Lightweight reinforcement learning for leader following in autonomous vehicle convoys

This work, presents an innovative strategy for autonomous convoy navigation that rethinks traditional approaches. Instead of equipping each vehicle with its own computationally intensive navigation system, our framework designates a single lead vehicle to perform the heavy lifting of global path planning and obstacle mapping. Follower vehicles adopt a lightweight, reinforcement learning (RL)–driven policy that allows for rapid local adaptations. Extensive validation through simulations in both PyBullet and Gazebo environments demonstrates that our decentralized scheme markedly reduces overall computational overhead and energy consumption while enhancing the synchronization and responsiveness of the convoy. We refer to this leader-follower framework throughout this paper as the Pied Piper Policy (PPP), reflecting its divide-and-conquer approach to convoy autonomy.