Cooperative Safety Intelligence in V2X-Enabled Transportation: A Survey

TL;DR

This survey reframes V2X safety as a cooperative safety intelligence problem by introducing the Sensor–Perception–Decision (SPD) framework. It unifies sensing, cooperative perception, and coordinated decision-making into a closed-loop reasoning process, emphasizing timing, trust, and human factors as cross-cutting constraints. The paper delivers a PRISMA-guided bibliometric analysis (2016–2025), surveys across sensing modalities, fusion strategies, and system architectures, and defines an evidence-contract approach for safety-ready outputs. It also outlines SPD-aligned datasets, benchmarks, and platforms, and offers a roadmap for scalable data infrastructures, embodied predictive intelligence, and human-centered SPD cooperation toward Zero-Accident Mobility.

Abstract

Vehicle-to-Everything (V2X) cooperation is reshaping traffic safety from an ego-centric sensing problem into one of collective intelligence. This survey structures recent progress within a unified Sensor-Perception-Decision (SPD) framework that formalizes how safety emerges from the interaction of distributed sensing, cooperative perception, and coordinated decision-making across vehicles and infrastructure. Rather than centering on link protocols or message formats, we focus on how shared evidence, predictive reasoning, and human-aligned interventions jointly enable proactive risk mitigation. Within this SPD lens, we synthesize advances in cooperative perception, multi-modal forecasting, and risk-aware planning, emphasizing how cross-layer coupling turns isolated detections into calibrated, actionable understanding. Timing, trust, and human factors are identified as cross-cutting constraints that determine whether predictive insights are delivered early enough, with reliable confidence, and in forms that humans and automated controllers can use. Compared with prior V2X safety surveys, this work (i) organizes the literature around a formal SPD safety loop and (ii) systematically analyzes research evolution and evaluation gaps through a PRISMA-guided bibliometric study of hundreds of publications from 2016-2025. The survey concludes with a roadmap toward cooperative safety intelligence, outlining SPD-based design principles and evaluation practices for next-generation V2X safety systems.

Figures (14)

• Figure 1: Road map of the survey.
• Figure 2: Comparative architectures of sensor-only, perception-only, decision-only, partial cooperation, and cooperative SPD intelligence frameworks.
• Figure 3: Annual publication counts across four major categories of V2X-enabled safety research from 2016 to 2025.
• Figure 4: Collaboration schemes for cooperative perception. (a) Single-vehicle baseline: raw data are encoded into features and decoded into results without collaboration. (b) Early fusion: agents first exchange raw observations to form a more complete scene, then a shared encoder–decoder produces cooperative perception results. (c) Intermediate fusion: each agent encodes its sensor data locally, exchanges latent features, and decodes fused features into cooperative perception results. (d) Late fusion: each agent produces local perception results, which are then exchanged and merged to obtain cooperative perception results. The gray panel depicts the communication topology (V2V/V2I), and arrows indicate the processing flow from sensing to fusion and outputs.
• Figure 5: Illustration of V2X communication modes in a connected intersection.