据一家乌克兰无人机制造商称,两年前在一次战场测试中,全自主无人机杀死了俄罗斯士兵。如果情况属实,这一事件将标志着这场战争中的又一个里程碑,这场战争推动了军用无人机、机器人和人工智能制导武器的空前发展。
乌克兰无人机制造商 Aero Center 的首席执行官亚历山大·科哈诺夫斯基在乌克兰驻伦敦大使馆主办的一场新闻发布会上接受《新科学家》采访时透露了这次一次性测试。科哈诺夫斯基描述称,这次测试(不涉及他目前所在的 Aero Center 公司)使用了预先编程的四旋翼无人机,它们飞往前线区域后,会激活一种由人工智能驱动的“终结者模式”,在该区域内搜寻并攻击任何目标。
显然,没有视频画面或其他任何东西来显示这些“终结者”无人机瞄准并攻击了什么。但科哈诺夫斯基告诉《新科学家》,派去检查后续情况的人工操控无人机发现了几名死亡的俄罗斯士兵,由此得出结论,是全自主无人机杀死了他们。
据《新科学家》报道,在乌克兰大使馆的活动上,国防公司的代表表示,乌克兰政府禁止在目标拦截的最后阶段使用人工智能。一名乌克兰军事指挥官也告诉《新科学家》,他的无人机操作员只使用半自主系统,始终由人类做出关键的控制决策。他阐述了乌克兰对“国际人道法”的承诺,同时强调军方始终“在决策过程中极其谨慎,以防止平民伤亡”。
考虑到这种方法的实际局限性以及国际人道主义法方面的考量,这项实验的一次性性质是合理的。派遣完全自主的无人机,在没有任何人类操作员干预的情况下,攻击给定区域内的任何目标,需要周密的预先规划,并且存在所谓的“友军误伤”事件或攻击平民非战斗人员的风险。目前也不清楚,与人类无人机飞行员相比,这些完全自主的四旋翼无人机在选择和攻击目标方面的效果如何。
根据联合国裁军事务厅的说法,目前对于何为致命性自主武器系统,尚无普遍认同的定义。但常见的描述将这类武器系统定义为具有自主性,能够“在没有人类行为者指导或输入的情况下执行其功能”。美国国防部的政策将致命性自主武器定义为“一旦启动,即可在没有人类操作员进一步干预的情况下选择和攻击目标的武器系统”。
自主AI在无人机武器库中的作用
据乌克兰政府前顾问卡特琳娜·邦达尔在为华盛顿特区的战略与国际研究中心(CSIS)智库撰写的报告中称,具备“在复杂和不可预测的环境中独立或极少监督下完成目标”能力的完全自主武器,在乌克兰战争中尚未成为战场现实。但她强调,越来越多的无人机集成了某些自主能力,用于导航,有时也用于目标锁定,即使人类操作员仍保持总体控制。
乌克兰和俄罗斯正在部署大量FPV无人机,用于侦察和打击车辆甚至单兵。这些无人机通常由训练有素的无人机飞行员操控,他们佩戴虚拟现实眼镜,从无人机的视角观察并瞄准敌方目标。此外,还有更大的四旋翼或多旋翼“轰炸机”无人机,能够携带更重的载荷,用于执行补给任务或在前线向敌方目标投掷爆炸物。
类似固定翼飞机的远程攻击无人机可能会融入更多自主决策能力。2025年,俄罗斯每晚向乌克兰城市发射数百架无人机,其中包括最初由伊朗提供、如今越来越多在俄罗斯本土生产的“沙希德”无人机。“沙希德”无人机通常预先编程,自动飞向目标,自主决策能力极低。但某些“沙希德”无人机变体,例如“天竺葵-2”,配备了走私的英伟达 Jetson Orin 微型计算机,能够提供机载视频处理和自主决策能力,包括自主目标识别与重新瞄准。
为防御俄罗斯“沙希德”无人机的这些攻击,乌克兰部署了涵盖低端与高端技术的防空系统,其中包括本土研发的低成本拦截无人机。据乌克兰政府平台 United 24 介绍,部分拦截无人机系统被设计为可自主飞抵拦截点并锁定目标——尽管仍需人类操作员进行初始目标选择、下达攻击指令,且始终保留取消攻击的能力。
与此同时,据乌克兰国防部称,乌克兰已具备技术和制造能力,每月可对超过20公里射程内的俄罗斯目标发动5000多次无人机打击。由于俄罗斯的电子战系统可能干扰人类操作员的通信链路,同时GPS干扰会扰乱GPS制导武器,这些中远程攻击无人机必然严重依赖自主导航能力。邦达尔在其战略与国际研究中心(CSIS)报告中写道,这种由AI驱动的导航已将乌克兰无人机打击的成功率从约10%至20%提升至70%至80%。
总体而言,乌克兰国防工业专注于在小型数据集上训练小型 AI 模型,以便在小型廉价芯片有限的算力上运行,邦达尔写道。这种方法催生了用于关键自主功能(包括导航和目标识别)的 AI 驱动软件,这些软件可与独立硬件模块打包,安装在小尺寸第一人称视角(FPV)无人机、远程打击无人机,甚至无人地面机器人的炮塔上。因此,即使完全自主的系统仍然罕见,但预计会有更多无人机和机器人获得特定的自主 AI 能力,以在战场上辅助人类决策。
本文于 6 月 12 日更新,补充了乌克兰当前反对使用此类完全自主无人机的政策的更多细节,并澄清科哈诺夫斯基的公司 Aero Center 并未参与该技术的一次性战场测试。
Fully autonomous drones killed Russian soldiers during a battlefield test two years ago, according to a Ukrainian drone manufacturer. If true, the incident would represent another milestone in a war that has spurred unprecedented developments in military drones, robots, and AI-guided weaponry.
The one-time test was revealed by Alexander Kokhanovskyy, CEO of the Ukrainian drone maker Aero Center, during an interview with New Scientist at a press event hosted by the Ukrainian embassy in London. Kokhanovskyy described the test—which did not involve his current company Aero Center—using quadcopter drones that were preprogrammed to fly to a front-line area before activating an AI-powered “Terminator mode” that would seek out and attack any target in the given area.
There was apparently no video feed or anything else to show what the “Terminator” drones targeted and attacked. But Kokhanovskyy told New Scientist that human-piloted drones sent to check out the aftermath found “a couple” of dead Russian soldiers, which led to the conclusion that the fully autonomous drones had killed them.
Defence company representatives at the Ukrainian embassy event said that the Ukrainian government bans the use of AI in the final stage of target interception, according to New Scientist. A Ukrainian military commander also told New Scientist that his drone pilots only use semi-autonomous systems that always have humans making crucial control decisions. He described Ukraine’s commitment to “international humanitarian law” while emphasizing that the military always exercises “great care in decision-making in order to prevent civilian casualties.”
The one-time nature of this experiment makes sense when considering the practical limitations of this approach, along with considerations regarding international humanitarian law. Sending fully autonomous drones to attack anything and everything in a given area without any human operator intervention requires careful preplanning and carries the risk of so-called “friendly fire” incidents or attacks on civilian noncombatants. It is also unclear how effective these fully autonomous quadcopter drones were in selecting and attacking targets compared to human drone pilots.
There is currently no commonly agreed definition of what constitutes a lethal autonomous weapon system, according to the United Nations Office for Disarmament Affairs. But common characterizations describe weapon systems with the autonomy to “perform their functions in the absence of direction or input from a human actor.” US Department of Defense policy has defined lethal autonomous weapons as “weapon system[s] that, once activated, can select and engage targets without further intervention by a human operator.”
The role of autonomous AI in drone arsenals
Fully autonomous weapons with the capability to “accomplish goals independently or with minimal supervision in complex and unpredictable environments” are not yet a battlefield reality in the war in Ukraine, according to Kateryna Bondar, a former advisor to the government of Ukraine, in her report for the Center for Strategic and International Studies (CSIS) think tank in Washington, DC. But she highlighted a growing number of drones integrating certain autonomous capabilities for navigation and sometimes targeting, even if human operators maintain overall control.
Ukraine and Russia are fielding many FPV drones for scouting and striking vehicles and even individual soldiers. These are typically controlled by trained drone pilots who wear virtual-reality goggles to see from the drone’s point of view while aiming for enemy targets. There are also larger quadcopter or multirotor “bomber” drones that can carry heavier payloads for either supply runs or for dropping explosives on enemy targets at the front lines.
Longer-range strike drones that resemble fixed-wing aircraft may incorporate more autonomous decision-making capabilities. In 2025, Russia launched hundreds of drones to attack Ukrainian cities each night, including Shahed drones originally provided by Iran and increasingly manufactured in Russia. Shahed drones are typically preprogrammed to fly automatically toward their targets with little autonomous decision-making capability. But some Shahed drone variants, such as the Geran-2, are equipped with smuggled Nvidia Jetson Orin microcomputers that provide onboard video processing and autonomous decision-making capabilities, including autonomous target recognition and retargeting.
To defend against those attacks by Russian Shahed drones, Ukraine has deployed low- and high-tech air defense systems that include homegrown, inexpensive interceptor drones. Some interceptor drone systems are designed to autonomously fly to the intercept point and lock onto targets—although a human operator is still required to perform initial target selection and initiate strike commands while always retaining the ability to cancel attacks, according to the Ukrainian government platform United 24.
Meanwhile, Ukraine has gained the technological and manufacturing capability to launch more than 5,000 drone strikes against Russian targets at ranges exceeding 20 kilometers every month, according to the Ukrainian Ministry of Defense. These mid- and long-range strike drones necessarily rely heavily on autonomous navigation capabilities because of Russian electronic warfare systems that can interfere with human operator communication links, along with GPS jamming that can throw off GPS-guided weapons. Such AI-driven navigation has boosted the success rate of Ukrainian drone strikes from around 10 to 20 percent to 70 to 80 percent, Bondar wrote in her CSIS report.
Overall, Ukraine’s defense industry has focused on training small AI models on small datasets to run on the limited computing power of small and inexpensive chips, Bondar wrote. That approach has created AI-driven software for crucial autonomous functions—including navigation and target recognition—that can be packaged with standalone hardware modules for installation on small first-person view (FPV) drones, long-range strike drones, or even the gun turrets of uncrewed ground robots. So even if fully autonomous systems remain rare, expect more drones and robots to gain specific autonomous AI capabilities that complement human decision-making on the battlefield.
This story was updated on June 12 to add more details on Ukraine’s current policy against using such fully autonomous drones, and to clarify that Kokhanovskyy’s company Aero Center was not involved in the one-time battlefield test of the technology.