Real-Time NMPC for Mechatronic Systems

Niels van Duijkeren

KU Leuven

Friday, July 24, 2015, 14:00 - 15:00

Room 01-014, Georges-Köhler Allee 103, Freiburg 79110, Germany

Nonlinear modelpredictive control is a control technique that enjoys an increasing interest from both academics and industry. The theoretic basis of NMPC is maturing and fast solution strategies become increasingly available. Because NMPC employs online optimization, these control algorithms can incorporate actuator constraints and more advanced functionalities such as collision avoidance. However, it remains a challenge to implement NMPC for fast and typically stiff mechatronic systems. This talk presents the progress and outlook for my PhD on the topic of “Real-Time NMPC for Mechatronic Systems” at the nine-month milestone. Traditional methods for control and motion planning of robotic manipulators is summarized along with a brief introduction to NMPC. A central topic of the ongoing work is the ABB IRB120 robotic arm, two explored strategies for time-optimal motion generation for this system are presented. This includes a bi-level programming approach to point-to-point motion planning, this strategy builds on the property that robot trajectory generation is convex if a path in joint space is known. Secondly, initial attempts towards a time-optimal NMPC approach for the robot arm are discussed. The talk continues with the planned doctoral training program and lists recent publications and outreach activities. Finally, prospects for future work (short and long term) are discussed which hopefully initiates a discussion with the audience.