Cooperative Driving and Collective Navigation Architectures for CCAVs: Contemporary Techniques, Simulators, Datasets, and Research Directions

Ahiska, Kenan · 2026 · Crossref

DOI: 10.5772/intechopen.1013696

archive: archived pipeline: cataloged verified

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

Summary

This chapter reviews contemporary architectures for cooperative driving and collective navigation in connected and cooperative autonomous vehicles (CCAVs). The research is motivated by the limitations of single-vehicle autonomous systems, which suffer from restricted perception ranges, occlusions, and an inability to coordinate with surrounding traffic. By leveraging vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communication, CCAVs aim to globally optimize safety and efficiency through shared sensor data and collaborative decision-making. The text categorizes CCAV autonomy into cooperative planning and control, as well as collective perception and navigation. The review analyzes two primary technical approaches for cooperative driving: multi-agent reinforcement learning (MARL) and end-to-end deep learning. MARL applications are categorized by motion dimensionality, addressing one-dimensional platooning, two-dimensional lane changes, and three-dimensional scenarios such as traffic signal control and unsignalized intersections. Specific methods include LSTM-based protocols for platooning, distributed proximal policy optimization for urban traffic, and value decomposition approaches like QMIX for complex intersections. End-to-end approaches map raw sensor data directly to control signals, bypassing modular pipelines. The chapter examines architectures such as CooperNaut, UniV2X, V2X-VLM, and SEAL, noting that recent models utilize vision-language models and semantic alignment to handle heterogeneous data and adverse weather conditions. For collective navigation, the text evaluates fusion strategies, identifying intermediate fusion as the optimal balance between communication bandwidth and perceptual accuracy. Unlike early fusion, which shares raw data, or late fusion, which shares detection outputs, intermediate fusion shares compressed neural features. The review details architectures like V2VNet, OPV2V, F-Cooper, and DiscoNet, which employ spatial alignment, attention mechanisms, and graph-based learning to improve detection under occlusion. Transformer-based solutions, including V2X-ViT, CoBEVT, CoFormerNet, and HM-ViT, are highlighted for their ability to handle multi-agent, multi-modal sensor fusion using heterogeneous self-attention and sparse attention mechanisms. The chapter further surveys the simulators and datasets essential for benchmarking these architectures. Simulators such as CARLA, SUMO, LGSVL, CityFlow, Apollo CyberRT, and MetaDrive are compared based on their physics fidelity, connectivity support, and suitability for reinforcement learning or traffic modeling. Datasets are categorized into simulated (e.g., COMAP, OPV2V, V2X-Sim 2.0, DeepAccident, AdverCity) and real-world (e.g., V2V4Real, TUMTraf) sources. The analysis emphasizes the gap between simulated and real-world data, particularly regarding adverse weather and communication limits, and identifies the need for robust, scalable solutions that can generalize across diverse traffic scenarios and sensor modalities.

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.

• v2x connected vehicle

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