在《宝可梦GO》全球热潮达到顶峰十年后,一家人工智能公司正利用数百万玩家拍摄的数十亿张真实世界图像,为送货机器人乃至可能的军用无人机开发导航技术。这代表着这款增强现实手机游戏留下了一份既引人入胜又可能令人不安的遗产——该游戏一直激励玩家用智能手机拍摄社区街道和地标的短视频。
这家名为Niantic Spatial的人工智能公司于2025年5月从《宝可梦GO》游戏开发商Niantic分拆出来,此前Niantic已将其《宝可梦GO》等授权游戏单独出售给沙特支持的视频游戏发行商Scopely。但在这笔交易之前,Niantic曾公开宣布计划利用数百万《宝可梦GO》玩家的扫描数据,以及该公司Scaniverse应用用户捕获的数据,来训练和开发一个“大型地理空间模型”——这是一个基于应用用户扫描真实地点所提供的地理定位图像训练而成的物理世界3D模型。
“地面扫描是帮助训练Niantic Spatial真实世界基础模型的组成部分之一——这些AI系统学习识别和解读物理空间,”一位Niantic Spatial发言人告诉Ars。“这些模型是训练过程的产物,而非底层扫描数据的副本或访问途径,这些扫描数据针对的是雕像、喷泉等公共兴趣点。”
据《麻省理工科技评论》报道,在Niantic Spatial作为独立公司分拆后,其模型在300亿张图像上进行了训练,这些图像大多集中在游戏玩家被激励前往的城市环境地点。这些图像通常从许多不同角度、在不同光照和天气条件下捕捉同一地点,并附有宝贵的元数据,显示了用户拍摄这些图像时手机的位置和朝向。
此类地面扫描“是游戏中的一项完全可选功能,用户可借此为公共地点创建一段短视频,”Niantic Spatial发言人表示。“自2019年以来,我们一直在隐私政策和公开声明中透明地说明,这些扫描将用于改进我们的技术平台。”
这使得 Niantic Spatial 能够开发出自己的视觉定位系统——一种通过将摄像头采集的视觉数据与详细环境三维地图中的参考数据进行比对,从而确定设备位置和朝向的技术。这类系统在室内、GPS 及其他全球导航卫星系统信号不可靠的城市环境,或存在主动 GPS 干扰的区域中尤其有用。
《麻省理工科技评论》在 2026 年 3 月重点报道了 Niantic Spatial 的技术,当时该公司宣布与 Coco Robotics 建立新的合作伙伴关系。这家机器人公司旨在利用 Niantic Spatial 的 AI 模型和视觉定位系统,帮助其四轮配送机器人车队在城市街道上导航。
但在 2025 年 12 月,Niantic Spatial 还宣布与空间智能公司 Vantor 达成一项协议,共同开发一种定位系统,帮助飞行无人机和地面车辆在无 GPS 环境中导航。Vantor 前身为航天与卫星公司 Maxar Intelligence,与美国国家地理空间情报局、美军多个分支以及国土安全部签有多项政府合同。
新的军工复合体
这套“综合定位系统”旨在将 Niantic Spatial 的视觉定位系统与 Vantor 的三维地形数据及 Raptor 软件整合在一起。在 2026 年 2 月于伦敦举行的国防地理空间情报(DGI)会议上,Niantic Spatial 的产品管理总监 Tory Smith 表示,该集成系统的早期测试结果显示,在许多场景下定位误差减少了 70%,精度可达 1.5 米以内。
Niantic Spatial 与 Vantor 之间的合作,因荷兰新闻媒体 Trouw 近日的一篇报道而获得了更多公众关注。“如果没有来自所有那些游戏玩家的大量扫描数据,这个系统的发展绝不可能如此迅速,”荷兰代尔夫特理工大学伦理与技术教授 Jeroen van den Hoven 在接受 Trouw 采访时表示。“玩家们间接地,或许是以一种微小但有效的方式,为军事应用做出了贡献。”
视觉定位系统并不一定充满伦理问题,即使在军事场景下也是如此。例如,在持续的俄乌战争中,乌克兰军方一直在部署配备自有视觉定位系统的战场机器人和无人机,以应对普遍存在的 GPS 干扰。“如果乌克兰人能够借此赢得对抗侵略者俄罗斯的正义战争,那将是一个好的发展,”Van den Hoven 告诉 Trouw。
但这家荷兰报纸还采访了长期玩 Pokémon Go 的玩家 Floris De Hingh,他对自己游戏数据被用于支持美国军事系统表示担忧。De Hingh 特别将自己描述为“强烈反对特朗普目前对伊朗发动的战争”。
“训练数据来自那些以为自己是在抓皮卡丘的人,他们根据一份大多数人从未读过的许可协议,将数据出售给了一条最终通向主权财富基金和国防巨头的供应链,”新闻网站 DroneXL 的创始人兼主编 Haye Kesteloo 写道。“为游戏获得的同意,并不等同于为武器计划获得的同意,即使最终用途被证明是合理的。”
Vantor 的一位发言人告诉 Ars,该公司“没有使用任何 Pokémon Go 数据,也无法访问 Pokémon Go 数据集中的任何信息。”同样,Niantic Spatial 的发言人表示,两家公司之间的协议不包括直接共享游戏数据。
但一些像 De Hingh 这样的《宝可梦 Go》玩家,可能会对自己游戏数据最初竟被用于训练 Niantic Spatial 的模型而感到不安——尤其是当该公司的视觉定位系统可能被用于军事用途时。Vantor 承认,他们正在“探索适配 Niantic Spatial 的地面视觉定位系统”,以便与 Vantor 现有的“无 GPS 环境下的定位能力”协同工作,而后者目前依赖卫星图像。
A decade after the global craze for Pokémon Go peaked, an AI company has been using billions of real-world images captured by millions of players to develop navigation technologies for delivery robots and possibly military drones. That represents an intriguing but potentially discomfiting legacy for an augmented reality mobile game that has incentivized gamers to capture short smartphone videos of physical neighborhoods and landmarks.
The AI company, Niantic Spatial, was spun out of Pokémon Go game developer Niantic in May 2025, after Niantic separately sold its licensed games such as Pokémon Go to the Saudi-backed video game publisher Scopely. But before that deal, Niantic publicly announced plans to use scans from millions of Pokémon Go players along with data captured by users of the company’s Scaniverse app to train and develop a “large geospatial model”—a 3D model of the physical world trained on the geolocated images provided by app users scanning real-world locations.
“Ground scans were one component to help train Niantic Spatial’s real-world foundation models —AI systems that learn to recognize and interpret physical spaces,” a Niantic Spatial spokesperson told Ars. “The models are the product of that training, not a copy of or a means of accessing the underlying scans, which were of public points of interest such as statues and fountains.”
After Niantic Spatial spun out as a standalone company, it trained its model on 30 billion images mostly clustered around urban environment locations that game players were incentivized to visit, according to MIT Technology Review. The images often captured the same location from many different angles under different lighting and weather conditions, and came with valuable metadata showing the location and orientation of user phones when they were capturing such images.
Such ground scans “were an entirely optional feature in games, where users created a short video of a public location,” the Niantic Spatial spokesperson said. “We’ve been transparent about the fact that the scans would improve our technology platform since 2019 in our privacy policy and public announcements.”
That allowed Niantic Spatial to develop its own visual positioning system—a type of technology that can provide a device’s position and orientation by comparing visual data from cameras with reference data from detailed 3D maps of environments. Such a system can be especially helpful indoors, in city environments where GPS and other global navigation satellite systems’ signals are unreliable, or in regions where there is active GPS jamming.
MIT Technology Review highlighted Niantic Spatial’s technology in March 2026, when the company announced a new partnership with Coco Robotics. The robotics company aimed to use Niantic Spatial’s AI model and visual positioning system to help its fleet of four-wheeled delivery robots navigate city streets.
But in December 2025, Niantic Spatial had also announced a deal with the spatial intelligence company Vantor to develop a positioning system that could help both flying drones and ground vehicles navigate GPS-denied environments. Vantor, formerly known as the space and satellite company Maxar Intelligence, has multiple US government contracts with the National Geospace-Intelligence Agency, various branches of the US military, and the Department of Homeland Security.
The new military-industrial complex
The “comprehensive positioning system” aimed to integrate Niantic Spatial’s visual positioning system with Vantor’s 3D terrain data and Raptor software. During the Defence Geospatial Intelligence (DGI) conference held in London in February 2026, Tory Smith, director of product management at Niantic Spatial, described early testing of the integrated system as leading to a 70 percent reduction in positioning error with accuracy to within 1.5 meters in many scenarios.
The partnership between Niantic Spatial and Vantor received more public attention through a recent story by Trouw, a Dutch news publication. “Without the large number of scans from all those gamers, the development of this system would never have progressed so quickly,” said Jeroen van den Hoven, professor of ethics and technology at Delft University of Technology in the Netherlands, in an interview with Trouw. “The players have indirectly, in a perhaps minimal but still effective way, made a contribution to military applications.”
Visual positioning systems are not necessarily fraught with ethical problems, even in a military scenario. For example, the Ukrainian military has been deploying battlefield robots and drones with their own visual positioning systems to survive the prevalence of GPS jamming in the ongoing Russo-Ukrainian war. “If the Ukrainians can win the just war against aggressor Russia with this, it is a good development,” Van den Hoven told Trouw.
But the Dutch newspaper also interviewed Floris De Hingh, a longtime Pokémon Go player who expressed concern about his gameplay data supporting US military systems. De Hingh specifically described himself as “strongly opposed to the war Trump is currently waging against Iran.”
“The training data came from people who thought they were catching Pikachu, under a license most never read, sold up a chain that ends at a sovereign wealth fund and a defense prime,” wrote Haye Kesteloo, editor in chief and founder of the news website DroneXL. “Consent obtained for a game is not consent for a weapons program, even if the end use turns out to be defensible.”
A Vantor spokesperson told Ars that the company “is not using any Pokémon Go data, nor do we have access to any information from the Pokémon Go dataset.” Similarly, Niantic Spatial’s spokesperson said that the agreement between the companies does not include direct sharing of game data.
But some Pokémon Go players, such as De Hingh, will probably be uncomfortable with the idea that their gameplay data helped train Niantic Spatial’s models in the first place—especially when the company’s visual positioning system may be used for military applications. Vantor acknowledged that it is “exploring adapting Niantic Spatial’s ground-based visual positioning system” to work alongside Vantor’s existing “GPS-denied positioning capabilities,” which currently rely on satellite imagery.