When the GPS got malfunctioning and missed the exit on the highway, we all went through that moment of frustration. The Tern AI team, building a low-cost GPS alternative, says it's because current technology is limited by its reliance on satellite positioning.
Tern AI says it found a way to find the vehicle's location using only map information and existing sensor data from the vehicle. The company's pitch: It's a cheap system that doesn't require any additional expensive sensors.
At SXSW, Austin-based startups have demonstrated that they can “derive positions from Nothing” exclusively for TechCrunch.
“There's no triangulation, satellites, wi-fi, nothing. Co-founder and president Brett Harrison told TechCrunch that Cyrus Behroozi, senior software developer at Tern, loaded a demo for the iPhone. “As we move away from the technology-limiting triangulation base, it really changes the game as we have the ability to completely remove that grid.”
Harrison says this breakthrough is important for several reasons. From a commercial standpoint, companies relying on GPS, including GPS, will reach the delivery company, but with flaws in GPS positioning, they lose time, money and gas every time the driver needs to double back.
More importantly, from aviation to disaster response to precision agriculture, our most important systems rely on GPS. Foreign enemies have already demonstrated that they can cause GPS signals, which could have devastating effects on both the economy and national security.
The US shows it wants to prioritize GPS alternatives. During his first term, President Donald Trump signed an executive order to reduce his reliance on a single PNT source (positioning, navigation, timing) service, such as GPS. There are also several other initiatives directing agencies and agencies such as the Department of Defense and the National Security Council to ensure resilient PNTs by testing and integrating non-GPS technologies.
“Deepseek came out and said it would cost USD 6 million to do what it cost. [OpenAI, Anthropic, and other AI companies] We do billions of things,” Harrison said. “It took the government billions and decades to move that dot across the map in real time in a vehicle. It was done for less than $2 million.”
Tern came out of Stealth in February 2024 and announced a $4.4 million seed round a few months later. This is a quick turnaround to achieve the type of positioning we experienced at SXSW this week.
Testing Austin Tern System
To begin the demonstration, Behroozi connected the 2019 Honda Civic to the mobile phone via Bluetooth, allowing the Tern application to draw data from existing sensors in the vehicle. He noted that Tern technology can be integrated directly into vehicles since 2009 and after the year of the model.
Normally Tern would manually set positions to speed things up, but for the demo, the team wanted a “cold start.” Behroozi turned off the location service on the phone, so the tern's intelligent system only had a cache map of 500 square miles around Austin and vehicle sensors.
As the car drove, the system took up road data and worked towards “convergence.” According to Behroozi, it took about 10 minutes for the system to reach full convergence from a cold start, due to our limited movement. Harrison usually ensures that convergence takes about 1-2 minutes without a starting point, and it takes just 1 minute.
Harrison said the tern system can also localize vehicles to parking garages, tunnels and mountains where GPS is struggling. Harrison didn't explain exactly how the information was “unique.”
After the system reached full convergence, we drove for a few more minutes and I saw it steadily track the precise movements in a way that seemed as good as GPS. It became more clear when we ran to downtown Austin. There, my Google regularly misunderstood me throughout the week as I navigated through urban streets dotted with towering buildings.
Harrison said “Tern's system is also safe from a privacy standpoint, as it uses GPS, as “if someone knows your ID, you can always find you.”
“Our system is a totally closed loop,” he said. “Now we are not releasing anything. It is independently deriving its own unique position. [via on edge computing]so there are no external touchpoints. ”
Built for scaling
“We set the company and solution to be scalable from the start. Looking at that Waymo car and all the embedded hardware, we don't know we'll be heading to Nissan Sentra anytime soon. Harrison said, pointing to Robotakshi on Waymober in front of us.
“At the manufacturer level, if [Tern] It is implemented within an infotainment system and is simply a software download, making it extremely scalable. Every new vehicle has the required sensor data. Map data is already present in all providers. It's very easy. ”
Tern's potential customers could be anyone from Google to Uber, from automakers to mobile phone makers. Harrison said the startup is open to the growth of the company as well as acquisitions.
“The main thing is to put this into the hands of the economy. There is the emergence of technologies that have not achieved their full potential due to the growth of threats and the limitations of triangulation,” Harrison said.
He noted that Terns are exploring possibilities with the government. The startup recently received a contract award from the US Department of Transportation after demonstrating its technology along with nine other companies around the world.
“We want to show the government what is possible with American innovation right now,” Harrison said.
Clarification: This article has been updated to clarify the periods that the terns claim normally takes to fully converge.