< Back to previous page


Position and Orientation Tunnel-Following NMPC of Robot Manipulators Based on Symbolic Linearization in Sequential Convex Quadratic Programming

Journal Contribution - Journal Article

The tunnel-following nonlinear model predictive control (NMPC) scheme allows to exploit acceptable deviations around a path reference. This is done by using convex-over-nonlinear functions as objective and constraints in the underlying optimal control problem (OCP). The convex-over-nonlinear structure is exploited by algorithms such as the generalized Gauss-Newton (GGN) method or the sequential convex quadratic programming (SCQP) method to reduce the computational complexity of the OCP solution. However, the {modeling effort and engineering time} required to implement these methods is high. We address the problem of reducing the {modeling effort} in the implementation of SCQP, focusing on a standard sequential quadratic programming (SQP) implementation where symbolic linearization is applied to the nonlinear part of the convex-over-nonlinear functions in the objective and constraints. The novelty of this paper is twofold. It introduces a novel operator that applies symbolic linearization in a transparent and easy way to solve nonconvex OCPs with the SCQP method, and introduces a meaningful representation of an orientation-tunnel for robotic applications by means of a convex-over-nonlinear constraint, which preserves the convexity exploitation by the SCQP method. The proposed technique is demonstrated in a tunnel-following task for a 7-degrees-of-freedom manipulator.
Journal: IEEE Robotics and Automation Letters
ISSN: 2377-3766
Issue: 2
Volume: 7
Pages: 2867 - 2874
Publication year:2022