Optimized predefined-time control for high-order nonlinear MASs via ICA and reinforcement learning

This paper presents an adaptive optimized controller for high-order nonlinear multi-agent systems (MASs) with practical predefined-time consensus. By integrating reinforcement learning (RL) and sliding mode control (SMC) within the identifier-critic-actor (ICA) framework, the proposed controller shapes the tracking errors through prescribed-performance bounds and drives the neighborhood sliding variables to a compact residual set within a predefined time. The convergence-time upper bound can be adjusted through design parameters and is independent of the initial conditions. A projection-correction technique is incorporated into the neural-network (NN) weight update laws to prevent weight drift and support the practical predefined-time stability analysis. The proposed framework also minimizes the cost function associated with the neighborhood consensus error and the control input. Finally, numerical simulations verify the effectiveness of the proposed method.