Connected Autonomous Vehicles

• Connected and Autonomous Vehicles (CAVs) are intelligent transportation systems that integrate sensor fusion, V2X communication, and autonomous control to optimize safety and efficiency.
• They leverage collaborative perception, BEV pipelines, and multi-agent coordination to reduce delays, congestion, and energy consumption across urban and highway settings.
• Key challenges include ensuring adversarial robustness, scalable communication protocols, and secure control architectures in mixed-autonomy environments.

Connected and Autonomous Vehicles (CAVs) comprise a transformative class of intelligent transportation systems, integrating networked vehicle-to-everything (V2X) communication with algorithmic autonomy in sensing, perception, decision-making, and control. This cyber-physical fusion enables cooperative perception, multi-agent coordination, adaptive traffic management, and new modalities of safety and efficiency not attainable by isolated human-driven or non-networked autonomous vehicles. CAV research draws on control theory, deep learning, communication systems, cybersecurity, large-scale optimization, and traffic modeling. Current deployments and proposals span urban corridors, highways, and mixed traffic, with quantitative demonstrations of dramatic gains in throughput, safety, and energy consumption. However, numerous technical, regulatory, and economic challenges remain, including adversarial robustness, scalable protocol design, and multi-agent consensus in mixed-autonomy environments.

1. Fundamentals of CAV Sensing, Communication, and Perception

CAVs utilize a heterogeneous array of onboard sensors—LiDAR, cameras, radar, IMU, GNSS—and V2X links to enable situational awareness well beyond the capabilities of classical automated vehicles. The Bird’s-Eye-View (BEV) pipeline, in which multi-camera images are unprojected onto a common grid, forms the dominant paradigm for ego-centric and collaborative perception, though BEV’s vulnerability to occlusion-induced blind spots motivates data fusion from neighboring vehicles.

Collaborative perception (CP) protocols such as PACP (Priority-Aware Collaborative Perception) align and fuse BEV outputs from multiple vehicles using geometric transformations and inter-BEV correlation metrics (e.g., IoU-based priority weights) (Fang et al., 2024). Efficient transmission is achieved by gating low-value neighbor views, optimizing resource allocation via submodular maximization, and compressing sensor data through adaptive, channel-aware deep autoencoders. This approach yields marked improvements in both perception coverage and detection accuracy relative to fairness- or throughput-optimized baselines.

Localization accuracy is further improved via multi-modal sensor fusion and cooperative localization. Graph Signal Processing (GSP)-based algorithms employ Laplacian regularization within dynamic vehicle graphs to integrate GPS anchors, LiDAR/RADAR ranges, and relative bearings, combined with temporal low-rank modeling to reduce GPS-induced errors by up to 94% (Piperigkos et al., 2021).

2. Control Architectures: From Vehicle to Network Scale

At the vehicle level, control strategies are anchored in dynamically- or kinematically-informed models with real-time parameter adaptation. Models such as the Information-Aware Driver Model (IADM) use both onboard sensor and V2X data to compute dynamic “safe gaps” and execute stable, comfortable, and efficient acceleration/braking decisions without explicit calibration for traffic regime (Rahman et al., 2020). Such controllers exhibit local and string stability, outperforming classical models (e.g., IDM) in trajectory tracking, jerk minimization, and flow (Rahman et al., 2020).

At the network scale, collaborative control frameworks include:

• Centralized and distributed Model Predictive Control (MPC) for traffic smoothing, corridor speed harmonization, and intersection coordination, leveraging real-time state exchange through V2V/V2I channels (Malikopoulos, 2022, Liu et al., 2023).
• Consensus protocols for platooning and Cooperative Adaptive Cruise Control (CACC), ensuring string stability and scalable performance through appropriate gain and topology selection.
• Bi-level methods combining local feedback (e.g., ACC) with network-wide information for routing and merging.
• Game-theoretic and variational equilibrium-based planners for multi-agent trajectory generation with coupled collision avoidance constraints and interaction-fairness, solved efficiently by multiplier consensus algorithms supported by RSU-mediated V2X (Liu et al., 2024).

Microscopic simulation studies (SUMO, VISSIM, CARLA) and scaled testbeds validate the superiority of these methods in metrics including average delay, comfort, and energy use. For example, cooperative schedule-driven intersection control can reduce average per-vehicle delay by 19–21% compared to decentralized schemes and retains ~14% delay savings even at 50% CAV penetration (Hu et al., 2019). Real-world CAV deployments and simulation demonstrate up to 4× congestion reduction, 300% speed gains, and 3× shorter trip times at full penetration (Mavromatis et al., 2020).

3. Communication Technologies: V2V, Cellular, and mmWave

Selecting communication modalities is application-dependent and regulated by the spatiotemporal requirements of control and perception tasks (Johri et al., 2015). V2V (IEEE 802.11p DSRC/C-V2X) delivers low-latency (<100 ms), small-scale (<300 m) communication for active safety, platooning, and collective perception. Cellular (LTE, 5G NR) supports large-scale, delay-tolerant applications such as infotainment, cloud analytics, and over-the-air updates.

mmWave (24–100 GHz) technology, via 3GPP 5G NR FR2 and IEEE 802.11ad/ay/bd, enables multi-Gb/s, sub-10 ms, high-reliability links to support sensor-data streaming for cooperative perception and platooning. Nevertheless, severe susceptibility to blockage, frequent beam alignment, and rapid handover necessitate multimodal beamforming, multi-connectivity, edge intelligence, and hybrid analog/digital arrays to maintain coverage and reliability (Mollah et al., 2023).

Hybrid networking strategies orchestrate both V2V and cellular for workloads in the medium–large Δt regime, dynamically offloading bulk or delay-tolerant data to infrastructure while reserving V2V for time-critical coordination (Johri et al., 2015).

4. Security and Resilience in CAV Networks

CAVs introduce a vastly expanded attack surface, as outlined by comprehensive taxonomies of vulnerabilities (Hossain et al., 2023, Pham et al., 2020). Attacks span:

• In-vehicle network (CAN/SAE J1939) spoofing, ECU firmware tampering, keyless-entry compromise.
• Internet-based exploits including V2V/V2I Sybil, DDoS, and cloud-side update interception.
• Sensor/algorithm attacks: GPS spoofing, LiDAR/camera jamming, adversarial ML perturbations.

Defenses combine cryptographic PKI-based authentication (IEEE 1609.2), real-time intrusion/anomaly detection (statistical, autoencoder-based), robust multi-sensor fusion, and DNN adversarial hardening (Pham et al., 2020). Performance trade-offs arise between detection accuracy, real-time latency (e.g., <10 ms on the CAN bus), and resource constraints on automotive-grade ECUs.

Emerging challenges include defending against generalized ML attacks, guaranteeing secure OTA update channels, privacy-preserving V2X protocols (pseudonymization), and digital-twin-based pre-deployment resilience testing (Hossain et al., 2023). The field urgently calls for harmonized protocols that scale across manufacturers and fleets, and for methods to bridge academia–industry gaps in fielded defenses.

5. Cooperative Planning, Traffic Management, and Mixed Autonomy

CAVs afford network-level optimizations that propagate benefits across urban and highway settings, even at modest penetration rates. Cooperative schedule-driven traffic signal management reshapes platoon arrivals via V2I speed guidance, achieving proved reductions in delay and emissions in both simulation and hardware-in-the-loop real-time experiments (Hu et al., 2019, Han, 2020). Hybrid traffic-law paradigms—differentiating CAV and HDV regulations and dynamically managing special-lane access—minimize average passenger delay by ~70% relative to legacy fixed schemes at low CAV rates (Kraicer et al., 18 Feb 2025).

In mixed traffic, CAV control must reconcile stability–safety trade-offs, especially when interacting with aggressive HDVs. Cooperative controllers augmented by Control Barrier Function (CBF)-based safety filters guarantee per-agent and platoon-level safety, while connectivity to downstream or upstream HDVs (via “look-ahead” or “look-behind” modes) further improves string stability or HV safety, respectively (Zhao et al., 2024).

Federated learning frameworks exploit V2X to facilitate privacy-preserving collaborative tuning of onboard controllers (e.g., autonomous PID), efficiently aggregating heterogeneous and non-iid data across distributed CAVs via dynamic proximal optimization and contract-theoretic incentives (Zeng et al., 2021).

6. Open Problems and Future Directions

Current research trajectories include scalable, real-time simulation and validation platforms (hardware-in-the-loop, VR-based mixed reality), quantifying and managing uncertainty in communication, automation, and human driver behavior, and developing robust, agent-based policy learning frameworks capable of generalization and rapid on-the-fly adaptation (Silva et al., 2017, Tzortzoglou et al., 1 Jul 2025).

Further challenges span:

• Adapting collaborative perception to multi-modal sensor suites (LiDAR/radar/camera fusion), under adversarial or degraded channel conditions (Fang et al., 2024).
• Efficient, real-time consensus and optimization protocols for dense, dynamically changing CAV graphs (Liu et al., 2024).
• Secure scalability in credential and OTA update management, especially in the face of regulatory, liability, and privacy fragmentation.
• Integration of machine learning with control and optimization to enable resilient, cooperative autonomy in highly uncertain, adversarial, and safety-critical scenarios.

The ongoing convergence of communication, control, perception, and security in CAV systems positions this field at the intersection of critical technological and societal infrastructure, with far-reaching implications for traffic efficiency, safety, energy sustainability, and urban planning.