Safe Nonlinear Model Predictive Control using a Learned Approximate Control-Invariant Set

In recent years, advanced model-based and datadriven control methods are unlocking agile and dynamic motions of robotics systems. However, ensuring safety with these control methods remains a challenge. This extended abstract presents a novel Model Predictive Control (MPC) scheme that can guarantee safety using a conservative approximation of a safe set (a.k.a. control-invariant set). This method replaces the classic terminal constraint with a novel receding constraint, which leads to a higher number of completed tasks, while retaining safety guarantees. We evaluated our approach on a simulated robot manipulator, empirically demonstrating its superiority to state-of-the-art MPC schemes.