Can V2I maneuver coordination improve traffic efficiency at intersections?

A controller replaces the traffic light

The paper studies a single-lane, four-way intersection where only CAVs circulate. The area is covered by roadside or cellular infrastructure and served by a multi-access edge computing (MEC) server, which hosts an intersection controller with low end-to-end delay. Vehicles talk to that controller over V2I using maneuver-coordination messages, reserving pieces of the road through the concept of a target road resource.

The geometry matters for the scheduling logic. A control zone extends 100 m back from the junction, and the junction itself is a conflict zone split into four subzones with 7.2 m sides. A right turn crosses one subzone, going straight crosses two, and a left turn crosses three. Only one vehicle may occupy a subzone at a time, though a car occupies two subzones while moving between them. The controller's job is to keep those subzones conflict-free while letting cars keep moving.

Why FIFS beats a plain FIFO queue

Each vehicle entering the control zone computes its own fastest trajectory to the junction, assuming an empty road, and sends the resulting entry and exit times for each subzone to the controller as a proposal. The controller checks for collisions before the conflict zone and inside it, postponing the proposal to the first acceptable time when conflicts appear, then returns a prescription. If the vehicle can follow it, it confirms and fixes its trajectory once.

The key difference from the FIFO benchmark is when a vehicle is allowed to claim space. Under FIFO, a vehicle may enter subzones only after the preceding vehicles have been scheduled, which the authors say produces high travel times even with ideal communication. FIFS instead schedules each arriving vehicle against the existing timetable, so the queue does not serialize the whole approach. The paper also stresses that each car defines its trajectory only once, avoiding repeated trajectory changes that would complicate a real protocol.

Backup mode is the congestion escape hatch

The design assumes cars will not stop, but it does not assume that always holds. A vehicle enters backup mode if its distance to the conflict area drops below a safety threshold, or if the prescribed speed profile falls below 3 m/s. In backup mode it ignores the controller and drives on its own sensors and normal road rules, then notifies the controller after crossing. The controller waits for all backup-mode vehicles to clear before re-initializing.

This mechanism explains the paper's main caveat. Backup mode is described as a last resort for heavy congestion, and it is exactly what erodes performance at the highest density tested. The protocol is also deliberately light: communication involves one vehicle at a time, a scheduled vehicle needs no further exchanges, and the number of message exchanges grows only linearly with the number of vehicles at the intersection.

SUMO simulations with realistic V2I delay

Results come from the SUMO traffic simulator driven by a Python script over the Traffic Control Interface, with each run lasting 1000 s. Four roads of one lane each feed the intersection, vehicles arrive as a Poisson process at equal rates from every direction, and each turns left, right or straight with uniform probability. Maximum speed is 50 km/h, maximum turning speed 20 km/h, acceleration is capped at 2.6 m/s² and braking at -4.5 m/s².

Communication is modeled as a uniformly random delay of 20 to 100 ms per message in either direction, which the authors call pessimistic for 5G cellular systems with a MEC. The comparisons are a priority case where horizontal roads have precedence, a traffic light with 35 s green and 3 s yellow phases, FIFO with ideal communication, and FIFS with ideal communication. The two metrics are travel time in seconds from simulation entry to exit, and CO2 emissions during that travel time under SUMO defaults.

FIFS wins until the heaviest load

At the lightest load, 0.05 vehicles/s per direction, the traffic light is the worst option because it forces unnecessary stops and waiting. Priority rules do better, though roughly 20% of vehicles are delayed when they arrive from the road that must yield. FIFO and FIFS both avoid stopping and perform best, and at this density they are very similar, with the communication protocol having negligible effect.

At 0.15 vehicles/s per direction, the ordering shifts. The traffic light now beats priority rules because it prevents a subset of vehicles from waiting for long intervals. FIFO shows its limits even with ideal communication, while FIFS keeps travel time per vehicle below 50 s and CO2 per vehicle below 200 g, and this holds with realistic V2I delays. Across densities from 0.05 to 0.20 vehicles/s per direction, FIFS with ideal communication is optimal at every density in both average and worst cases, and FIFS with the real protocol still clearly beats priority, traffic lights and FIFO. Only at 0.20 vehicles/s per direction does a large share of vehicles enter backup mode and performance become non-optimal.

Single-lane CAV-only scope limits transfer

The validated scenario is narrow in ways that matter for anyone extrapolating. It is a classic single-lane, four-way intersection with only CAVs, no human-driven vehicles and no vulnerable road users such as pedestrians or cyclists. The authors state that extending the algorithm to other intersection types, roundabouts, human-driven vehicles and non-connected objects is left to future work, and they suggest cameras as a possible first approach for detecting non-connected objects and VRUs.

The architecture is also centralized: vehicles coordinate only through the V2I controller, and adaptation to a distributed vehicle-to-vehicle setting plus a field test campaign are named as future activities. The performance numbers therefore describe simulated CAV-only traffic under a modeled delay distribution, not measured road outcomes.

The decision hinges on density, not on the protocol

The paper's most useful comparison is not FIFS versus FIFO in the abstract but where each option sits as density rises. Priority rules are preferable to a traffic light while congestion is moderate, and the reverse holds once traffic is congested; at 0.20 vehicles/s per direction, priority rules leave some vehicles waiting so long that average travel time and emissions rise sharply. FIFS with ideal communication sits above all of them at every tested density.

With the real protocol, the practical reading is that FIFS delivers its clearest advantage in the light-to-moderate range, where it removes stops entirely, and that its edge depends on keeping vehicles out of backup mode. The 0.20 vehicles/s per direction case is the threshold the paper itself identifies as the point where heavy congestion pushes many cars into sensor-based driving and the coordinated schedule stops being optimal. Anyone judging whether such a controller is worth deploying should therefore ask how close their intersection runs to that load, not just whether V2I coordination is available.

Sources used in this answer