Transportation is a critical national infrastructure: We use public roads every day; by moving goods and people, roadways are vital to our national well-being. Safety is critical to transportation, but the United States has long suffered a road safety crisis. According to the US Department of Transportation website: “More than 370,000 people died in transportation incidents over the last decade (2011-2020) in the United States." Connected and autonomous transportation is envisaged as the future of transportation that will help achieve zero fatality on roads. Since the DARPA Grand Challenge in 2005, tremendous progress has been made in the development of connected and autonomous vehicles (CAVs). GM SuperCruise and Teslas with “autopilot” are already roaming on our streets; and pilot “robotaxi” services are now available in several major US cities.

Today’s CAVs can, at best, be rated as SAE Level-4: Namely, these vehicles are designed with a specific set of conditions, referred to as its Operational Design Domain (ODD), outside which a CAV must come to a safe stop. Level-5 or fully autonomous driving still has a long way to go, despite the rapid advances in artificial intelligence and machine learning (AI/ML) in recent years. To partly circumvent these challenges, (partial) remote driving has been proposed as an alternative or complementary approach, where a remote human operator takes over the operation of a CAV when needed, e.g., before a CAV is about to encounter a situation outside its ODD and has to stop or after such an event. The potential of remote driving– or teleoperated driving (ToD) in the 3GPP parlance– is inspired by the promise of 5G and is considered one of the key use cases for 5G. Remote driving has been tested in mostly ideal and restricted environments; several start-up companies are promoting remote driving for certain use cases.

Can 5G really fulfill its promise and meet the requirements of remote driving? Two key latency bottlenecks in current 5G networks that pose unique challenges to remote driving are: a) highly fluctuating radio channel conditions induced by high mobility of vehicles, thereby causing longer delays due to high block error rates (BLERs) and retransmissions; b) frequent handovers (HOs) further induced by vehicle mobility, which can cause prolonged bursts of high packet losses, thereby further worsening end-to-end latency performance, as our measurement studies and remote driving experiments over 5G networks have shown. All these can have cascading cumulative effects, severely degrading the end-to-end performance of remote driving.

In this Ideas Lab Breaking Low proposal– nicknamed DRIVE-SAFE, we target remote– and cooperative driving as the vertical use case to break the low latency barrier in wireless cellular networks from cross-layer and end-to-end vertical application perspectives. We focus, in particular, on tackling the above two latency bottlenecks in today’s 5G networks. We advance two major innovations: i) We put forth a novel vertical-driven, mobility-aware framework with innovative proactive mechanisms (e.g., trajectory-aware HOs) to reduce the impacts of high mobility and handovers on the tail latency performance of the target vertical application. ii) We seamlessly integrate C-V2X technologies with 5G networks by advancing situation-based V2V communications for cooperative situation awareness. By incorporating cooperative driving to remote driving, we not only circumvent the challenge in egocentric (remote) driving (“ego-driving”)– despite being equipped with a full suite of sensors, an ego-vehicle can’t see everything all the time– but also effectively leverage the advantages of 5G and V2V technologies while overcoming their respective shortcomings. These ideas will be expanded on in the following sections.

Clearly, tackling all the low latency challenges in enabling remote and cooperative driving over 5G (and next generation (NextG)) networks require enormous efforts that go far beyond this two-year project. To cope with this complexity, we take a phased approach by narrowing down the scope and focusing on those critical problems that pose the most significant barrier to achieving low latency for the target vertical use case. Based on their Technology Readiness Levels (TRLs), we organize our research tasks in two phases with minimum viable and optional components. Such a phased approach will enable us to start integrating, testing, and evaluating the proposed (Phase 1) solutions using the NSF POWDER testbed earlier, while we work on Phase 2 solutions in parallel. Toward the end of the two-year project, we will conduct pilot demonstrations using the Mcity test-track facilities. We are confident that this phased approach will ensure a high chance of successfully executing this high-risk and yet high-reward project, truly moving the needle with transformative impacts.

The project represents close collaborations among three academic institutions – University of Minnesota (UMN), University of Utah (UoU), and University of California, Riverside (UCR) – and three industrial organizations – General Motors, a vehicle OEM (original equipment vendor), Nokia Bell Labs, part of a network equipment vendor, and AT&T, a mobile network operator (MNO). With rich experience, diverse expertise, and the resources needed, the team is well-placed to carry out this project successfully. As argued in the proposal, we believe that the timing is right; we have laid out a 5-year roadmap to amplify the impacts of this project. 
NSF Awards: 2453815, 2453816, and 2453817.