Inside Anduril and Meta’s quest to make smart glasses for warfare

深入安杜里尔（Anduril）与 Meta 的合作：打造战争用智能眼镜

EXECUTIVE SUMMARY The defense-tech company Anduril has shared new details about the augmented-reality headset for the military it’s prototyping with Meta, including a vision for ordering drone strikes via eye-tracking and voice commands. 执行摘要 国防科技公司安杜里尔（Anduril）近日披露了其与 Meta 合作研发的军用增强现实（AR）头显的更多细节，其中包括通过眼球追踪和语音指令下达无人机打击任务的构想。

Quay Barnett, who leads the efforts as a vice president at Anduril following a career in the Army’s Special Operations Command, says his fundamental goal is to optimize “the human as a weapons system.” The vision is undoubtedly cyborg-inspired: Barnett wants drones and soldiers to see together, share information seamlessly, and make decisions as one. 曾在陆军特种作战司令部服役、现任安杜里尔副总裁并负责该项目的 Quay Barnett 表示，他的根本目标是优化“作为武器系统的人类”。这一愿景无疑受到了“赛博格”（半机械人）概念的启发：Barnett 希望无人机与士兵能够实现视觉共享、信息无缝同步，并作为一个整体做出决策。

Anduril actually has two such projects in the works. The first is the Army’s Soldier Born Mission Command, or SBMC, for which the company won a $159 million prototyping contract last year to work with Meta on augmented-reality glasses to attach to existing military helmets. 安杜里尔目前实际上有两个此类项目正在进行中。第一个是陆军的“士兵任务指挥系统”（SBMC），该公司去年赢得了 1.59 亿美元的原型设计合同，将与 Meta 合作开发可安装在现有军用头盔上的增强现实眼镜。

But Anduril has also embarked on a self-funded side quest, announced in October, to design its own helmet and headset combo called EagleEye. This is something the military has not asked for, but Anduril insists it will prefer it and purchase it in the end. 但安杜里尔还在 10 月宣布启动了一个自筹资金的副项目，旨在设计名为“鹰眼”（EagleEye）的头盔与头显组合。虽然军方并未提出此项需求，但安杜里尔坚信军方最终会青睐并采购该产品。

So far, both systems are years away. The Army isn’t expected to move its top choice for the SBMC program into production until 2028, if it picks one at all (the previous lead for the effort, Microsoft, was set to receive a $22 billion production contract that was ultimately cancelled when the glasses didn’t prove viable). 到目前为止，这两个系统距离投入使用还有数年时间。预计陆军要到 2028 年才会将其 SBMC 项目的首选方案投入生产（前提是最终选定方案；该项目的前任主导者微软曾有望获得一份 220 亿美元的生产合同，但因眼镜产品被证明不可行而最终被取消）。

But Barnett told MIT Technology Review about where both Anduril’s prototypes are headed. Depending on the situation, the glasses for either prototype will overlay certain information onto a soldier’s field of view. This might be as simple as a compass or as complex as an entire map of the area, information about where nearby drones are flying, or AI-driven recognition of a target like a truck. 不过，Barnett 向《麻省理工科技评论》介绍了安杜里尔两个原型机的研发方向。根据不同情况，这两款原型眼镜都会在士兵的视野中叠加特定信息。这些信息可能简单如指南针，也可能复杂如整个区域地图、附近无人机的飞行位置，或是由人工智能识别出的目标（如卡车）。

The soldier would then speak to the interface in plain language—for example, to order an evacuation for someone who’s been injured or to plan a route taking into account which areas are off limits. A large language model—Anduril is in tests with Google’s Gemini, Meta’s Llama, and even Anthropic’s Claude, despite the company’s conflict with the Pentagon—will be used to help translate a soldier’s speech into commands the software can follow. 士兵随后可以用通俗语言与界面对话——例如，下令撤离伤员，或规划一条避开禁区的路线。安杜里尔正在测试谷歌的 Gemini、Meta 的 Llama，甚至 Anthropic 的 Claude（尽管该公司与五角大楼存在冲突），这些大语言模型将被用于帮助将士兵的语音转化为软件可执行的指令。

And the engine for it all will be Anduril’s software Lattice, which incorporates data from lots of different military hardware into one picture. The Army announced in March that it would spend $20 billion to integrate Lattice with essentially its entire infrastructure. 这一切的核心引擎将是安杜里尔的软件 Lattice，它能将来自多种不同军事硬件的数据整合到一个画面中。陆军在 3 月宣布，将斥资 200 亿美元将 Lattice 与其几乎所有的基础设施进行整合。

Barnett’s team is designing the headset to carry out multi-step tasks. A soldier might send a drone to surveil an area and instruct it to come back once it’s found something that looks like an artillery unit; then the system would recommend courses of action, like sending a nearby drone to strike, that would have to be approved by the normal chain of command. Barnett 的团队正在设计能够执行多步骤任务的头显。士兵可以派遣无人机侦察区域，并指示其在发现疑似炮兵单位时返回；随后，系统会推荐行动方案（例如派遣附近的无人机进行打击），这些方案仍需通过正常的指挥链进行审批。

Leading the system through this, if all goes to plan, might not even require speech; the soldier could instead communicate through tracked eye movements and subtle taps. That’s the idea, anyway. It’s worked on early prototypes, Barnett says, but there aren’t yet versions ready for the Army to test at scale. The component parts began arriving in March. 如果一切按计划进行，引导系统执行这些任务甚至无需语音；士兵可以通过眼球追踪和轻微的触碰动作进行交互。至少构想是如此。Barnett 表示，该功能在早期原型机上已实现，但目前还没有可供陆军进行大规模测试的版本。相关组件已于 3 月开始到货。

Because of federal military contracting rules, these parts—unlike Meta’s commercial smart glasses—required new supply chains that don’t rely on Chinese companies. 由于联邦军事合同规则的限制，这些零部件与 Meta 的商用智能眼镜不同，必须建立不依赖中国企业的新供应链。

It’s a lot for soldiers already bogged down in information overload, says Jonathan Wong, a former US Marine who works as a senior policy researcher at RAND on Army efforts to buy new tech. Both smart glasses projects aim to create a clean interface that presents only the right information at the right time. But it’s a product that soldiers will reject if it costs more of their attention than it saves. 兰德公司（RAND）高级政策研究员、前美国海军陆战队队员 Jonathan Wong 表示，对于已经深陷信息过载的士兵来说，这负担太重了。这两个智能眼镜项目都旨在创建一个简洁的界面，只在正确的时间呈现正确的信息。但如果该产品消耗的注意力多于节省的注意力，士兵们将会拒绝使用它。

“How much mental bandwidth do you have to be both aware of your surroundings and to operate this technology in a way that makes you and your whole unit better?” he says. Wong recalls that as a platoon commander, for example, he had a radio that operated on three different channels at once. “The moment that two people were on different channels talking at the same time, I immediately couldn’t comprehend anything that either one of them was trying to tell me, and I was probably not aware of my own surroundings,” he says. “I think there are limits to what you can take in.” “你到底有多少心理带宽，既能保持对周围环境的感知，又能操作这种技术来提升你和你整个单位的作战能力？”他问道。Wong 回忆说，作为排长时，他曾使用过一台同时运行三个频道的无线电。“当两个人同时在不同频道说话时，我立刻就无法理解他们中任何一个人想表达的内容，而且我可能也失去了对周围环境的感知，”他说，“我认为人类接收信息是有极限的。”

Ideally, Barnett says, smart glasses can ease that information overload. Anduril’s approach is to get creative with ways the user can access necessary information quickly. Voice commands and eye tracking are a piece of that strategy. But even if it’s all technically feasible, it might take years of field testing to know if the system is actually useful for soldiers, Wong says. Barnett 表示，理想情况下，智能眼镜可以缓解这种信息过载。安杜里尔的方法是通过创造性的方式让用户快速获取必要信息。语音指令和眼球追踪只是该策略的一部分。但 Wong 指出，即使这一切在技术上可行，也可能需要多年的实地测试才能确定该系统对士兵是否真正有用。

Such a system would mark a major escalation in how closely soldiers rely on imperfect AI systems. While computer vision models used to identify objects have long been employed by militaries, and chatbots have recently entered decision-making during the war in Iran, these technologies have not yet made their way to most frontline soldiers. A smart glasses system tasked with identifying threats and recommending strikes would introduce massive new risks of errors. 这样的系统标志着士兵对不完善的人工智能系统的依赖程度大幅升级。虽然用于识别物体的计算机视觉模型早已被军队采用，聊天机器人也已进入伊朗战争的决策环节，但这些技术尚未普及到大多数前线士兵。一套负责识别威胁并推荐打击目标的智能眼镜系统，将带来巨大的新错误风险。

Anduril is not the only one competing to develop smart goggles for combat. Rivet, which specializes in wearable sensors for the military, received a $195 million prototyping contract the same time, and in March the Israeli defense-tech company Elbit received its own $120 million contract. This all comes after Microsoft lost its role leading the Army’s smart glasses effort, following a Pentagon audit that found the Army wasn’t properly testing the glasses, a mistake that could have wasted $22 billion. 安杜里尔并非唯一一家竞标开发战斗智能眼镜的公司。专注于军用可穿戴传感器的 Rivet 公司在同一时间获得了 1.95 亿美元的原型设计合同，以色列国防科技公司埃尔比特（Elbit）也在 3 月获得了 1.2 亿美元的合同。在此之前，微软因五角大楼的一项审计而失去了陆军智能眼镜项目的主导权，该审计发现陆军未能对眼镜进行适当测试，这一失误本可能导致 220 亿美元的浪费。

For both Anduril’s prototypes, the company is testing a new system for digital night vision, which uses electronic sensors and algorithms to boost low levels of light. It’s been a promised technology for decades but has tended to… 对于安杜里尔的两个原型机，该公司正在测试一种新型数字夜视系统，该系统利用电子传感器和算法来增强微光。这项技术几十年来一直被寄予厚望，但往往……