Small AI models let drones autonomously identify and attack battlefield targets
Small AI models let drones autonomously identify and attack battlefield targets
小型人工智能模型助力无人机在战场上自主识别并攻击目标
As European militaries adapt to the use of AI and drones in modern warfare, a NATO-backed startup is helping to deploy AI-driven target detection and selection that can run on small drones for surveillance and attack missions. 随着欧洲各国军队不断适应人工智能和无人机在现代战争中的应用，一家由北约支持的初创公司正在协助部署人工智能驱动的目标探测与选择系统，该系统可在小型无人机上运行，用于执行侦察和攻击任务。
The company, Scaleout Systems, was originally founded by researchers from Uppsala University in Sweden in 2018, and initially focused on training and deploying machine learning models directly on the hardware available in commercial trucks and other vehicles. But once Russia launched its full-scale invasion of Ukraine in 2022, the company pivoted toward defense applications. 这家名为 Scaleout Systems 的公司最初由瑞典乌普萨拉大学的研究人员于 2018 年创立，起初专注于在商用卡车及其他车辆的现有硬件上直接训练和部署机器学习模型。但在 2022 年俄罗斯对乌克兰发动全面入侵后，该公司将业务重心转向了国防应用。
“With the war in Ukraine and a shifting world, we realized that this technology can be very important to operationalize edge data and sensor data for machine learning to make sure that NATO allies have found that strategic advantage,” Andreas Hellander, cofounder and CEO of Scaleout Systems, told Ars. Scaleout Systems 的联合创始人兼首席执行官安德烈亚斯·赫兰德（Andreas Hellander）在接受《Ars Technica》采访时表示：“随着乌克兰战争的爆发和世界局势的变迁，我们意识到这项技术对于将边缘数据和传感器数据转化为机器学习应用至关重要，这能确保北约盟国获得战略优势。”
Instead of using frontier AI models from OpenAI or Anthropic, Scaleout is harnessing leaner ones, such as machine learning models that can perform computer vision tasks on the hardware of drones or computers used at forward bases. “They need to fit on forward-deployed hardware and edge hardware, which can vary quite a bit from small embedded devices to quite powerful edge workstations,” Hellander said. Scaleout 没有使用 OpenAI 或 Anthropic 的前沿人工智能模型，而是利用更精简的模型，例如能够在无人机硬件或前线基地计算机上执行计算机视觉任务的机器学习模型。赫兰德说：“它们必须适配前线部署的硬件和边缘硬件，这些硬件差异很大，从小型嵌入式设备到功能强大的边缘工作站不等。”
Scaleout was selected to join NATO’s Defence Innovator Accelerator for the North Atlantic (DIANA) Challenge Program in 2025. There, it has worked on the Federated Aerial Intelligence for Recon project to adapt machine learning models for the edge computing hardware found in drones, drone pilot tablets, and field command posts. Scaleout 入选了北约 2025 年“北大西洋国防创新加速器”（DIANA）挑战计划。在该项目中，他们致力于“联邦空中侦察智能”项目，旨在使机器学习模型能够适配无人机、无人机飞行员平板电脑以及野战指挥所中的边缘计算硬件。
Such AI models can help the drone operators with tasks like target identification, even as the drone’s cameras and sensors collect data from the surrounding battlefield environment. They can then intermittently share selective updates with computing nodes at the local platoon or company headquarters without transmitting sensitive raw data. 即使在无人机的摄像头和传感器从周围战场环境收集数据的同时，这些人工智能模型也能协助无人机操作员完成目标识别等任务。随后，它们可以间歇性地与当地排级或连级指挥部的计算节点共享选择性更新，而无需传输敏感的原始数据。
Those headquarters computing nodes help to retrain the AI models on the new battlefield data aggregated from multiple sources, before pushing the updated capabilities out to the edge devices when the opportunity arises. 这些指挥部的计算节点利用从多个来源汇总的新战场数据来重新训练人工智能模型，并在时机成熟时将更新后的能力推送到边缘设备。
“Models might have been trained in a desert environment, and if we try to deploy them in an urban environment, they’re not going to perform well,” Hellander told Ars. “If we can release several new versions of this model that—during the course of a single day or certainly an operation—keep learning and keep improving from this massive amount of sensor data that is generated at a practical edge, that is the sustainable advantage.” “模型可能是在沙漠环境中训练出来的，如果我们试图将其部署在城市环境中，它们的表现就不会理想，”赫兰德告诉《Ars》。“如果我们能发布该模型的几个新版本，使其在一天之内或某次行动过程中，能够不断从实际边缘产生的海量传感器数据中学习并改进，这就是可持续的优势。”
Testing the technology in live exercises
在实战演习中测试该技术
The technology means military drones and devices could benefit from local AI models that operate independently without relying on continuous communications with a central server hosting larger AI models in a data center. Reliance on a centralized location to run AI models looks riskier at a time when large data centers have been targeted and destroyed during the war between the US and Iran. This approach is also incredibly useful on modern battlefields where electronic warfare and enemy jamming can frequently interfere with communication signals. 这项技术意味着军用无人机和设备可以受益于本地人工智能模型，这些模型能够独立运行，无需依赖与托管大型人工智能模型的数据中心服务器进行持续通信。在美伊冲突期间，大型数据中心成为攻击目标并被摧毁，这表明依赖中心化位置运行人工智能模型风险更高。这种方法在现代战场上也极其有用，因为电子战和敌方干扰经常会干扰通信信号。
A growing number of Ukrainian military drones are already incorporating onboard AI capabilities into cheap kamikaze drones. “We built a functioning concept of how we do this for a surveillance and reconnaissance system based on a drone,” Hellander told Ars. “This is something we have also done in public demonstrations in Sweden.” 越来越多的乌克兰军用无人机已将机载人工智能功能集成到廉价的自杀式无人机中。“我们构建了一个功能性概念，展示了如何将其应用于基于无人机的监视和侦察系统，”赫兰德告诉《Ars》。“这也是我们在瑞典公开演示中所展示的内容。”
Scaleout is participating in the Affordable Loitering Modular Ammunition (ALMA) project headed by BAE Systems Bofors that aims to develop a low-cost, autonomous kamikaze drone. The ALMA concept was first publicly demonstrated during a Winter Demo 2026 event held in Sweden in January. That demonstration showed how a drone could autonomously “detect, identify and geolocate all potential spotted threats with the use of AI,” according to a Scaleout Systems presentation about ALMA’s capabilities. “All data is handled by dedicated onboard computing, allowing the system to perform in real-time without any external processing.” Scaleout 正在参与由 BAE 系统公司博福斯（BAE Systems Bofors）牵头的“经济型巡飞模块化弹药”（ALMA）项目，旨在开发一种低成本的自主自杀式无人机。ALMA 概念于今年 1 月在瑞典举行的“2026 冬季演示”活动中首次公开展示。根据 Scaleout Systems 关于 ALMA 能力的介绍，该演示展示了无人机如何利用人工智能自主“探测、识别并定位所有潜在的威胁目标”。“所有数据均由专用的机载计算设备处理，使系统能够在无需外部处理的情况下实时运行。”
Using its onboard AI capabilities, the drone automatically prioritized the highest-value target as defined by its mission—in this case, an armored engineering vehicle—and flew to that target to drop an explosive on it. A human operator could still control and direct the drone, but the drone carried out the mission on its own without direct human commands. 利用其机载人工智能功能，无人机会自动优先处理任务定义中价值最高的目标（在本例中为一辆装甲工程车），并飞向该目标投下爆炸物。人类操作员仍然可以控制和指挥无人机，但无人机在没有直接人工指令的情况下也能自主执行任务。
In June, Scaleout also tested its technology at a Swedish Air Force base in Uppsala, where the Swedish military already has a license to use Scaleout’s main software platform. That demonstration showed how a forward-deployed computing node at the military base could still run its own AI inference and active-learning processes after losing connection with a central computing node in Scaleout Systems’ lab. Once the connection was restored, the local AI model updates were shared with the central computing node. 今年 6 月，Scaleout 还在乌普萨拉的一个瑞典空军基地测试了其技术，瑞典军方已获得使用 Scaleout 主软件平台的许可。该演示展示了部署在军事基地的前线计算节点在与 Scaleout Systems 实验室的中央计算节点失去连接后，仍能运行其人工智能推理和主动学习进程。一旦连接恢复，本地人工智能模型的更新就会与中央计算节点共享。
This federated learning strategy, which allows the decentralized network of AI models to learn from aggregated data, could eventually scale across entire geographic regions or countries, Hellander explained. “In principle, you can unlock collaboration between NATO member states,” he said. 赫兰德解释说，这种联邦学习策略允许去中心化的人工智能模型网络从汇总的数据中学习，最终可以扩展到整个地理区域或国家。“原则上，你可以开启北约成员国之间的协作，”他说。