Adaptive Nonlinear Control with Online Identification and Receding-Horizon Optimization

Abstract

An adaptive nonlinear optimal-control scheme is developed by combining receding-horizon iLQR, state estimation, online parameter identification, and actuator constraints. AMIGO (Adaptive Model-based Intelligent Guidance and Orchestration) organizes the computation into three Time Phases: identification, planning, and closed-loop control. A supervisory adaptive loop monitors predictive consistency during closed-loop operation and can repeat identification and planning when persistent parameter mismatch is detected. The nonlinear transition is evaluated by the fourth-order Runge-Kutta method (RK4), and model parameters are refined by the Levenberg-Marquardt method (LM). The method is illustrated by a Van der Pol oscillator, a quadcopter, and an autonomous lunar-lander descent.