Nonlinear Model Predictive Control in the Application of Constrained Manipulator Control

The goal of this thesis is to evaluate model predictive control (MPC) in the context of constrained redundant robot manipulators. At first MPC is applied to linear systems and compared with traditional finite-horizon optimal control. In the second part the model predictive- and finite-horizon kinematic control strategy proposed by Schuetz [1] is implemented. Additionally an alternative MPC and finite horizon control algorithm is developed based on the shooting method. To contrast these methods with the widely-used 'instantaneous' kinematic control, a simulation is carried out, where a redundant manipulator is controlled to perform obstacle avoidance. In this simulation it is shown that Schuetz' algorithm only approximates the optimal solution. This is achieved by demonstrating that the newly suggested method yields lower accumulated cost-values for the finite horizon than Schuetz' method.