Differentially Flat Learning-Based Model Predictive Control Using a Stability, State, and Input Constraining Safety Filter

Learning-based optimal control algorithms unknown systems using past trajectory data and a learned model of the system dynamics. These controllers use either linear approximation dynamics, trading performance for faster computation, or nonlinear optimization methods, which typically perform better but can limit real-time applicability. In this work, we present novel controller that exploits differential flatness to achieve similar state-of-the-art learning-based with significantly less computational effort. Differential is property dynamical whereby be exactly linearized through input mapping. Here, transformation as Gaussian process used in safety filter guarantees, high probability, stability well flat state constraint satisfaction. This then refine inputs from predictive constrained two successive convex optimizations. We compare our method strategies performance, efficiency, while also respecting constraints, guaranteeing stability.