pith. sign in

arxiv: 2512.11551 · v2 · pith:IORJ6I2Bnew · submitted 2025-12-12 · 💻 cs.RO

CarlaNCAP: A Framework for Quantifying the Safety of Vulnerable Road Users in Infrastructure-Assisted Collective Perception Using EuroNCAP Scenarios

J\"org Gamerdinger , Sven Teufel , Simon Roller , Oliver Bringmann This is my paper

Pith reviewed 2026-05-25 07:27 UTC · model grok-4.3

classification 💻 cs.RO
keywords collective perceptioninfrastructure-assistedvulnerable road usersEuroNCAP scenariosCARLA simulatorautonomous drivingsafety evaluationaccident avoidance
0
0 comments X

The pith

Infrastructure sensors achieve up to 100 percent accident avoidance for vulnerable road users in simulated EuroNCAP scenarios where vehicle sensors alone reach only 33 percent.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents CarlaNCAP as a framework and dataset of 11,000 frames drawn from EuroNCAP safety-critical scenarios to quantify how infrastructure-mounted sensors improve perception for vulnerable road users. Simulations demonstrate that collective perception from fixed roadside units can eliminate accidents in these cases by overcoming occlusions that vehicle sensors cannot. A sympathetic reader would care because the work supplies concrete numbers on safety gains that could inform decisions about deploying roadside infrastructure to protect pedestrians and cyclists in urban settings.

Core claim

Infrastructure-assisted collective perception, using sensors on elements such as traffic lights or lamp posts, significantly reduces accident rates for vulnerable road users in safety-critical EuroNCAP scenarios by providing enhanced viewpoints that overcome occlusions, achieving up to 100 percent accident avoidance compared to 33 percent with a vehicle equipped only with onboard sensors, as measured in the CarlaNCAP dataset.

What carries the argument

The CarlaNCAP framework and dataset, which runs EuroNCAP scenarios in the CARLA simulator to compare accident outcomes with and without infrastructure-assisted collective perception for vulnerable road users.

If this is right

  • Infrastructure sensors can supply viewpoints that reduce occlusions limiting vehicle perception in urban environments.
  • The CarlaNCAP dataset enables standardized evaluation of perception methods for vulnerable road user safety across multiple scenarios.
  • Decision makers gain quantitative evidence on safety improvements to assess infrastructure-assisted collective perception deployments.
  • Collective perception from fixed units can complement vehicle sensors to address perceptual limitations in occluded situations.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the simulation-to-reality gap proves small, cities could prioritize roadside sensor installations to protect cyclists and pedestrians more effectively than vehicle upgrades alone.
  • Extending the framework to include varying weather, lighting, or traffic densities would test whether the reported avoidance gains persist outside the chosen scenarios.
  • Linking CarlaNCAP results to real vehicle-to-infrastructure communication protocols could identify practical steps for scaling the approach.

Load-bearing premise

The CARLA simulator with the chosen sensor placements and EuroNCAP scenarios produces accident outcomes that match real-world safety results for vulnerable road users.

What would settle it

A field test using real infrastructure sensors and vehicles in matching EuroNCAP-style scenarios that records accident avoidance rates differing from the simulated 100 percent versus 33 percent.

Figures

Figures reproduced from arXiv: 2512.11551 by J\"org Gamerdinger, Oliver Bringmann, Simon Roller, Sven Teufel.

Figure 1
Figure 1. Figure 1: EuroNCAP VRU AEB scenarios for the CARLA-NCAP dataset. Figures from [19] [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Symbolic representation of the camera placement for [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Heatmap for the detection on the CPNC-50 scenario [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Heatmap for the detection on the CBNA scenario [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗
Figure 3
Figure 3. Figure 3: Results of the analysis of accident prevention capabilities on the EuroNCAP VRU AEB scenarios CPNC-50 (a), [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 6
Figure 6. Figure 6: Heatmap for the detection on the CBLA scenario [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗
read the original abstract

The growing number of road users has significantly increased the risk of accidents in recent years. Vulnerable Road Users (VRUs) are particularly at risk, especially in urban environments where they are often occluded by parked vehicles or buildings. Autonomous Driving (AD) and Collective Perception (CP) are promising solutions to mitigate these risks. In particular, infrastructure-assisted CP, where sensor units are mounted on infrastructure elements such as traffic lights or lamp posts, can help overcome perceptual limitations by providing enhanced points of view, which significantly reduces occlusions. To encourage decision makers to adopt this technology, comprehensive studies and datasets demonstrating safety improvements for VRUs are essential. In this paper, we propose a framework for evaluating the safety improvement by infrastructure-based CP specifically targeted at VRUs including a dataset with safety-critical EuroNCAP scenarios (CarlaNCAP) with 11k frames. Using this dataset, we conduct an in-depth simulation study and demonstrate that infrastructure-assisted CP can significantly reduce accident rates in safety-critical scenarios, achieving up to 100% accident avoidance compared to a vehicle equipped with sensors with only 33%. Code is available at https://github.com/ekut-es/carla_ncap

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The paper introduces CarlaNCAP, a simulation framework and dataset (11k frames) based on CARLA and EuroNCAP scenarios to quantify safety gains for vulnerable road users from infrastructure-assisted collective perception. It reports that infrastructure CP yields up to 100% accident avoidance in safety-critical scenarios versus 33% for a vehicle equipped only with onboard sensors.

Significance. If the simulation outcomes prove representative, the work supplies concrete, scenario-specific quantitative evidence supporting infrastructure CP adoption for VRU protection, backed by public code release. This addresses a practical gap in demonstrating safety benefits beyond qualitative arguments.

major comments (1)
  1. [Simulation study and results (inferred from abstract and framework description)] The central quantitative claims (up to 100% vs. 33% accident avoidance) rest on CARLA-generated collision outcomes in the simulation study; no calibration of sensor models, occlusion physics, or vehicle dynamics against real EuroNCAP crash statistics or field data is presented, leaving transferability unverified.
minor comments (2)
  1. Clarify the exact composition of the 11k-frame dataset (number of distinct scenarios, runs per scenario, and randomization parameters) to support reproducibility claims.
  2. The abstract states 'Code is available'; confirm whether the released repository includes the exact sensor placement configurations and controller logic used for the reported runs.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the detailed review and constructive comment. We address the major concern point-by-point below and outline planned revisions.

read point-by-point responses

  1. Referee: The central quantitative claims (up to 100% vs. 33% accident avoidance) rest on CARLA-generated collision outcomes in the simulation study; no calibration of sensor models, occlusion physics, or vehicle dynamics against real EuroNCAP crash statistics or field data is presented, leaving transferability unverified.

    Authors: We agree that the quantitative results are derived from CARLA simulations without direct calibration to real EuroNCAP crash statistics or field data. The CarlaNCAP framework and dataset are explicitly positioned as a simulation-based tool to enable standardized, scenario-specific evaluation of infrastructure-assisted collective perception using established EuroNCAP test protocols. CARLA's underlying models for sensors, occlusions, and vehicle dynamics follow documented physical approximations commonly used in the AD research community, but we did not perform empirical matching to crash databases. In the revised manuscript we will (1) add an explicit Limitations section that states the simulation-to-real gap and the unverifiable transferability of the reported percentages, (2) qualify all safety-gain claims as 'potential benefits under the modeled conditions,' and (3) outline concrete next steps for future real-world validation. These changes will make the scope and limitations of the study transparent without altering the core contribution of the open framework and dataset. revision: yes

Circularity Check

0 steps flagged

No circularity: results are direct simulation outputs

full rationale

The paper reports accident avoidance percentages obtained by running CARLA simulations on a fixed set of EuroNCAP scenarios under three sensor configurations (vehicle-only, infrastructure-assisted CP, etc.). No equations, fitted parameters, or self-citations are used to derive the reported rates; the 100 % vs. 33 % figures are literal counts of collision events produced by the simulator. The derivation chain therefore consists solely of experimental execution rather than any reduction of outputs to inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No free parameters, axioms, or invented entities are described in the abstract. The work relies on the established CARLA simulator and standard EuroNCAP scenarios.

pith-pipeline@v0.9.0 · 5756 in / 1078 out tokens · 30105 ms · 2026-05-25T07:27:43.699287+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

22 extracted references · 22 canonical work pages

  1. [1]

    Road safety: Commission elcomes agreement on new EU rules to help save lives,

    European Commission, “Road safety: Commission elcomes agreement on new EU rules to help save lives,” 2019, [Online]. Available: https: //ec.europa.eu/commission/presscorner/detail/en/IP 19 1793, [Accessed: December 5, 2022]

    work page 2019

  2. [2]

    Accounting for the special role of infrastructure-assisted collective perception,

    F. A. Schiegg, A. Rueeck, J. Gamerdinger, H. Tchouankem, E. Xhoxhi, and G. V olk, “Accounting for the special role of infrastructure-assisted collective perception,” in2023 IEEE 26th International Conference on Intelligent Transportation Systems (ITSC), 2023, pp. 189–195

    work page 2023

  3. [3]

    Cooperative infrastructure perception,

    F. Ahmad, C. S. Shin, W. Pang, B. Leong, P. Ghosh, and R. Govindan, “Cooperative infrastructure perception,” in2024 IEEE/ACM Ninth Inter- national Conference on Internet-of-Things Design and Implementation (IoTDI), 2024, pp. 61–72

    work page 2024

  4. [4]

    Optimal placement of roadside infrastructure sensors towards safer autonomous vehicle deployments,

    R. Vijay, J. Cherian, R. Riah, N. De Boer, and A. Choudhury, “Optimal placement of roadside infrastructure sensors towards safer autonomous vehicle deployments,” in2021 IEEE International Intelligent Trans- portation Systems Conference (ITSC). IEEE, 2021, pp. 2589–2595

    work page 2021

  5. [5]

    A comprehensive safety metric to evaluate perception in autonomous systems,

    G. V olk, J. Gamerdinger, A. von Bernuth, and O. Bringmann, “A comprehensive safety metric to evaluate perception in autonomous systems,” in2020 IEEE 23rd International Conference on Intelligent Transportation Systems (ITSC). IEEE, 2020, pp. 1–8

    work page 2020

  6. [6]

    Dolphins: Dataset for collaborative perception enabled harmonious and interconnected self-driving,

    R. Mao, J. Guo, Y . Jia, Y . Sun, S. Zhou, and Z. Niu, “Dolphins: Dataset for collaborative perception enabled harmonious and interconnected self-driving,” inProceedings of the Asian Conference on Computer Vision, 2022, pp. 4361–4377

    work page 2022

  7. [7]

    Deepaccident: A motion and accident prediction benchmark for v2x autonomous driving,

    T. Wang, S. Kim, W. Ji, E. Xie, C. Ge, J. Chen, Z. Li, and P. Luo, “Deepaccident: A motion and accident prediction benchmark for v2x autonomous driving,”arXiv preprint arXiv:2304.01168, 2023

    work page arXiv 2023

  8. [8]

    Tumtraf v2x cooperative perception dataset,

    W. Zimmer, G. A. Wardana, S. Sritharan, X. Zhou, R. Song, and A. C. Knoll, “Tumtraf v2x cooperative perception dataset,”arXiv preprint arXiv:2403.01316, 2024

    work page arXiv 2024

  9. [9]

    Scope: A synthetic multi-modal dataset for collective perception including physical-correct weather conditions,

    J. Gamerdinger, S. Teufel, P. Schulz, S. Amann, J.-P. Kirchner, and O. Bringmann, “Scope: A synthetic multi-modal dataset for collective perception including physical-correct weather conditions,” in2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), 2024, pp. 1–8

    work page 2024

  10. [10]

    Quantifying the performance and optimizing the placement of roadside sensors for cooperative vehicle-infrastructure systems,

    Y . Du, F. Wang, C. Zhao, Y . Zhu, and Y . Ji, “Quantifying the performance and optimizing the placement of roadside sensors for cooperative vehicle-infrastructure systems,”IET intelligent transport systems, vol. 16, no. 7, pp. 908–925, 2022

    work page 2022

  11. [11]

    Roadside lidar placement for cooperative traffic detection by a novel chance constrained stochastic simulation optimization approach,

    Y . Chen, L. Zheng, and Z. Tan, “Roadside lidar placement for cooperative traffic detection by a novel chance constrained stochastic simulation optimization approach,”Transportation Research Part C: Emerging Technologies, vol. 167, p. 104838, 2024

    work page 2024

  12. [12]

    Collective perception datasets for autonomous driving: A comprehen- sive review,

    S. Teufel, J. Gamerdinger, J.-P. Kirchner, G. V olk, and O. Bringmann, “Collective perception datasets for autonomous driving: A comprehen- sive review,” in2024 IEEE Intelligent V ehicles Symposium (IV), 2024, pp. 1548–1555

    work page 2024

  13. [13]

    Dair-v2x: A large-scale dataset for vehicle- infrastructure cooperative 3d object detection,

    H. Yu, Y . Luo, M. Shu, Y . Huo, Z. Yang, Y . Shi, Z. Guo, H. Li, X. Hu, J. Yuanet al., “Dair-v2x: A large-scale dataset for vehicle- infrastructure cooperative 3d object detection,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 21 361–21 370

    work page 2022

  14. [14]

    V2x-seq: A large-scale sequential dataset for vehicle-infrastructure cooperative perception and forecasting,

    H. Yu, W. Yang, H. Ruan, Z. Yang, Y . Tang, X. Gao, X. Hao, Y . Shi, Y . Pan, N. Sun, J. Song, J. Yuan, P. Luo, and Z. Nie, “V2x-seq: A large-scale sequential dataset for vehicle-infrastructure cooperative perception and forecasting,” 2023

    work page 2023

  15. [15]

    V2x-vit: Vehicle-to-everything cooperative perception with vision transformer,

    R. Xu, H. Xiang, Z. Tu, X. Xia, M.-H. Yang, and J. Ma, “V2x-vit: Vehicle-to-everything cooperative perception with vision transformer,” inEuropean conference on computer vision. Springer, 2022

    work page 2022

  16. [16]

    CARLA: An Open Urban Driving Simulator,

    A. Dosovitskiy, G. Ros, F. Codevilla, A. Lopez, and V . Koltun, “CARLA: An Open Urban Driving Simulator,” inConference on robot learning. PMLR, 2017, pp. 1–16

    work page 2017

  17. [17]

    V2x-sim: Multi-agent collaborative perception dataset and benchmark for autonomous driving,

    Y . Li, D. Ma, Z. An, Z. Wang, Y . Zhong, S. Chen, and C. Feng, “V2x-sim: Multi-agent collaborative perception dataset and benchmark for autonomous driving,”IEEE Robotics and Automation Letters, 2022

    work page 2022

  18. [18]

    V2x-real: a largs-scale dataset for vehicle-to- everything cooperative perception,

    H. Xiang, Z. Zheng, X. Xia, R. Xu, L. Gao, Z. Zhou, X. Han, X. Ji, M. Li, Z. Menget al., “V2x-real: a largs-scale dataset for vehicle-to- everything cooperative perception,”arXiv preprint arXiv:2403.16034, 2024

    work page arXiv 2024

  19. [19]

    Test protocol – aeb/lss vru systems,

    EUROPEAN NEW CAR ASSESSMENT PROGRAMME, “Test protocol – aeb/lss vru systems,” Feb 2024, version 4.5.1

    work page 2024

  20. [20]

    Carla-bsp: a simulated dataset with pedestrians,

    M. Wielgosz, A. M. L ´opez, and M. N. Riaz, “Carla-bsp: a simulated dataset with pedestrians,” 2023. [Online]. Available: https://arxiv.org/abs/2305.00204

    work page arXiv 2023

  21. [21]

    Faster r-cnn: Towards real-time object detection with region proposal networks,

    S. Ren, K. He, R. Girshick, and J. Sun, “Faster r-cnn: Towards real-time object detection with region proposal networks,”IEEE transactions on pattern analysis and machine intelligence, vol. 39, no. 6, pp. 1137– 1149, 2016

    work page 2016

  22. [22]

    Towards Realistic Evaluation of Collective Perception for Connected and Automated Driving,

    G. V olk, Q. Delooz, F. A. Schiegg, A. V on Bernuth, A. Festag, and O. Bringmann, “Towards Realistic Evaluation of Collective Perception for Connected and Automated Driving,” in2021 IEEE International Intelligent Transportation Systems Conference (ITSC), 2021, pp. 1049– 1056

    work page 2021