Advanced V2X Communication, MARL, and Platoon Coordination
Multi-agent systems (MAS) have emerged as a critical enabling technology for autonomous vehicle coordination, fundamentally transforming how vehicles communicate, cooperate, and navigate in complex traffic environments. These systems enable vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communication, essential for autonomous traffic systems, platooning, and dynamic route management, with industry leaders like Waymo, Tesla, and NVIDIA integrating multi-agent logic for decentralized coordination between vehicles and infrastructure.
Cooperative Perception (CP), enabled by Vehicle-to-Everything (V2X) communication, has emerged as a promising paradigm for enhancing situational awareness in autonomous driving by enabling vehicles to exchange perception data with surrounding vehicles, infrastructure, pedestrians, and networks, extending visibility beyond line-of-sight limitations. Through data exchange, CP enhances decision-making, enabling vehicles to better anticipate the movements of other road users, synchronize actions, and execute coordinated maneuvers, such as platooning and safe lane changes.
DSRC Technology: DSRC, based on the IEEE 802.11p protocol, emerged as the first standard for V2X communications, operating in the 5.9 GHz frequency band. The system supports vehicle speeds up to 200 km/h, nominal transmission range of 300 meters (up to 1000 meters), and default data rates of 6 Mbps (up to 27 Mbps). However, the IEEE 802.11p protocol lacks sufficient safety guarantees for critical communications in fully autonomous driving scenarios.
C-V2X Technology: C-V2X represents a more recent standard based on 3GPP LTE and 5G cellular standards. The architecture operates through two communication modes: direct short-range communication (PC5) for real-time safety applications and network-based communication (Uu) for broader coverage using 4G/5G infrastructure. C-V2X leverages cellular networks, offering broader coverage up to 10 kilometers and supporting data rates up to 100 Mbps, with latencies as low as 20 milliseconds in direct communication mode.
Multi-agent reinforcement learning (MARL) not only learns control policies but also considers interactions with all other agents in the environment, mutual influences among different system components, and distribution of computational resources. Achieving safety through MARL is challenging due to non-stationarity, partial observability, and the need for effective coordination among agents.
A 2025 study presented the first investigation into autonomous vehicle merging into existing platoons, proposing an MA-DRL-based cooperative control framework that achieved:
A novel MARL framework that enhances safety by incorporating transmission of vehicle intent and adaptive communication scheduling into a unified end-to-end learning paradigm. This approach addresses the complex coordination challenges in mixed traffic environments where autonomous and human-driven vehicles coexist.
Vehicle platooning represents one of the most mature applications of multi-agent coordination, involving groups of connected autonomous vehicles traveling in coordinated formations while maintaining short gaps through real-time communication and synchronization. Through V2V communication, connected and automated vehicles (CAVs) share position, speed, and acceleration information, enabling coordinated moves such as forming platoons and convoys to save fuel by optimizing inter-vehicle distances to reduce aerodynamic drag.
Extensive field experiments have demonstrated significant fuel savings from truck platooning, with reductions up to 20% in total fuel consumption. When two semi-trucks were platooned at 64 mph with a 36-foot following distance, results showed average fuel consumption savings of 4.5% for the leading truck and 10% for the following truck. Experiments on coordinated automatic longitudinal control of platoons of three Class 8 tractor-trailer trucks demonstrated 4-5% fuel reduction for leading trucks and 10-14% for following trucks.
China has emerged as the global leader in C-V2X deployment. In 2024, C-V2X pre-installations in Chinese passenger cars reached approximately 500,000 units with an assembly rate of 2.21%, representing significant growth from 2023's 270,000 units (1.2% penetration). By 2028, installations are expected to exceed 2 million units with installation rates exceeding 8%. China's Ministry of Industry and Information Technology announced the first batch of 20 pilot cities for V2X applications in July 2024, marking the transition from testing to large-scale implementation.
The U.S. Department of Transportation unveiled a new automated vehicle framework in 2025, with NHTSA releasing amendments to its Standing General Order for automated driving systems and Level 2 advanced driver assistance systems. NHTSA's framework emphasizes three key principles: prioritizing safety of ongoing AV operations on public roads, unleashing innovation by removing unnecessary regulatory barriers, and enabling commercial deployment to enhance safety and mobility.
A 2025 IEEE ComSoc Technology Blog explored how Agentic AI can address limitations in 5G-based V2X communications through distributed artificial intelligence at edge nodes. The proposed framework centers on the OODA model (Observe, Orient, Decide, Act), enabling real-time autonomous decision-making, multi-agent coordination without central orchestration, adaptive protocol switching based on network conditions, and embedded security monitoring.
China's Vehicle-Road-Cloud Integration and C-V2X Industry Research Report 2025 highlights advancement of road-side infrastructure from 5G-Advanced (5G-A) to 6G and beyond in intelligent transportation systems. These next-generation networks promise ultra-low latency (sub-millisecond), massive connectivity (millions of devices per square kilometer), and integrated sensing and communication capabilities that will enable even more sophisticated multi-agent coordination scenarios.
A major ongoing challenge concerns mixed traffic flow systems composed of connected autonomous vehicles and human-driven vehicles, requiring strategies to improve overall efficiency and safety by assigning appropriate control strategies to CAVs while accounting for unpredictable human driver behavior. Research continues to develop robust coordination algorithms that perform well in these transitional environments.