Verkennend Onderzoek naar Predictieve Controle en Schatting voor Sensorgebaseerde Robotica: Toepassingen op het Volgen van Oppervlakken en het Vormgeven van Kabels

Robot manipulators perform well in structured environments but face challenges in unstructured settings due to disturbances caused by changes in the environment. While traditional sensor-based robot control allows the robot to react to disturbances, tracking errors can occur when disturbances are not properly estimated and compensated for. Furthermore, traditional sensor-based robot control typically relies on instantaneous controllers, which are susceptible to local minima and struggles to remain within the limits of the robot, such as actuator limits. To address these challenges, this research explores the use of predictive control and estimation. First, an estimator formulation is proposed which can be easily integrated into existing constraint-based robot task specification frameworks. This approach was validated on a force-controlled contour-following task, where it was used to estimate and compensate for disturbances occurring during task execution, resulting in reduced tracking errors and improved contour-following speed. Next, the use of Model Predictive Control (MPC) is explored for two sensor-based tasks—surface following and cable shaping. In the case of the surface-following application, the main contribution lies in analysing the impact that the choice of surface model has on performance, as well as proposing a novel surface model based on radial basis functions. Moreover, an estimation scheme is provided which can estimate the unknown states of the proposed surface model in real-time. In the case of the cable-shaping application, the main contribution lies in proposing a real-time-feasible MPC formulation that can be used to improve the reliability and execution speed compared to state-of-the-art approaches. For both applications, the improved performance that results from using MPC is clearly demonstrated, highlighting its potential to improve the performance of sensor-based robot applications in various domains.