Niantic Spatial & Vantor: Pokémon Go scans train drone navigation
Voluntary AR scans from Pokémon Go players trained Niantic Spatial models now paired with Vantor to guide drones when GPS is denied.
TL;DR
- 01Voluntary AR scans from Pokémon Go players trained Niantic Spatial models now paired with Vantor to guide drones when GPS is denied.
- 02The partnership was announced in December 2025, and early tests of the combined system showed error reduction of up to 70 percent and accuracy of about 1.5 meters.
- 03In a move that links consumer AR mapping to defense applications, Niantic Spatial is pairing its ground-level Visual Positioning System with Vantor’s Raptor software and Vantor’s 3D terrain data.
Niantic Spatial has combined a foundation model trained on voluntary Pokémon Go player scans with Vantor’s spatial software to help drones and other autonomous systems navigate when satellite signals are unavailable. The partnership was announced in December 2025, and early tests of the combined system showed error reduction of up to 70 percent and accuracy of about 1.5 meters.
What happened
In a move that links consumer AR mapping to defense applications, Niantic Spatial is pairing its ground-level Visual Positioning System with Vantor’s Raptor software and Vantor’s 3D terrain data. The two companies say the combo produces a shared coordinate system for drones, vehicles, and AR headsets that can operate when GPS is jammed, spoofed, or otherwise denied. Niantic’s blog post cited early test results showing error reduction of up to 70 percent and accuracy around 1.5 meters, and described the visual system as immune to standard signal jammers.
The raw material for Niantic Spatial’s foundation models traces back to a 2021 Pokémon Go update, which added in-game incentives for players to scan real-world locations using their smartphones. Millions of players opted in and generated, according to DroneXL, billions of visual mapping data points from streets, buildings, parks, and trees. Niantic and Vantor told Guardian Australia that the ground-level scans were not handed directly to Vantor; instead the scans were used to train Niantic’s foundation models. Niantic has said the scans were submitted voluntarily and covered by the privacy and terms-of-service policies in place at the time.
Niantic split its gaming business from its geospatial AI business in March 2025, when Scopely acquired the games side for $3.5 billion, leaving Niantic Spatial as a standalone company focused on spatial AI models. Separately, Vantor won a February 2026 US Army contract worth up to $217 million for the One World Terrain program, which focuses on high-precision 3D terrain data for the Army’s Synthetic Training Environment. There is no public evidence that Pokémon Go scans will be part of that specific contract.
How the technology is combined
Niantic Spatial’s visual system, built from ground-level AR scans and foundation models, provides a camera-based Visual Positioning System that determines location without GPS. Vantor contributes Raptor software and its 3D terrain data, compiled from over two decades of satellite imagery. Together the components create a shared coordinate system that aligns data for drones, vehicles, and AR headsets, enabling navigation and positioning when satellite signals are compromised. Early integrated testing produced the error reductions and meter-level accuracy cited by Niantic Spatial.
Both companies framed the integration around GPS denial scenarios; the partnership specifically targets GPS denial, spoofing, interference, and jamming. The overall system is presented as resistant to standard signal jammers because it relies on ground-level visual cues rather than satellite navigation signals.
Why it matters
This work ties large-scale, voluntarily collected consumer AR data into systems that can support military and defense applications. The involvement of Pokémon Go scans in training foundation models that underpin a GPS-free navigation system raises questions about the boundary between consumer-facing data collection and downstream defense uses, even where companies say raw scans were not handed to a contractor. The partnership also responds to a real operational problem: GPS jamming and spoofing are already being used in conflicts in Ukraine and Iran to disrupt drones and guided systems.
What to watch
Watch for any disclosures or procurement documents showing whether Niantic-origin scans are used under Vantor’s One World Terrain work, and for public results from larger field tests validating sustained meter-level accuracy and resistance to jamming. Also monitor company statements about data governance and consent around AR scans as Niantic Spatial pursues more defense and enterprise partnerships.
Written by The Brieftide · Source: The Decoder
The Brieftide Daily · 06:00
Briefs like this one, in your inbox every morning.
Continue reading
More in AI SafetyAnthropic essay: Dario Amodei's Cold War playbook for AI
Anthropic published a sweeping essay plus two policy frameworks calling for binding audits of frontier models and a strategic national.
Germany approves DE-AISI to test Anthropic frontier models
Germany's National Security Council greenlit DE-AISI, modeled on the UK's AISI, to evaluate Anthropic frontier models and national security
DeepMind $10M fund for multi-agent AI safety research
DeepMind and partner organisations have opened a $10 million funding call to support research into multi-agent coordination.
OpenAI shifts automation policy: no full automation by 2028
OpenAI says entirely automating everything isn't the future and emphasizes a human-machine tandem plus new safeguards.