A Simple and Efficient Algorithm for Nonlinear Model Predictive Control

摘要

We present PANOC, a new algorithm for solving optimal control problemsarising in nonlinear model predictive control (NMPC). A usual approach to thistype of problems is sequential quadratic programming (SQP), which requires thesolution of a quadratic program at every iteration and, consequently, inneriterative procedures. As a result, when the problem is ill-conditioned or theprediction horizon is large, each outer iteration becomes computationally veryexpensive. We propose a line-search algorithm that combines forward-backwarditerations (FB) and Newton-type steps over the recently introducedforward-backward envelope (FBE), a continuous, real-valued, exact meritfunction for the original problem. The curvature information of Newton-typemethods enables asymptotic superlinear rates under mild assumptions at thelimit point, and the proposed algorithm is based on very simple operations:access to first-order information of the cost and dynamics and low-cost directlinear algebra. No inner iterative procedure nor Hessian evaluation isrequired, making our approach computationally simpler than SQP methods. Thelow-memory requirements and simple implementation make our method particularlysuited for embedded NMPC applications.