Routing Around the Queue: An Evaluation of Eco-Routing Protocol on different Traffic Scenarios

I. INTRODUCTION

• Road transport emissions reflect idling, queueing, and repeated acceleration, motivating routes informed by current road conditions rather than static maps, –.The introduction notes that combustion-engine vehicles remain more than nineteen in twenty vehicles currently driving through cities, based on the cited fleet-share estimate.Source: §I · p. 1
• The evaluation varies traffic regimes on a purpose-built two-route network to expose a congestion dependency in ErouVe’s effectiveness.The paper asks how ambient traffic conditions affect whether eco-routing has anything to offer, testing conditions not directly examined by prior evaluations.Source: §I · p. 1
• Under saturated traffic, ErouVe reduces fleet-wide CO2 by 3.28%, while the evaluation identifies three weaknesses and proposes a corrective method for each.The weaknesses are cold-start knowledge asymmetry, decision flapping under noisy reports, and a per-second CO2 metric that rewards idling; each proposed method measurably restores effectiveness.Source: §I · p. 1

II. RELATED WORK

• Related work combines cooperative awareness, traffic statistics, intelligent traffic-light control, and inter-vehicle information systems with routing or emissions reduction –.Congestion estimation has also used cooperative RSU-mediated measurement, while VANET misbehavior and intrusion detection address the trustworthiness of reported data,.Source: §II · p. 2
• Neither ErouVe’s original evaluation nor its robustness follow-up isolates ambient congestion, which this work varies alongside road architecture.The earlier studies effectively hold traffic conditions fixed while varying attacker presence; this evaluation instead varies traffic regime and road architecture with protocol and attack surface held fixed.Source: §II · p. 2
• Recent work addresses dynamic V2X routing, road-architecture bottlenecks, and misbehavior detection, but does not isolate ambient congestion as the determinant of routing opportunity –.The cited examples include SUMO studies of tunnel congestion and bus-lane rerouting, as well as federated-learning-based detection –.Source: §II · p. 2

III. THE EROUVE PROTOCOL

• ErouVe’s three phases collect vehicle measurements at RSUs, share segment averages among neighboring RSUs, and provide route guidance to requesting vehicles.The architecture is decentralized and the phases run continuously between vehicles and RSUs.Source: §III · p. 2, Figure 1 · p. 2, Figure 2 · p. 3
• RSUs aggregate per-vehicle travel-time and CO2 reports into running segment means, which are exchanged and compared instead of raw reports.The averages use reports still inside the RSU’s current averaging window.Source: §III · p. 2
• The original rule normalizes time, CO2, and distance, combines them with fixed weights, and selects the minimum-cost option.The robustness follow-up replaces the weighted sum with majority voting across the same three features; this evaluation implements the original rule.Source: §III · p. 3

IV. SYSTEM DESIGN

• The testbed is designed to execute ErouVe's three-phase protocol under varied road architectures and traffic regimes, and to assess weaknesses and corrective methods.Source: §IV · p. 3

A. Platform

• The Veins subproject links Veins 5.2, OMNeT++ 5.7, and SUMO 1.11 through TraCI for road-traffic simulation.Source: A. Platform · p. 3
• Vehicles and RSUs use IEEE 802.11p at 6 Mbps and 20 mW, while RSUs gossip every 2 s.Source: A. Platform · p. 3

B. Adapting a Paper Protocol to a Real Router

• RSUs send adapted costs for known segments, and vehicles apply them through changeRoute() so SUMO finds a minimum-cost path beyond the next hop.This avoids implementing a separate shortest-path search despite Veins not exposing road-graph connectivity to RSUs.Source: B. Adapting a Paper Protocol to a Real… · p. 3

C. Weakness 1: A Cost Metric That Rewards Idling

• Per-second CO2 rates can make congested segments appear clean because idling emits little, even when crossing the same distance burns more fuel.The method retains observed mean travel time directly as a segment cost because it reflects length and congestion.Source: C. Weakness 1: A Cost Metric That… · p. 3
• The revised CO2 cost uses emissions per meter driven, min–max normalized across the segments currently known to the RSU.The best-known segment maps to 0 and the worst to 1; the resulting value is the CO2 term in routing.Source: C. Weakness 1: A Cost Metric That… · p. 3
• Max-only normalization failed because per-meter CO2 values cluster far from zero, compressing the useful signal.Source: C. Weakness 1: A Cost Metric That… · p. 4

D. Weakness 2: Cold-Start Knowledge Asymmetry

• An RSU may divert vehicles from a good road to an unmeasured, possibly bad alternative whose default static weight looks cheap.The authors describe these problems as applying generally to systems using crowd-sourced, incrementally built estimates.Source: D. Weakness 2: Cold-Start Knowledge… · p. 4
• Knowledge-gated cost acceptance prevents adjusting an alternative's cost unless the RSU also knows the vehicle's currently planned next-segment cost.Source: D. Weakness 2: Cold-Start Knowledge… · p. 4

E. Weakness 3: Decision Flapping

• A later, noisier estimate with fewer samples can undo a correct diversion when fresh costs are reapplied at roughly 2 s gossip intervals.Diversion locking stops a vehicle from reprocessing gossip for its current segment after it leaves its default plan; the paper later discusses a residual case these methods do not fully resolve.Source: E. Weakness 3: Decision Flapping · p. 4

V. EXPERIMENTAL SETUP

• The bundled Erlangen example offered effectively one shared corridor, so the authors instead sought a network with a measurable route choice.Source: §V · p. 4

A. Network Topology

• The direct route is nominally faster and SUMO's default, but crosses two signalized junctions; the bypass is 6% longer, has a lower speed limit, and is signal-free.Source: Figure 3 · p. 4
• Each junction uses a static 90 s signal program, yielding roughly 47% effective green time and estimated one-lane saturation throughput of about 0.24 veh/s (≈880 veh/h).The authors describe this throughput as rather optimistic and note it could affect future results.Source: A. Network Topology · p. 4

B. Traffic Demand

• Heavy traffic pushes mainFlow arrivals to roughly twice the signal’s capacity, forming a real, growing queue behind C1.Under light traffic, demand remains below capacity and delay is limited to each vehicle waiting through one red phase.Source: B. Traffic Demand · p. 5
• All vehicles represent conventional petrol passenger cars using the HBEFA3/PC_G_EU4 emission class.Each vehicle is 5 m long, with 2.6/4.5 m/s² acceleration/deceleration and a 0.5 driver-imperfection sigma.Source: B. Traffic Demand · p. 5

C. Protocol Parameters

• ErouVe’s configuration is listed in Table III.Source: Table III · p. 5

A. Isolated Route Comparison (No Rerouting)

• The bypass’s CO₂ advantage grows from 8.9% to 20.5% under heavy traffic as the direct route’s mean travel time rises from 161 s nominal to 339 s observed.The maximum observed queue was 29 vehicles.Source: A. Isolated Route Comparison (No… · p. 5, Table IV · p. 5

B. Protocol-Driven Results

• Under light traffic, Baseline and EcoRouting fleet mean CO₂ are statistically indistinguishable.Source: Figure 4 · p. 5
• EcoRouting reduces the fleet mean CO₂ by 3.28%.Source: B. Protocol-Driven Results · p. 5, Table V · p. 5
• In heavy traffic, 92 of 150 mainFlow vehicles diverted and saved 7.7% on average, while 29 of 63 rerouted exploreFlow vehicles regressed 10.4%.Source: Figure 5 · p. 5, B. Protocol-Driven Results · p. 5, Table VI · p. 5

VII. DISCUSSION

• The bypass is 6% longer in every run, but its value depends on whether queueing makes the direct route costly enough to justify diversion.Under heavy traffic, direct travel time rises from 161 s nominal to 339 s observed; under light traffic, the fleet-wide effect is indistinguishable from noise.Source: §VII · pp. 5–6
• Even after diversion locking and knowledge-gated cost acceptance, 29 of 63 bypass-assigned vehicles reroute and that group’s mean CO₂ rises.The authors identify incomplete knowledge of the competing route as the most likely remaining cause.Source: §VII · p. 6
• The emission model represents a petrol passenger car, a vehicle type that accounts for upwards of 95% of vehicles on the road.The paper cites for the fleet-share figure.Source: §VII · p. 6

• 3.28% fleet-wide mean CO2 reduction and 7.7% lower emissions for diverted vehicles were reported under traffic dense enough to saturate a signalized route.Source: §VIII · p. 6
• Distance-normalized CO2 cost, knowledge-gated cost acceptance, and diversion locking measurably restored protocol effectiveness.Source: §VIII · p. 6
• Future work targets the ExploreFlow regression and tests the majority-voting decision rule from the follow-up study against the same scenarios.Source: §VIII · p. 6
• Further evaluation will test multiple diverge points and multi-hop I2I communications, then repeat the light/heavy comparison on a real OpenStreetMap-derived network.Source: §VIII · p. 6