applications. Quadruped robots, for example, can navigate terrain more quickly and efficiently, according to Fankhauser of Robozaps.
He also notes that commercial humanoid technology remains immature. Current commercial humanoid robots struggle with basic tasks such as warehouse packing and opening doors.
"If there was a war in Taiwan today, the likelihood that China is going to militarise these humanoids and fight effectively is fanciful,"Fankhauser adds.
While Chinese robots have demonstrated impressive displays, these have occurred in highly controlled or structured environments, which contrast with the unpredictability of real-world warfare. Fankhauser suggests the situation might evolve over the next five to ten years.
Academic Perspectives and Practical Limitations
Robert Griffin, who works on humanoid robots at the non-profit Florida Institute for Human and Machine Cognition, which conducts military-funded humanoid projects focused on non-combat applications, acknowledges the potential of humanoids to reduce risks to human soldiers. One of the institute's spin-out companies was later acquired by Foundation Robotics for part of its core technology.
Griffin highlights that unpredictable environments remain a significant challenge. Tasks such as jumping through a window of unknown height, landing on uneven ground, and navigating unfamiliar interiors are difficult for robots.
"You get an impression of human-level capability by seeing the human form… but [these autonomous systems] don't know how to handle open-ended uncertainty yet,"Griffin states.
He notes that human soldiers have effectively thwarted AI systems by performing unexpected actions, such as somersaulting or placing cardboard boxes over their heads.
Practical issues such as runtime also persist. Griffin describes power consumption for locomotion and joint movement as a problem "plaguing every humanoid company." Achieving six hours of operation would be "very impressive."
Whether Foundation Robotics can develop hands capable of manipulating weapons designed for humans remains uncertain. Griffin comments that the company is setting extremely challenging engineering goals.
Ethical Concerns
Ethical considerations are prominent in discussions about lethal autonomous weapons. Nicole van Rooijen, executive director of Stop Killer Robots, a global coalition of non-governmental organizations, emphasizes that such weapons lower the threshold for warfare, dehumanize conflict, and obscure accountability.
She finds the humanoid form particularly concerning, as human-like machines may appear familiar and trustworthy as their civilian use expands, increasing the risk of misjudging danger.
Van Rooijen advocates for international regulations to de-escalate the current technological arms race.
- Inside the secretive and lucrative world of orchid breeding
- Not so dusty: How tech is changing woodworking
- Why Spotify has no button to filter out AI music
