Modelvoorspellende controle en schatting voor robotprogrammering op basis van beperkingen: Praktische benaderingen voor vereenvoudiging van het ontwerp

Although robot manipulators excel in highly structured environments, they encounter difficulties in less structured environments due to the greater variety of conditions and potential disturbances. To enable robots to operate in these environments, previous work introduced a constraint-based robot programming approach, in which robots are programmed to react to changes in the environment based on sensor measurements. Although this approach works well, the commonly used controllers are purely instantaneous and struggle to predict future constraint violations, leading to suboptimal performance. To address this issue, this research investigates two key improvements: the integration of Model Predictive Control (MPC) and a disturbance estimation framework. First, the use of MPC is explored for a surface-following task, in which the robot maintains a specified distance and orientation relative to an unknown surface using laser sensor data. A novel MPC formulation is proposed that preserves the familiar tuning properties of instantaneous controllers while benefiting from a prediction horizon. This approach demonstrated improved performance in real-world experiments and aims to enable easier adoption and tuning in practical applications. Second, a method is introduced to estimate disturbances and position uncertainties within an existing constraint-based control framework. This estimation approach is validated on a force-controlled contour-following task, where it effectively reduces force-tracking errors and increases task execution speed by compensating for external disturbances. Together, these contributions enhance the performance of constraint-based robot programming, further enhancing the use of advanced sensor-based control strategies in unstructured environments.