Platooning Connected, Autonomous, and Human-Driven Vehicles: A Deep Reinforcement Learning-based Approach

Researchers propose a hybrid vehicle platooning system using deep reinforcement learning that allows non-connected vehicles to safely join autonomous platoons while managing traffic flow stability. The approach addresses real-world mixed traffic conditions by dynamically controlling platoon structures to suppress disturbance propagation, reduce fuel consumption, and improve safety—demonstrating significant improvements in balancing traffic capacity with stability.

This research tackles a fundamental challenge in autonomous vehicle deployment: managing mixed traffic where connected and non-connected vehicles operate simultaneously. Current platooning systems assume all vehicles are connected, creating a deployment gap between idealized models and actual road conditions where legacy vehicles, older models, and non-connected autonomous vehicles coexist. The study's innovation lies in explicitly designing for this heterogeneous environment rather than ignoring it.

The core technical challenge is preventing platoon instability when unregulated non-connected vehicles join formations. The authors discovered that without controls, such integration triggers uncontrolled platoon expansion and amplifies disturbance propagation—where minor speed variations cascade through the traffic flow. Their deep reinforcement learning solution uses multi-level state representation to dynamically decide which vehicles can join, how platoons should restructure, and when to dissolve formations. This enables real-time optimization between competing objectives: maximizing traffic throughput and maintaining system stability.

For autonomous vehicle deployment and smart transportation infrastructure, this work bridges theory and practice. Rather than waiting for universal vehicle connectivity, the solution accommodates gradual market adoption of connected systems. The demonstrated reduction in fuel consumption and emissions provides environmental benefits that regulators value. For developers, this suggests that robust autonomous systems must account for imperfect information and mixed environments—not idealized scenarios.

Future validation will likely focus on real-world testing with actual traffic data and edge cases. The DRL approach's computational requirements for roadside infrastructure and communication latency remain open questions. As vehicle fleets gradually transition to connected autonomy, solutions that gracefully handle heterogeneous traffic become critical infrastructure components.

• →Deep reinforcement learning enables dynamic control of mixed vehicle platoons containing both connected and non-connected vehicles without causing traffic instability.
• →Unregulated integration of non-connected vehicles into platoons can cause rapid expansion and disturbance propagation, requiring intelligent gatekeeping and restructuring.
• →The solution achieves simultaneous improvements in traffic throughput, stability, fuel consumption, and emissions through multi-level state representation.
• →The approach addresses a critical deployment gap between idealized autonomous systems and real-world mixed traffic conditions during transition periods.
• →Platoon topology dynamically adapts based on vehicle connectivity status and traffic flow conditions rather than using fixed formation rules.