Optispline: a toolbox for robust (dynamic) optimization using spline relaxations

Joris Gillis

KU Leuven

Tuesday, December 12, 2017, 11:00

"Room 02-012, Georges-Köhler Allee 102, Freiburg 79110, Germany"

Building on the CasADi infrastructure, we created an abstraction for the differentiation and evaluation of N-dimensional symbolic BSplines.

By parametrizing a function as a linear combination of positive basis functions (bspline), we can transcribe infinitely-dimensional positivity constraints on a domain to a finite set of positivity constraints on bspline coefficients.

In this talk, we demonstrate how this technique can be applied to 1) path constraints in a continuous-time optimal control problem, 2) robustifying constraints in a non-linear program, and 3) construction of smooth interpolating functions with desirable properties.

(Joint work with Goele Pipeleers, Wannes Vanloock, and Erik Lambrechts, MECO, KULeuven)