New DRL Strategy Enables Hybrid Platoons with Non-Connected Vehicles
Deep reinforcement learning balances traffic throughput and stability in mixed-autonomy platoons.
A team led by Zhen Qin et al. has introduced a novel hybrid platooning approach that conditionally allows non-connected vehicles (both autonomous and human-driven) to join connected vehicle platoons. Most existing platooning methods assume all vehicles are connected, which doesn't reflect real-world mixed traffic. The key challenge: unregulated inclusion of non-connected vehicles causes rapid platoon expansion and amplifies disturbance propagation, worsening the inherent conflict between traffic throughput and stability.
To solve this, the authors developed a deep reinforcement learning (DRL)-based hybrid platooning control strategy. It integrates vehicle dynamics, platoon topology, and traffic flow states via a multi-level state representation network, enabling a dynamic trade-off between capacity and stability. Numerical simulations demonstrate that the DRL strategy effectively suppresses velocity disturbances by dynamically optimizing platoon structures, significantly enhancing stability and safety while reducing fuel consumption and emissions. The paper is available on arXiv (2606.20648).
- Enables non-connected AVs and human-driven cars to join platoons, improving real-world applicability.
- DRL strategy dynamically trades off traffic capacity against stability to prevent disturbance propagation.
- Simulations show reduced fuel consumption and emissions alongside enhanced safety in mixed traffic.
Why It Matters
Makes autonomous platooning practical today by integrating non-connected vehicles, improving safety and efficiency for all road users.