Optimal Formation Control for Autonomous Vehicles: A Bilayer Predefined-Time Fuzzy Reinforcement Learning Approach

This paper develops a bilayer predefined time fuzzy reinforcement learning (PT-FRL) control strategy to improve the efficiency of autonomous vehicle formation execution and reduce energy consumption. First, the fixed constraints of communication connectivity and collision avoidance are reconstructed into performance constraints, and normalized error mapping techniques are used to transform them into a new unconstrained error system. Then, based on the system, a cost function was constructed that balances cost control and performance. The control strategy adopts a bilayer architecture: In the first layer, a feedforward controller is designed to provide a more concise control object for subsequent PT-FRL controllers by compensating for known nonlinear coupling terms in advance, and can significantly reduce fuzzy logic systems computational load. and in the second layer, a PT optimal formation controller is designed using FRL to ensure that the autonomous vehicles complete the formation task within the predefined time. Finally, the effectiveness of the proposed method was verified through simulation and experiments.