PaperlayerView exact revision on arXiv

Source-linked AI summary

ITS Fairy: Occlusion Assistance Selected Against a Recipient's Own Perception Reports

Yenan Wang, Oscar Karlsson, Elad Michael Schiller, Francesco Raviglione, Claudio Casetti

arXiv:2609.31429v1cs.NIcs.DC

TL;DR

Sensing occlusion can leave a local safety application without an object needed for collision computation, while infrastructure may hold its state. The ITS Fairy uses a recipient’s own CPM-reported awareness to select missing conflict-relevant object state for local insertion; in the tested settings, assisted operation yields larger minimum-TCA margins, with robustness conditioned on omission and analysis rate.

  • Problem

    Sensing occlusion can hide a conflict-relevant object from a vehicle’s local environment model even when infrastructure holds its state.

    Source: §I · pp. 1–2

  • Method

    The ITS Fairy selects objects absent from a recipient’s CPM-derived reported awareness that pass a TCA conflict test, then supplies their state for local recomputation and control.

    Source: §I · p. 2

  • Results

    3.83–9.47 s: assisted lane-merge medians at 40–120 km/h, versus local-only medians at or below 0.08 s; at every evaluated speed in both geometries, the smallest assisted per-encounter minimum TCA exceeds the largest local-only value.

    Source: §I · p. 2, §V · p. 7

  • Takeaways & Limitations

    The tested results show that supplying conflict-relevant state selected against a recipient’s own reports improves information available to an unchanged local safety controller under sensing occlusion.

    Source: §I · p. 2, §V · p. 7

    The primary per-encounter minimum TCA is defined only once the detector holds both objects, so it measures margin after the conflict becomes locally computable.

    Source: §III · p. 4

Abstract

from arXiv · show

Cooperative perception can expose object state beyond a vehicle's onboard sensors, but sensing occlusion can still leave a local safety application without the objects its collision computation needs. To tackle this challenge, we present the ITS Fairy, an infrastructure-side Server Local Dynamic Map (S-LDM) service whose decision unit is the pair (recipient, missing conflict-relevant object): among objects absent from a recipient's CPM-derived reported awareness, it sends only those relevant to a Time of Closest Approach (TCA) conflict test. Comparable services predict what a vehicle can perceive; the ITS Fairy instead reads what it has already reported. The recipient inserts the selected state into its local LDM and uses its unchanged collision-avoidance controller. We evaluate this application-level mechanism in SUMO--ms-van3t--S-LDM emulation, since extended as VaN3Twin, using a sensing-occluded lane merge and four-way intersection scenario. At every main-sweep speed, the smallest assisted per-encounter minimum TCA exceeds the largest local-only value in the archived data. Additionally, assisted medians remain in the multi-second range where local-only operation repeatedly approaches zero. In the lane-merge robustness data, the median benefit persists at 80% configured assistance omission with 10 and 5 Hz analysis, but largely disappears at 100-120 km/h when 80% omission is combined with 1 Hz analysis. These results demonstrate the application-level value of supplying object state selected against what a recipient has itself reported. They are not a vehicular wireless-channel evaluation, and they do not quantify what selectivity saves relative to forwarding every nearby object.

I. INTRODUCTION

The ITS Fairy addresses recipient-specific cooperative-perception assistance by selecting missing, conflict-relevant object state from what a vehicle has itself reported, while leaving collision assessment and maneuver control local. The evaluation reports higher minimum-TCA margins under sensing occlusion and examines how assistance omission and analysis rate condition that benefit.

Open Issues and Positioning

  • The open issue is selecting assistance from evidence of what a recipient reported perceiving while limiting it to missing objects relevant to an actual conflict.Sender-side generation, risk, or informativeness rules do not determine what a particular vehicle is missing for collision avoidance.Source: §I · pp. 1–2

Proposed Approach

  • The Fairy derives each recipient’s reported-awareness set from its CPM history and sends state only for residual objects that pass a Time of Closest Approach conflict test.The recipient inserts the state into its local LDM and recomputes TCA with its unchanged avoidance controller; the service sends neither a collision verdict nor a maneuver command.Source: §I · pp. 2–3
  • Because the S-LDM already ingests vehicle CPMs, reported awareness is available as an aggregation byproduct without an additional message.The design also keeps the local controller unchanged so differences reflect information availability rather than a different predictor.Source: §I · p. 2

Evaluation

  • The evaluation compares local-only and assisted operation with the same local controller in sensing-occluded lane-merge and four-way-intersection scenarios, using per-encounter minimum TCA as its primary metric.It also examines lane-merge robustness across configured omission rates and Fairy analysis rates.Source: §I · p. 2

Effectiveness

  • At every evaluated speed in both geometries, the smallest assisted per-encounter minimum TCA exceeds the largest local-only value; assisted medians remain multi-second while local-only medians are at or below 0.08 s over most of the sweep.The four-way intersection reproduces the effect in a multi-vehicle geometry.Source: §I · p. 2

Robustness

  • At 10 Hz, the benefit survives 80% assistance omission almost intact, whereas at 1 Hz the same omission leaves a benefit only at the lowest tested speed.The reported collapse at 1 Hz is abrupt rather than gradual.Source: §I · p. 2

Contributions

  • The contributions include an open-source SUMO–ms-van3t–S-LDM evaluation environment and a joint characterization of minimum-TCA benefits across omission and analysis rate.The environment implements the Fairy and the sensing-occluded lane-merge and four-way-intersection scenarios.Source: §I · p. 3

II. SYSTEM MODEL AND ITS FAIRY

The ITS Fairy runs at an infrastructure S-LDM and supplements each recipient’s local LDM with missing objects selected for TCA relevance. Collision-risk recomputation and maneuver control remain with the recipient.

II. SYSTEM MODEL AND ITS FAIRY

  • The infrastructure S-LDM aggregates vehicle CAM/CPM reports and hosts the Fairy, while collision-risk recomputation and control remain at the recipient.The evaluated system uses reported awareness rather than simulator ground truth.Source: §II · p. 3, Figure 2 · p. 3
  • For each recipient, fresh S-LDM objects in a dynamic neighborhood minus reported objects form the missing-object set.The neighborhood radius is max{∥v_v∥T_h, d_base}, with T_h = 10 s and d_base = 5 m.Source: §II · p. 3
  • The Fairy uses the recipient’s CPM history as its reported-awareness set, assumed to correspond to objects in that vehicle’s local LDM.Source: §II · p. 3
  • The Fairy applies a 2-D constant-velocity TCA test over 10 s and selects objects predicted to pass within 5 m.Source: §II · p. 3
  • For each selected object, the Fairy sends a custom CAM-like message; the recipient inserts its state into the local LDM and recomputes TCA.The message carries object state but no acceleration, confidence, or TCA verdict.Source: §II · p. 3, Figure 2 · p. 3

III. EXPERIMENTAL METHODOLOGY

The evaluation compares assisted and local-only operation with the same local controller in SUMO–ms-van3t–S-LDM emulation, using occluded merge and intersection scenarios and omission/rate sweeps. The configured omissions are application-level, and the experimental distributions use one fixed SUMO seed.

III. EXPERIMENTAL METHODOLOGY

  • Local-only and Fairy-assisted operation use the same vehicle-local collision-avoidance mechanism in SUMO–ms-van3t–S-LDM co-simulation.The emulated path uses UDP/IPv4 and virtual Ethernet to a UDP-to-AMQP relayer and S-LDM; no vehicular wireless PHY/MAC or propagation model is instantiated.Source: §III · p. 3
  • The occluded scenarios are a building-separated lane merge and a four-way intersection; per-encounter minimum TCA is primary and route traversal time secondary.The nominal speed sweep is 30–120 km/h, with 140–200 km/h as an extreme-speed stress regime.Source: Figure 3 · p. 4, §III · p. 4
  • Configured assistance omissions are application-level drops before local-LDM insertion, not measured wireless PDR or a model of correlated wireless loss.Vehicle applications do not consume direct peer V2V messages.Source: §III · p. 4
  • Table II’s frequencies denote Fairy analysis rates; robustness tests vary configured message omission across 10, 5, and 1 Hz sweeps.The slower-rate sweeps omit the 15% setting; each Fairy-to-vehicle message is dropped independently while CAM/CPM reporting remains unaffected.Source: Table II · p. 4, §III · p. 4
  • Each speed and configuration uses sequential encounters with fixed routes and nominal speed under SUMO seed 10, so distributions are descriptive rather than independently seeded replications.There are 100 encounters per configuration except two 10 Hz, 120 km/h cells with 99 and 92 usable encounters.Source: §III · p. 4

IV. RESULTS

In both tested geometries, Fairy assistance yields higher minimum-TCA margins than local-only operation under sensing occlusion. The lane-merge robustness results show that retained benefit depends jointly on assistance omission and analysis frequency.

Effect of missing-object assistance

  • At 40–70 km/h, lane-merge assisted median minimum TCA is 7.47–9.47 s, versus local-only medians at or below 0.08 s from 40–120 km/h.At 120 km/h, the assisted median is 3.83 s; across every evaluated lane-merge speed, the smallest assisted value exceeds the largest local-only value.Source: §IV · p. 4, Figure 4 · p. 5
  • Lane-merge mean traversal-time reductions grow from about 11% to 45–47% across 30–120 km/h.The reported account attributes the pattern to local-only vehicles braking late and hard, often stopping, while assisted vehicles adjust speed gradually; above 140 km/h the difference narrows without an attributed cause.Source: §IV · p. 4
  • At 40–80 km/h in the intersection, local-only medians are 0.01–0.03 s, while assisted medians decline from 4.65 to 2.24 s.The lowest assisted values across 100 encounters remain above the largest local-only value at every speed.Source: §IV · pp. 4–5

Robustness to missed assistance opportunities

  • At 1 Hz and 80% omission, the 90 km/h median remains 2.29 s, while distributions at higher tested speeds become bimodal as encounters retain or lose most of the assisted margin.At 10 Hz, recipients are re-examined every 100 ms; at 1 Hz and 120 km/h, a vehicle travels roughly 33 m between analyses against a 50 m sensing model.Source: §IV · pp. 5–7
  • At 80% omission, lane-merge median minimum TCA remains 5.49, 4.68, and 3.61 s at 90, 100, and 120 km/h with 10 Hz analysis, but at 1 Hz falls to 0.04 and 0.08 s at 100 and 120 km/h.At 5 Hz, the median remains similar, though the 120 km/h first quartile falls from 2.29 s to 1.36 s.Source: §IV · p. 5, Figure 5 · p. 6, Table III · p. 6

Interpretation

  • Minimum TCA measures the margin after the local detector can compute a conflict, not collision rate or physical proximity independently.Under occlusion, the detector cannot evaluate the conflict until the other vehicle enters the 50 m sensing model, so part of the separation reflects when the conflict becomes locally computable.Source: §IV · p. 6

V. CONCLUSION

The ITS Fairy uses an S-LDM to send recipient-specific conflict-relevant object state absent from reported awareness, while leaving collision-risk recomputation and maneuver control local. The conclusion reports improved minimum-TCA margins in the tested settings and identifies limits of the evaluation and next steps.

V. CONCLUSION

  • The ITS Fairy sends conflict-relevant object state missing from a vehicle’s reported awareness, while collision-risk recomputation and maneuver control remain local.The conclusion reports assisted lane-merge minimum TCA of 3.83–9.47 s versus at most 0.08 s without assistance from 40–120 km/h, with 80% omission benefit at 10 Hz but not 1 Hz.Source: §V · p. 7
Loading 2609.31429v1…