Trajectory tracking for systems with unknown time-varying disturbances

M. Cao, A. Harapanahalli, E. Morales Cuadrado, S. Coogan
IEEE Open Journal of Control Systems, 2026

Abstract

We consider trajectory tracking in nonlinear systems with partially unknown time-dependent and state-dependent dynamics. Given a reference trajectory, we leverage the mixed monotonicity property of dynamical systems to craft an embedding system, a single trajectory of which efficiently computes a forward invariant tube around the reference. We incorporate the unknown portions of the dynamics by modeling them as time-varying Gaussian processes, an existing machine learning technique, and combine them with tools from control theory via a general inclusion function which generates an associated embedding system that accounts for the first order interactions between the disturbance behavior, feedback controller, and system state. We illustrate the efficacy of this formulation through a trajectory tracking scenario in a software-in-the-loop simulation and demonstrate the viability of this method in practical implementation via physical hardware experiments on a quadrotor.