Editor’s note: This article was first published by the University of Utah's John and Marcia Price School of Engineering and is adapted here with permission. Zhi-Li Zhang, UMN professor of computer science and engineering and a CTS scholar, serves as the project's principal investigator.
Self-driving cars and “robotaxis” need detailed mappings of the real-world environments they operate in and require massive data for extensive training and testing before they can be deployed. Even with this data, and despite their increasing sophistication, these vehicles can can get themselves into tricky situations that require a human driver to fix. This can present compounding traffic problems if there’s no qualified driver near the vehicle. But with all of the sensors and connectivity in these vehicles, why can’t a remote driver take control of a vehicle from afar?
The issue is latency — the time it takes for information to get back and forth between driver and vehicle. Anything less than instantaneous feedback is prohibitively dangerous when operating a two-ton machine, and current cellular networks are not up to the task.
Researchers from the University of Minnesota’s College of Science and Engineering, the University of Utah’s John and Marcia Price College of Engineering, and the University of California, Riverside, in collaboration with researchers from Nokia Bell Labs and GM Research and Development, are working on how remote driving and self-driving can complement each other. By solving the latency problem, remote drivers at teleoperation centers up to hundreds of miles away could take over in situations in which self-driving vehicles get stuck, lost, or confused. They could also operate vehicles in remote areas that lack the detailed map data self-driving modes rely on.
In July 2026, the team members demonstrated their work in a real-world setting using Minnesota's custom CAV (connected and automated vehicle) on Utah’s POWDER mobile and wireless test bed, equipped with the network technologies and end-to-end teleoperation application the team developed.
Their efforts are part of a U.S. National Science Foundation program called Breaking the Low Latency Barrier for Verticals in Next-G Wireless Networks (NSF 24-545). The researchers from Minnesota, Utah, California, Nokia, and GM are part of an NSF “Breaking Low” initiative, funded by the NSF Directorate for Technology, Innovation and Partnerships (NSF TIP), and aptly called DRIVE-SAFE.
"While there were a lot of great expectations initially regarding 5G enabling a range of emerging vertical industries, many of them did not take off as hoped, partly due to the perception that 5G technology was really not up to meeting their performance requirements,” says Sudharman K. Jayaweera, program director, Emerging Technologies Section, NSF TIP. “The NSF Breaking Low program was motivated by the need for 5G/NextG technology solutions that can verifiably demonstrate specific low-latency performance targets required by a particular wireless vertical application,” he says. “Remote and cooperative driving is one such use-case for which reliable low-latency connectivity is absolutely critical for it to be viable for mass adoption. DRIVE-SAFE is an industry and academia collaboration where wireless networking experts work side-by-side with autonomous vehicular technology experts to develop, demonstrate, and verify networking technology innovations that meet the specific latency performance needs of remote and cooperative driving under real operating conditions. This project is developing a bold new approach to make NextG networking enable vertical applications, as NSF TIP hoped to achieve from the Breaking Low program."
The current 5G cellular network that our phones connect to was designed largely with individual consumers, not vehicles, in mind. Dropped calls or slow upload speeds are annoyances rather than dealbreakers. The fact that cars travel rapidly exacerbates both of those problems and creates a new one: handoffs. Traveling through overlapping coverage zones from multiple cell towers creates risks for packets of information to be lost.
“So many of these types of projects end up in the ‘valley of death’ — great preliminary findings but no ability to turn them into results,” says project lead Zhi-Li Zhang, Distinguished McKnight University Professor and Qwest Land Grant Chair in Telecommunications at the University of Minnesota. “We’re committed to making teleoperation feasible in the real world.”
One of the key enablers for the DRIVE-SAFE project is the availability of open source cellular network technologies, which allows researchers to explore novel network solutions. One example is the OCUDU project (open centralized unit/distributed unit), the so-called “Linux of RAN (Radio Access Networks),” which is funded by the FutureG Office within the Office of the Under Secretary of Defense for Research and Engineering. Another example is the Duranta OAI RAN 5G open source stack developed and maintained by the OpenAirInterface Software Alliance. These open source stacks foster the creation of real-world end-to-end solutions, like that created by the DRIVE-SAFE team.
“Your car already has a lot of relevant information, like speed and direction, that could help these networks maintain a faster, better connection,” says Kobus Van Der Merwe, the Jay Lepreau Professor in the Kahlert School of Computing and co-principal investigator of the DRIVE-SAFE project. “Ultimately, we need closed-loop networks, where cars and base stations are talking to each other so they can take advantage of this information and make better decisions.”
The demonstration utilizes UMN's connected and automated vehicle (MnCAV), which has played a critical role in advancing research through UMN's MnCAV Ecosystem program. The UMN team modified it to be remote-drivable for this demonstration where the CAV made several laps of a parking lot on the University of Utah campus, all while being piloted from inside a conference room in the Merrill Engineering Building. Rather than using a commercial 5G network to send information back and forth to the vehicle, the teleoperation station connected to the vehicle via an OCUDU-enabled 5G network instantiated in the POWDER testbed and enhanced with network technologies developed by the DRIVE-SAFE team.
This setup allowed the researchers to explore trajectory-driven network optimization techniques to reduce the latency typically associated with a mobile network. One trajectory-driven approach the team developed optimizes the handoff process essential to any mobile network. In essence, interaction between the teleoperation service and the network allows the network to control the handoff process based on the trajectory of the vehicle, thus avoiding unnecessary handoffs that might add latency or other undesirable network effects. This approach is in contrast with the default handoff behavior in today’s 5G networks, in which cellular devices always switch to a stronger connection when one is available. But that behavior can be counterintuitively counterproductive if the device is a moving car: switching multiple times in succession is worse for latency than staying on a single network, even if that network is not the fastest.
Another approach the team developed is adapting the modulation and coding scheme associated with a wireless link based on its trajectory. That capability, developed by Nokia Bell Labs, is called Semantic-Aware RAN Adaptation, or SARA. For example, rather than reacting to transmission errors after they occur — which is too slow for a moving vehicle — SARA uses the vehicle's reported position and speed to adjust the link before conditions degrade; it also protects safety-critical driving commands more strictly than the video feed. In the Utah trials, this technology held the teleoperation session together at the edge of cell coverage, where the unmodified network lost the link entirely.
"The vehicle already knows where it is going and how fast. Our contribution makes that information actionable inside the radio scheduler itself," says Nakjung Choi, project co-principal investigator and department head at Nokia Bell Labs. "The same mechanism generalizes well past driving. Drones, remote-controlled machinery, and industrial automation all need a network that adapts to what the application is doing, and building that in natively is one of the key pillars of upcoming AI-native 6G networks."
The researchers will demonstrate the DRIVE-SAFE capabilities again in October, kicking off the second year of their project. Plans for the second year include working with the other DRIVE-SAFE partner institutions, the University of California, Riverside, and GM to explore the feasibility of combining 5G and vehicle-to-everything capabilities to further enhance teleoperated driving.
