ABSTRACTThis paper proposes the reinforcement learning‐based prescribed performance optimal formation control approach for a fleet of unmanned surface vehicles (USVs) with collision avoidance and preserved connectivity. Firstly, to improve the prescribed performance of formation errors while ensuring collision avoidance and preserving connectivity between two successive USVs, the monotone tube boundary functions are designed. Compared to the traditional exponential form of prescribed performance control, the formation error can converge accurately to the vicinity of the origin within a predefined time, and the overshoot of the formation error is constrained. Secondly, in order to derive realistic optimal solutions, an actor‐critic reinforcement learning algorithm is implemented and the unknown parameters are approximated using identifier neural networks, which simplifies the Hamilton‐Jacobi‐Bellman equation to obtain optimal control solutions. Meanwhile, to simplify the computational complexity, a simple positive function is designed, and the gradient descent method is applied to it such that the simplified critic and actor weight updating laws are acquired. Finally, within the framework of the backstepping method, the adaptive optimal formation control strategy ensures that USVs follow a given reference trajectory and maintain the required line‐of‐sight (LOS) range between two consecutive USVs. The feasibility and effectiveness of the proposed control approach are well illustrated by the simulation results.
Read full abstract