Tuesday, December 03, 2019, 11:30
Room 01-012, Georges-Köhler-Allee 102, Freiburg 79110, Germany
Parallel computing for the optimization of nonlinear model predictive control (NMPC) faces the problem of lacking dedicated algorithms for multi-core processors and the challenge in designing parallel programs. This talk presents a highly parallelizable Newton-type algorithm with a superlinear rate of convergence and its corresponding toolkit that can automatically generate parallel C/C++ code for real-time Linux. The efficiency and low latency of the introduced parallel method are shown with numerical experiments including a quadrotor reference tracking example and a 7-DOF robot manipulator path following example.