Abstract
Adaptive Dynamic Programming (ADP), synergizing reinforcement learning, neural networks, and optimal control theory, has emerged as a powerful data-driven methodology for solving complex decision-making and control problems in nonlinear systems. This survey provides a comprehensive and unified overview of the core principles and recent advances in ADP-based control, structured around an integrated "Intelligent Critic-Decision-Communication" framework. We dissect this framework into four foundational pillars: (1) the theoretical underpinnings rooted in the Hamilton-Jacobi-Bellman/Hamilton-Jacobi-Isaacs (HJB/HJI) equations and the mitigation of the curse of dimensionality; (2) the algorithmic core revolving around the adaptive critic-actor architecture and its enhanced learning mechanisms, including experience replay, concurrent learning, meta-reinforcement learning, and transformer architectures; (3) the augmentation of control objectives to handle real-world constraints such as safety, input saturation, and external disturbances through barrier functions and game theory; and (4) the resource-aware implementation strategies with a particular emphasis on the deep integration of event-triggered mechanisms-from static and dynamic event-triggering to periodic, learning-based, and self-triggered control-along with output feedback observers for multi-agent systems. By synthesizing a vast body of literature spanning foundational theory to industrialgrade applications, this paper offers a clear and structured perspective on the state of the art, elucidating the deep interconnections between algorithmic design, constraint satisfaction, and practical deployment.