Mixed Use of Analytical Derivatives and Algorithmic Differentiation for NMPC of Robot Manipulators

Alejandro Astudillo (1) , Justin Carpentier (2) , Joris Gillis (1) , Goele Pipeleers (1) , Jan Swevers (1)

1 KU Leuven - Catholic University of Leuven = Katholieke Universiteit Leuven
2 WILLOW - Models of visual object recognition and scene understanding

In the context of nonlinear model predictive control (NMPC) for robot manipulators, we address the problem of enabling the mixed and transparent use of algorithmic differentiation (AD) and efficient analytical derivatives of rigid-body dynamics (RBD) to decrease the solution time of the subjacent optimal control problem (OCP). Efficient functions for RBD and their analytical derivatives are made available to the numerical optimization framework CasADi by overloading the operators in the implementations made by the RBD library Pinocchio and adding a derivative-overloading feature to CasADi. A comparison between analytical derivatives and AD is made based on their influence on the solution time of the OCP, showing the benefits of using analytical derivatives for RBD in optimal control of robot manipulators.

Mots clés: Analytical derivatives; Algorithmic differentiation; Robot manipulators; Optimal control; Predictive control

Domaines: Automatique / Robotique

Alejandro Astudillo, Justin Carpentier, Joris Gillis, Goele Pipeleers, Jan Swevers. Mixed Use of Analytical Derivatives and Algorithmic Differentiation for NMPC of Robot Manipulators. MECC 2021 - 1st IFAC Modeling, Estimation and Control Conference, Oct 2021, Austin, United States. ⟨hal-03541487⟩