Geometric Conditions for Lossless Convexification in Fuel-Optimal Control of Linear Systems with Discrete-Valued Inputs

Abstract: Trajectory generation for autonomous systems with discrete-valued actuators is challenging due to the mixed-integer nature of the resulting optimization problems, which are generally intractable for real-time, safety-critical applications. Lossless convexification offers an alternative by reformulating mixed-integer programs as equivalent convex programs that can be solved efficiently with guaranteed convergence. This paper develops a lossless convexification framework for the fuel-optimal control of linear systems with discrete-valued inputs. We extend existing Mayer-form results by showing that, under simple geometric conditions, system normality is preserved when reformulating Lagrange-form problems into Mayer-form. Furthermore, we derive explicit algebraic conditions for normality in systems with cross-polytopic input sets. Leveraging these results and an extreme-point relaxation, we demonstrate that the fuel-optimal control problem admits a lossless convexification, enabling real-time, discrete-valued solutions without resorting to mixed-integer optimization. Numerical results from Monte Carlo simulations confirm that the proposed approach consistently yields discrete-valued control inputs with computation times compatible with real-time implementation.