Collective Perception Virtual Safety Validation in Urban Environments: Scenarios, Tools, Metrics

Bolovinou, Anastasia; Panagiotopoulos, Ilias; Ballis, Athanasios; Amditis, Angelos · 2026 · Crossref

DOI: 10.1007/978-3-032-06763-0_113

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the critical need for virtual safety validation of Collective Perception (CP) in urban environments for Connected and Automated Vehicles (CAVs). CP allows CAVs to exchange perception data, effectively extending their field of view beyond on-board sensor limitations, such as blind spots and non-line-of-sight scenarios. While CP is standardized by ETSI, previous research has predominantly focused on network connectivity and VANET aspects through large-scale simulations, neglecting the perception engineering perspective. The authors identify a gap in evaluating how shared information content affects object detection and fusion under occlusions and sensor uncertainties. Consequently, the study aims to bridge this gap by reviewing recent advancements in synthetic datasets and benchmarks, and by establishing high-level safety validation requirements for CP testing in simulation. The methodology involves a comprehensive state-of-the-art review of recent CP datasets, benchmarks, and simulation tools. The authors analyze simulation-generated datasets like OpenCDA, OPV2V, and V2X-Sim, which utilize the CARLA simulator to support multi-agent perception tasks. They also examine real-world datasets such as DAIR-V2X. The review highlights the shift from network-centric testing to perception-centric evaluation, focusing on late fusion techniques where bounding boxes and confidence scores are shared. Based on this analysis, the authors propose a conceptual architecture for a CP simulation framework and derive specific validation requirements categorized into scenario generation, metrics, and test environments. The findings present a structured set of high-level validation requirements for urban CP testing. For scenario generation, the authors specify the need for multi-agent scenarios involving connected and non-connected agents, vulnerable road users, and smart infrastructure nodes in complex urban domains like intersections and roundabouts. Regarding metrics, they propose Key Performance Indicators (KPIs) including fused object localization, detection, classification, clutter rate, runtime performance, and robustness to perception errors and message delays. These metrics must account for varying CAV penetration rates and realistic perception uncertainty models. For test environments, the paper outlines three approaches: physical testing with real-world data, pure simulation using tools like CARLA, and hybrid setups combining physical CAVs with digital twins. The significance of this work lies in providing a foundational framework for the safety validation of CP systems, moving beyond network performance to assess actual perception benefits. By defining specific scenarios, metrics, and tools, the paper enables researchers to evaluate CP modules rigorously in virtual environments, addressing safety-of-intended-functionality (SOTIF) concerns. The authors conclude that future work will apply these requirements to assess Bayesian-based CP modules in CARLA, thereby advancing the development of safe, cooperative driving automation systems in mixed traffic environments.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• situational awareness

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Methodological Resource: tool software