In early 2024, the Defense Advanced Research Projects Agency — DARPA, the U.S. military's primary research funding body — quietly updated its robotics solicitation portfolio to include humanoid-form platforms under its Robotic Autonomy in Complex Environments with Resiliency programme. It was not a headline announcement. The language was technical and the funding amounts modest relative to DARPA's overall budget. But the signal was clear: U.S. defense research had formally identified humanoid robots as a relevant technology category, not just for logistics and support functions, but for autonomous operation in environments that are explicitly described as contested and degraded.
That development sits awkwardly with the way humanoid robotics companies talk about their products in public. The dominant commercial narrative is warehouse logistics, automotive manufacturing, and eventually elder care. The defense angle is rarely raised in press materials, investor briefings, or the trade press coverage that follows every product announcement. It exists in a separate conversation — in government procurement documents, defence contractor white papers, and academic robotics conferences with dual-use research tracks — that does not often intersect with the consumer-facing story of helpful humanoid robots learning to fold laundry and carry boxes.
The gap between those two conversations is worth examining directly.
What Defense Agencies Are Actually Interested In
Defense interest in humanoid robots is not new. Boston Dynamics — whose Atlas robot is the most technically recognisable humanoid platform in public awareness — was founded in 1992 and spent its first two decades primarily on defense contracts, including substantial DARPA funding for legged robot research. The BigDog quadruped, which preceded Atlas, was explicitly developed for military logistics: carrying equipment over terrain that wheeled vehicles couldn't navigate. That lineage is not incidental. It shaped the technical priorities of the company and, by extension, of the broader legged robotics field.
What defense agencies are interested in today has several distinct components. The first is logistics and sustainment: the unglamorous work of moving supplies, ammunition, and equipment in environments where roads are unreliable, vehicles are vulnerable, and human exposure to risk is a genuine operational constraint. A robot that can carry a load over uneven terrain, navigate debris, and operate for extended periods without human guidance has clear military utility — and many of the technical capabilities that make a humanoid robot useful in a warehouse are directly transferable to that mission profile.
The second is hazardous environment operations: working in spaces contaminated by chemical, biological, radiological, or nuclear material; clearing buildings prior to human entry; handling improvised explosive devices. These are tasks that militaries already use robotic systems for — typically wheeled or tracked platforms with manipulator arms — but where a humanoid form factor offers potential advantages in accessing spaces designed for human bodies: climbing stairs, opening doors, operating equipment that requires two-handed interaction.
The third, and most contested, is combat-adjacent autonomy: robots that can operate independently in conditions where communications are degraded, where human oversight is limited, and where the operating environment is hostile. This is where the language of military robotics procurement becomes careful and where the distinction between logistics support and weapons platform begins to matter enormously.
What Companies Are Saying and Not Saying
The major commercial humanoid robotics companies occupy a spectrum on this question. At one end are companies that have made explicit public commitments to prohibit weapons development on their platforms. Boston Dynamics published an open letter in October 2022 — co-signed by several other robotics companies — pledging not to weaponise their robots and calling on the industry to adopt similar norms. The letter was notable for being proactive rather than reactive, and for naming the concern directly rather than deflecting it.
At the other end are companies that have pursued defense contracts explicitly and positioned their platforms as military-capable. Ghost Robotics, which makes a quadruped platform that visually resembles Boston Dynamics' Spot, has demonstrated its Q-UGV robot with rifle-mounting hardware at defence exhibitions and has active contracts with U.S. military branches. Ghost Robotics is not a humanoid company, but it illustrates that the commercial-to-military pipeline in legged robotics exists and is operational.
The major humanoid companies — Figure AI, Apptronik, Agility Robotics, 1X Technologies — have generally positioned themselves in the commercial camp, focusing their public communications entirely on industrial and consumer applications. What they have not done, with a few exceptions, is publish binding commitments about what their technology will not be used for, or established contractual restrictions in their licensing and sales agreements that would prevent defense use. The distinction between "we don't plan to pursue military contracts" and "our platform cannot be used for military purposes" is significant, and the former is the position most companies are actually in.
Tesla's Optimus occupies a different position by virtue of its scale ambitions. Elon Musk has stated publicly that he expects Optimus to be produced in quantities of millions. A general-purpose humanoid robot manufactured at that scale, with software that can be updated remotely and capabilities that are deliberately broad, is a dual-use technology by definition — not because Tesla has military intentions, but because at sufficient volume and capability, the technology becomes available to any state or non-state actor that can acquire it, and no commercial deployment policy prevents that.
The Capability Transfer Problem
The most technically substantive dimension of the dual-use question is not about intent. It is about the relationship between the capabilities that commercial humanoid development is producing and the capabilities that would be required for military applications.
Commercial development is currently focused on several capability areas that have direct military relevance. Navigation in unstructured environments — the ability to walk over uneven surfaces, climb obstacles, and maintain balance in conditions that weren't specifically prepared for the robot — is a core commercial challenge and a core military requirement. Manipulation of objects that weren't specifically designed for robot handling, including objects with variable weight and geometry, is another. Long-duration operation with limited human supervision is a third.
Each of these capabilities advances primarily through the commercial development pipeline right now, because that pipeline has access to more data, more compute, more deployment environments, and more engineering talent than any defence-specific programme could efficiently marshal. The result is that military-relevant robotic capability is being developed at scale, at speed, and at private expense — and the institutional mechanisms for thinking about what that means for international security are not keeping pace.
Arms control frameworks for autonomous weapons systems exist in draft form at the United Nations level, where a Group of Governmental Experts has been meeting since 2014 to discuss lethal autonomous weapons systems (often abbreviated as LAWS). Those discussions have produced statements of principles but no binding treaty. The pace of commercial humanoid development is likely to outrun whatever regulatory framework eventually emerges from that process, not because the process is moving slowly — though it is — but because the technology is advancing faster than any multilateral negotiation has historically moved.
The Export Control Dimension
One area where policy has moved faster than the broader dual-use debate is export controls. The United States has progressively tightened export restrictions on advanced robotics and AI-adjacent technologies, particularly in the context of competition with China. The Commerce Department's Entity List includes several Chinese robotics and AI companies, and the October 2023 export control rules on advanced semiconductors have downstream implications for the computing hardware that current-generation humanoid robots depend on.
China's own humanoid robotics industry has developed rapidly under state direction, with companies including Unitree, UBTECH, and Fourier Intelligence producing platforms that are increasingly competitive with their U.S. and European counterparts on specific capability metrics. The Chinese government has explicitly identified humanoid robotics as a strategic technology priority, with a Ministry of Industry and Information Technology roadmap published in 2023 targeting domestic humanoid production at scale by 2025–2027. The dual-use framing is built into that strategic posture from the outset, not treated as a separate concern to be managed after commercial development succeeds.
The export control response to this dynamic has been to restrict component and technology flows rather than to address the strategic question directly. It slows technology transfer but does not resolve the underlying question of how to think about general-purpose humanoid platforms that exist in both commercial and military capability spaces simultaneously.
What an Honest Conversation Looks Like
The humanoid robotics industry has, to its credit, produced more public engagement with the dual-use question than many previous robotics generations. The 2022 Boston Dynamics letter was a genuine attempt to establish a norm before the norm was needed. Academic robotics researchers have published on the ethics of autonomous systems with unusual candour relative to other technology fields.
What the industry has not yet produced is a systematic framework for what commercial humanoid companies are actually committing to — not in terms of their own intentions, but in terms of verifiable restrictions on how their technology can be used by downstream customers, modified by third parties, or acquired by state actors. The gap between stated position and binding commitment is where the real dual-use risk lives.
This is not a problem unique to humanoid robotics. It is the standard shape of dual-use technology governance across most technology categories: voluntary norms, export controls at the margins, and a structurally inadequate international framework that lags the technology by years or decades. What is different in the humanoid case is the combination of general-purpose physical capability, civilian-scale manufacturing ambitions, and the speed at which the technology is advancing.
The companies building humanoid robots are, in most cases, genuinely focused on commercial applications. That focus is real, and the warehouse logistics story is not a cover for anything. But technology at sufficient capability and scale does not stay within the boundaries of its developers' intentions. The dual-use question will arrive for humanoid robotics whether the industry engages with it or not. The only variable is whether it arrives with a framework in place or without one.