An efficient bounded-variable nonlinear least-squares algorithm for embedded MPC

This paper presents a novel approach to solve linear and nonlinear model predictive control (MPC) problems that requires small memory footprint throughput is particularly suitable when the and/or controller parameters change at runtime. The contributions of include: (i) formulation MPC problem as bounded-variable least-squares (BVNLS) problem, demonstrating use an appropriate solver can outperform industry-standard solvers; (ii) easily-implementable library-free BVNLS with proof global convergence; (iii) matrix-free method for computing products vectors Jacobians, required by BVNLS; (iv) efficient updating sparse QR factors using active-set methods problems. Thanks explicitly parameterizing optimization algorithm in terms tuning parameters, resulting inherently immediately adaptive any changes formulation. same algorithmic framework cope linear, nonlinear, variants based on broad class prediction models sum-of-squares cost functions.