What are humanoid robots actually for on the battlefield?

3d rendering humanoid robots working with headset and notebook (Getty images)

Humanoid robots are approaching the battlefield , but not because their robotic legs and arms make them better infantrymen. Their near-term military value lies in doing dangerous physical work inside environments, vehicles and infrastructure designed for human bodies.

The decisive test will be whether they can sustain the wider unmanned force at acceptable cost — not whether they can imitate a soldier in a demonstration. Militaries should therefore treat humanoids as a specialized enabling capability, not a replacement army.

The war in Ukraine has moved unmanned ground systems from the margins of force design to the center of daily operations. In the first quarter of 2026, Ukrainian ground robotic systems conducted about 24,500 missions , including more than 9,000 in March. The number of Ukrainian units employing them rose from 67 in November 2025 to 167 four months later, and Ukraine’s Ministry of Defense subsequently announced plans to contract 25,000 ground robots in the first half of 2026, setting the aspirational goal of transferring all frontline logistics to robotic systems where possible.

These operations are all being done by simple wheeled or tracked machines — affordable, replaceable and adequate. But experimentation with humanoid systems is beginning. Foundation sent two Phantom MK-1 humanoids to Ukraine in February 2026 for evaluation, although available reporting suggests testing rather than verified direct combat employment. In July, Ukraine’s Brave1 initiative identified humanoid robots as a priority for forthcoming defense-technology grant competitions. China has also demonstrated a teleoperated humanoid that mirrored the movements of a human controller.

These cases do not generate combat operations utility, but they do show that several defense entrepreneurs and businesses now consider the question serious enough to fund and test prototypes. The relevant question for these experiments to solve is not whether humanoids will supersede unmanned aerial vehicles (UAVs), unmanned ground vehicles (UGVs) or soldiers. It is whether a human-shaped machine can perform important tasks that these alternatives cannot perform safely or economically.

That is a narrower proposition, but also a more credible one.

The humanoid form has one compelling advantage: most of the physical world is built for people. Doors, stairs, ladders, hatches, tools, control panels, vehicle cabins, ship passageways and industrial machinery assume human height, reach and dexterity. A tracked robot may carry more weight and a quadruped may cross rough ground more reliably, but neither can readily enter a building, climb to another floor, open an electrical cabinet, replace a component and use the tools already stored there.

Building logistics networks for contested environments

Explore how software, edge AI and resilient connectivity can strengthen military logistics in contested environments. Join Breaking Defense on Oct. 21 and earn 1 CPE credit.

This advantage points to three initial mission sets.

First, humanoids can perform hazardous work in human-designed spaces. Urban breaching, tunnel reconnaissance, explosive-ordnance disposal, inspection of damaged infrastructure, shipboard damage control and handling chemical or radiological hazards all expose personnel to substantial risk. Early systems will probably be teleoperated because these environments are too complex, communications too uncertain and the consequences of error too high for broad autonomy. Even then, a robot capable of opening doors, climbing stairs and manipulating standard equipment could extend the reach of engineers, medical teams and explosive-ordnance specialists.

Second, humanoids may provide a physical interface between commanders and autonomous systems. As formations acquire hundreds of heterogeneous robots, soldiers cannot individually service and control every platform. A humanoid could eventually execute a commander’s intent locally: moving sensors, distributing batteries, organizing launch cycles, recovering disabled systems and adapting a position to changing requirements. This is not the same as giving a robot command authority. It is delegating physical tasks within boundaries established by human leaders.

Third, and perhaps most consequentially, humanoids may become the maintainers of the unmanned force. Large robotic fleets come with physical burdens that software cannot solve on its own. Drones require batteries, payloads, inspection, launch preparation, recovery, cleaning and repair. Ground vehicles require recharging, damaged-component replacement, payload changes and retrieval. Maritime systems return to piers and support ships whose fittings were designed for sailors. A sufficiently dexterous humanoid could service several types of systems using existing tools and infrastructure, allowing human technicians to remain farther from enemy observation and strike.

If one humanoid can turn around multiple UAVs or UGVs, establish a temporary launch site, or keep a robotic logistics detachment operating through the night, its value comes from the additional sorties and reduced human exposure it enables. The appropriate measure of success is therefore not how closely it resembles a soldier; it is how many human labor hours, risky movements, or aborted robotic missions it eliminates.

Technology, Survivability and Control

The obstacles are formidable. Commercial humanoids usually operate on smooth floors, near reliable power and within reach of technicians. Battlefields add mud, rubble, rain, thermal extremes, blast effects, electromagnetic interference and deliberate attack. A robot that falls and cannot recover becomes an obstacle. A sophisticated platform that requires factory repair after minor damage may be less useful than a crude UGV repaired by soldiers with common parts.

Military requirements should consequently emphasize endurance, field repair and gradual degradation over theatrical performance. A useful system needs swappable power, sealed electronics, back-drivable and durable actuators, self-recovery after falls and hands or interchangeable end-effectors that can manipulate existing tools and connectors. It also needs navigation without continuous satellite positioning; resilient, low-probability-of-detection communications; and control modes that shift among teleoperation, supervised autonomy and preplanned action when links fail.

Survivability will also be a factor, but unlike protecting a real human, armor may not be a good solution. Weight devoted to protection reduces endurance and payload, while a human-sized thermal and visual signature may attract immediate fire. Dispersion, concealment, low acoustic and electromagnetic signatures, rapid repair and the ability to abandon a mission may matter more than hardening. Planners must also assume cyber compromise, spoofing and capture. Sensitive mission data and software should be compartmented, encrypted and erasable.

Weapons pose a separate issue. The legal and ethical problem is not a robot’s shape but the degree of autonomy in selecting and engaging targets. Unarmed logistics, reconnaissance and engineering roles present a much lower barrier to fielding. Arming humanoids would require weapons reviews, explicit command responsibility, technically reliable abort mechanisms and rules governing human judgment over lethal force. International debate over autonomous weapons remains unsettled, and the International Committee of the Red Cross, for example, continues to advocate for strict limits and effective human supervision . Early military adoption should not wait for every legal question to be resolved, but it should begin with missions that do not require machines to make life-and-death decisions.

Force Design and Acquisition

Humanoid robots should initially be attached to organizations that already understand specialized equipment: engineers, explosive-ordnance teams, logistics units, maintenance formations and unmanned-systems units. Creating independent “robot infantry” formations would put institutional branding ahead of demonstrated capability. Small operational test detachments can instead identify where the humanoid form produces a measurable advantage and where wheels, tracks, quadrupeds, or human labor remain superior.

Acquisition authorities should demand comparisons, not demonstrations. Every proposed mission should be tested against a human team and the cheapest suitable robotic alternative. Relevant measures include cost per completed mission, mean time between failures, recovery and repair rates, operator workload, battery demand, transport burden and human exposure avoided. These metrics would prevent an impressive prototype from becoming an expensive program without a defensible concept of operations.

The procurement model should also reflect the technology’s immaturity. Militaries should buy limited prototype batches, expose them to realistic field conditions and update hardware and software in short cycles. Open interfaces for batteries, tools, payloads, control software and data links are essential; otherwise, each manufacturer will create a closed ecosystem that cannot support the mixed robotic fleets already emerging. Most near-term resources should remain with proven, cheaper unmanned systems, while humanoid programs compete for expansion by demonstrating unique operational value.

Doctrine would likely evolve alongside the hardware. Commanders need to know who controls the robot, who authorizes its actions, what happens when communications fail and when recovery is worth risking other assets. Units will require new maintenance skills and stocks of actuators, batteries, sensors and computing modules. Training should emphasize human-machine teaming under electronic attack rather than choreographed peacetime demonstrations.

Humanoid robots are unlikely to arrive first as massed mechanical infantry. Their more plausible path is quieter: the robot that enters a contaminated compartment, carries equipment up a stairwell, services a drone launch point or repairs another machine while soldiers remain under cover. Such tasks lack the drama of a humanoid assault, but they address real manpower and survivability problems.

In other words, humanoid robots are not standalone combat platforms; rather, they are platforms that ensure the continuity of the combat robotic ecosystem. Humanoids are the humans missing from the digital battlefield.

The Ukrainian experience shows that robotic warfare rewards systems that are useful, numerous, adaptable and integrated with existing forces. Humanoids will have to meet the same standard. If they can exploit human-designed infrastructure and sustain a larger robotic ecosystem, they may become an important layer of future force design. If they cannot outperform simpler alternatives on cost, reliability and operational effect, they will remain impressive machines in search of a mission.

Yavuz Turkgenci is a retired three-star general of the Turkish Armed Forces whose career traversed across several offices, including Western European and NATO posts and the Commandant of the Turkish Third Field Army. He holds a doctorate in security strategy design and management.

3D渲染人形机器人佩戴头戴式显示器和笔记本电脑（Getty图片）

人形机器人正逐渐接近战场，但这并非因为它们的机械腿和手臂使它们成为更优秀的步兵。它们近期的军事价值在于能够在专为人类设计的环境、车辆和基础设施中执行危险的体力劳动。

决定性的考验在于它们能否以可接受的成本维持更广泛的无人部队，而不是它们能否在演示中模仿士兵。因此，各国军队应该将人形机器人视为一种专门的支援能力，而不是一支替代军队。

乌克兰战争已将无人地面系统从部队设计的边缘推向日常作战的核心。2026年第一季度，乌克兰地面机器人系统执行了约24500次任务，其中3月份就超过9000次。使用此类系统的乌克兰部队数量从2025年11月的67支增加到四个月后的167支。随后，乌克兰国防部宣布计划在2026年上半年采购25000台地面机器人，并设定了尽可能将所有前线后勤保障转移到机器人系统的宏伟目标。

这些行动目前都由简单的轮式或履带式机械完成——价格低廉、易于更换且足以胜任。但人形机器人系统的试验正在启动。2026年2月，基金会向乌克兰派遣了两台“幻影”MK-1型人形机器人进行评估，但现有报道表明，此次评估更多是测试，而非已证实的直接作战应用。7月，乌克兰的“勇敢一号”计划将人形机器人列为即将举行的国防技术资助竞赛的优先项目。中国也展示了一款能够模仿人类操控者动作的遥控人形机器人。

这些案例虽然不能直接应用于实战，但确实表明一些国防领域的企业家和企业已经足够重视这个问题，并愿意资助和测试原型机。这些实验需要解决的关键问题并非人形机器人是否会取代无人机、无人地面车辆或士兵，而是人形机器能否安全或经济地执行其他替代方案无法完成的重要任务。

这是一个范围较窄的提议，但也更可信。

人形机器人有一个显而易见的优势：现实世界的大部分设施都是为人类设计的。门、楼梯、梯子、舱口、工具、控制面板、车辆驾驶室、船舶通道和工业机械都以人类的身高、臂展和灵活性为标准。履带式机器人或许能承载更大的重量，四足机器人或许能更可靠地穿越崎岖地形，但它们都无法轻易进入建筑物、爬到另一层楼、打开配电柜、更换部件并使用那里已有的工具。

为冲突环境构建物流网络

探索软件、边缘人工智能和弹性连接如何增强对抗环境下的军事后勤保障。参加10月21日举行的Breaking Defense研讨会，即可获得1个CPE学分。

这一优势指向三个初始任务集。

首先，人形机器人可以在人类设计的空间中执行危险作业。城市突围、隧道侦察、爆炸物处理、受损基础设施检查、舰船损管以及化学或放射性危害物处理等工作都会使人员面临巨大风险。早期系统可能需要远程操控，因为这些环境过于复杂，通信过于不稳定，而且一旦出错后果过于严重，难以实现完全自主。即便如此，能够开门、爬楼梯和操作标准设备的机器人也能拓展工程师、医疗队和爆炸物处理专家的工作范围。

其次，人形机器人可以为指挥官和自主系统之间提供物理接口。随着部队装备数百台异构机器人，士兵无法单独维护和控制每个平台。人形机器人最终可以在局部范围内执行指挥官的意图：移动传感器、分配电池、组织发射周期、回收故障系统以及根据不断变化的需求调整位置。这与赋予机器人指挥权不同，而是在人类指挥官设定的范围内委派实际任务。

第三，或许也是最重要的一点，人形机器人可能成为无人部队的维护者。庞大的机器人舰队会带来软件无法独自解决的物理负担。无人机需要电池、有效载荷、检查、发射准备、回收、清洁和维修。地面车辆需要充电、更换损坏部件、更换有效载荷和回收。海上系统需要返回码头和支援舰艇，而这些设施原本是为水手设计的。足够灵巧的人形机器人可以使用现有工具和基础设施来维护多种类型的系统，从而使人类技术人员能够远离敌方的观察和攻击。

如果一个人形机器人能够调转多架无人机或无人地面车辆的航线、建立临时发射场，或者维持一支机器人后勤分队通宵运作，那么它的价值就体现在它能够执行更多架次任务并减少人员暴露风险上。因此，衡量其成功的恰当标准并非它与士兵的相似程度，而是它能够节省多少人力工时、减少多少危险行动或避免多少次机器人任务的中止。

技术、生存能力和控制

障碍重重。商用人形机器人通常在平坦的地面上运行，靠近可靠的电源，并且便于技术人员维护。而战场则不然，泥泞、瓦砾、雨水、极端温度、爆炸冲击、电磁干扰以及蓄意攻击，都让它们难以应对。一旦机器人跌倒且无法恢复，就会成为战场上的障碍。即使是需要返厂维修的精密平台，在轻微损坏后也可能不如士兵用普通零件就能修复的简易无人地面车辆实用。

因此，军事需求应更侧重于耐久性、野外维修和渐进式性能衰减，而非舞台表现。一个实用的系统需要具备可更换电源、密封电子元件、可反向驱动且耐用的执行器、跌落后自动恢复功能，以及能够操作现有工具和连接器的可互换末端执行器。此外，它还需要无需持续卫星定位的导航能力；具有高可靠性且低被探测概率的通信能力；以及在链路故障时能够在远程操作、受控自主和预设动作之间切换的控制模式。

生存能力固然重要，但与保护真人不同，装甲可能并非理想之选。用于防护的重量会降低续航能力和有效载荷，而人体大小的热成像和视觉特征可能会立即招致火力攻击。分散性、隐蔽性、低声学和电磁特征、快速维修以及放弃任务的能力可能比加固更为重要。规划者还必须考虑到网络攻击、欺骗和数据截获的风险。敏感的任务数据和软件应进行隔离、加密并可擦除。

武器问题则另当别论。法律和伦理问题不在于机器人的外形，而在于其在选择和攻击目标方面的自主程度。非武装的后勤、侦察和工程任务部署门槛要低得多。而为类人机器人配备武器则需要武器审查、明确的指挥责任、技术上可靠的中止机制以及规范人类对致命武力使用判断的规则。关于自主武器的国际争论仍未有定论，例如，红十字国际委员会仍在倡导对自主武器的使用进行严格限制和有效的人工监督。早期军事应用不应等到所有法律问题都得到解决才开始，而应首先从那些不需要机器做出生死抉择的任务入手。

部队设计与采购

人形机器人最初应配属给那些已经熟悉专用设备的单位，例如工程师、爆破小组、后勤部队、维修单位和无人系统单位。组建独立的“机器人步兵”部队会将机构品牌置于实际能力之上。小型作战测试分队则可以更有效地识别人形机器人在哪些方面具有显著优势，以及在哪些方面轮式、履带式、四足式或人工操作仍然更胜一筹。

采购部门应该要求进行对比测试，而不是仅仅进行演示。每一项拟议任务都应与人类团队和成本最低且适用的机器人替代方案进行对比测试。相关衡量指标包括每次任务完成成本、平均故障间隔时间、恢复和维修率、操作员工作量、电池需求、运输负担以及避免的人员暴露风险。这些指标可以防止一个令人印象深刻的原型机变成一个缺乏可靠运行概念的昂贵项目。

采购模式也应体现技术的不成熟性。军方应采购少量原型机，将其置于真实的战场环境中进行测试，并定期更新硬件和软件。电池、工具、有效载荷、控制软件和数据链路的开放式接口至关重要；否则，每个制造商都会构建一个封闭的生态系统，无法支持目前正在涌现的混合型机器人集群。近期大部分资源应继续投入到成熟且成本更低的无人系统中，而人形机器人项目则应通过展现独特的作战价值来竞争扩张。

作战理论很可能会随着硬件的发展而演变。指挥官需要了解谁控制机器人，谁授权其行动，通信中断时会发生什么，以及何时值得冒着损失其他资产的风险进行恢复。部队需要掌握新的维护技能，并储备执行器、电池、传感器和计算模块。训练应侧重于在电子战攻击下的人机协同作战，而不是精心编排的和平时期演示。

人形机器人不太可能以大规模机械步兵的形式率先抵达战场。它们更可能采取的行动方式更为隐蔽：机器人潜入受污染的舱室，携带装备沿楼梯向上移动，维护无人机发射点或维修其他机器，而士兵则留在掩体后方。这类任务虽然不像人形机器人突袭那样引人注目，但却能切实解决人力和生存能力方面的问题。

换句话说，人形机器人并非独立的作战平台；相反，它们是确保作战机器人生态系统持续运转的平台。人形机器人弥补了数字战场上人类角色的缺失。

乌克兰的经验表明，机器人战争青睐那些实用、数量众多、适应性强且能与现有部队整合的系统。人形机器人也必须达到同样的标准。如果它们能够利用人类设计的基础设施并维持一个更大的机器人生态系统，它们就可能成为未来部队设计的重要组成部分。如果它们在成本、可靠性和作战效果方面无法超越更简单的替代方案，那么它们就只能是令人印象深刻却尚未找到合适任务的机器。

亚武兹·图尔克根奇是土耳其武装部队的一位退役三星将军，他的职业生涯横跨多个部门，包括西欧和北约的多个职位，以及土耳其第三野战军司令。他拥有安全战略设计与管理博士学位。