Surrey's AI navigates dense cities without GPS
Researchers at the University of Surrey just developed PEnG, an AI navigation system that bypasses GPS satellites entirely. By pairing satellite maps with feeds from a simple monocular camera, it slashed urban positioning errors from 734 meters down to just 22 meters. Why it matters: it offers a reliable, low-cost fix for autonomous cars and delivery robots navigating tunnels and high-rise concrete jungles.
PEnG (Pose-Enhanced Geo-Localisation) is an open-source visual navigation pipeline designed to fix position tracking in dense urban areas without relying on GPS satellites. By cross-referencing top-down satellite imagery with ground-level camera feeds, it works out both where a machine is and which way it is pointing.
How it navigates
The system runs on a two-step localisation pipeline. It first narrows down the rough location using street-level feature matching, then applies relative pose estimation to calculate the camera's precise orientation and angle.
Its main technical advantage is low hardware overhead. Instead of demanding expensive multi-beam LiDAR arrays or specialized multi-sensor rigs, PEnG runs off a simple monocular camera, the exact type already installed in most road cars and low-cost delivery bots.
In urban benchmark tests where tall buildings typically shatter satellite reception, baseline visual drift reached up to 734 meters. PEnG slashed that error down to just 22 meters, an almost 40-fold accuracy improvement.
Code availability and roadmap
Developed by researchers at the University of Surrey, the underlying research has been published in IEEE Robotics and Automation Letters.
The project team has made the research open source so roboticists and developers can build on their findings. They are now working on a hardware prototype designed for continuous testing in live traffic environments.
Why it matters
For developers building delivery robots, drones, or autonomous platforms, PEnG shows how far visual data alone can go when satellite signals get jammed or obstructed by skyscrapers. Achieving a 22-meter accuracy window without LiDAR or RTK ground stations lowers the barrier to building resilient navigation.
Still, a 22-meter margin is not sharp enough for autonomous cars needing lane-level precision. We see it working best as a failsafe layer: paired with wheel odometry and IMU sensors, it gives autonomous robots a cheap, dependable backup whenever the sky gets blocked.
Sources (1)
- interestingengineering.com(opens in a new tab)Originalinterestingengineering.com