Zephr has emerged from stealth with radical plans to dramatically improve GPS — using cell phones with zero hardware modifications.
Mobile devices have abysmal GPS accuracy, which prevents location app developers in verticals spanning gaming to fitness tracking from truly taking advantage of the powerful little computer everyone carries in their pocket. Two of Zephr’s four founders, Sean Gorman and Pramukta Rao, felt this problem acutely during a previous venture, when they were tasked by a customer to build a visual positioning system for a company campus.
To do so, Gorman and Rao set up a highly accurate differential GPS receiver known as a base station. These have huge antennas and can connect to a network of other base stations to provide what’s called an error correction. Using that error correction (and some differential calculations), one can get up to 1-2 centimeter-level accuracy. But the pair realized that they were surrounded by tons of raw GPS data — from people’s cell phones.
“Why can’t we just turn everybody’s phone into a base station, and use those raw satellite measurements to try to figure out an error correction?” Gorman, who is Zephr’s CEO, said in a recent interview. “Instead of having one super precise signal, what if you had a whole bunch of mediocre signals, and then we could use some ensemble optimizations to get that converge on reality?”
The answer, of course, is that the GPS accuracy on cell phones just isn’t good enough. The issues with GPS are especially acute in urban areas, where GPS signals often bounce off buildings or suffer from line-of-sight issues.
Zephr says it has found a solution. Gorman and Rao teamed up with Kostas Stamatiou and Scotty Nelson to start the company in September last year. The founding team have major bona fides in the mapping world: Among other achievements, Gorman was Snap’s former mapping engineering manager; Rao was a computer vision engineer at Snap and head of engineering at Violet Labs; Stamatiou co-founded greenhouse gas emissions analytics company BlueSky Resources; and Nelson was a senior data scientist at Twitter, now X.
The startup’s solution works by directing a group of phones in a given area to ping their GPS measurements to satellites, which then uses a software server to calculate an error correction similar to the one generated by receiver base stations. That correction is sent back to the phones and is used to improve the device’s GPS.
“You can crowdsource the measurements across a bunch of phones to get a better version of reality by looking at more satellites and getting more measurements,” Gorman said. “But then you can also do that anonymously, because we’re just looking at the measurements and then we send the error correction, so we don’t know the location of the device but the device gets the correction to fix it.”
The company says it doesn’t even need very many phones to improve GPS accuracy on each mobile device: just 10-15 mobiles within 10 kilometers of each other.
Their plan has caught investor attention, with the company landing a $3.5 million seed round led by Space Capital and First Spark Ventures.
Zephr validated its concept with help from Silicon Valley research giant SRI International, which operates a lab focused on positioning, navigation, timing (PNT) and GPS, where engineers focus on sophisticated simulations, modeling and benchmarking. Testing at SRI validated Zephr’s concept — and provided the startup with an investment from the research firm’s venture arm, SRI Ventures.
The team, which stands at six full-time and two part-time, has since been doing field testing using real-world cell phones and differential RTK-GPS systems as ground truth. It’s now ready to start doing pilots using a demo app that can collect data and compare the accuracy of the standard GPS on the phone to the improved GPS. Zephr is also planning on customizing its models for customers (which is pretty much anyone that is developing a mobile app that uses GPS).
Gorman says Zephr has seen a lot of interest from rideshare, location-based gaming and advertisement technology verticals. But the team, which Gorman hopes to grow to 10 full-time by the end of this year, has its sights set on longer-term markets that are still nascent and evolving.
For example, autonomy, robotics and augmented reality all require highly precise, inexpensive positioning measurements, Gorman said.
“We think this new form of positioning could go a long ways to augmenting or replacing visual positioning systems, which are incredibly expensive,” he said. “Google and Apple have to map the whole world in 3D to get the visual positioning to work. It’s just really expensive to have a fleet of street view vehicles and planes to capture the aerial views. Potentially if you can do this just with sensors, you don’t have to map it ahead of time, and you don’t have to invest so much in compute and data collection to get some of these things to run in the future.”