State-Dependent Uncertainty Modeling in Robust Optimal Control Problems through Generalized Semi-Infinite Programming

Abstract

Generalized semi-infinite programs (generalized SIPs) are problems featuring a finite number of decision variables but an infinite number of constraints. They differ from standard SIPs in that their constraint set itself depends on the choice of the decision variable. Generalized SIPs can be used to model robust optimal control problems where the uncertainty itself is a function of the state or control input, allowing for a less conservative alternative to assuming a uniform uncertainty set over the entire decision space. In this work, we demonstrate how any generalized SIP can be converted to an existence-constrained SIP through a reformulation of the constraints and solved using a local reduction approach, which approximates the infinite constraint set by a finite number of scenarios. This transformation is then exploited to solve nonlinear robust optimal control problems with state-dependent uncertainties. We showcase our proposed approach on a planar quadrotor simulation where it recovers the true generalized SIP solution and outperforms a SIP-based approach with uniform uncertainty bounds.