Robotic systems suffer from inaccuracies stemming from a variety of sources, such as manufacturing errors, backlash, thermal expansion and compliance, limiting their absolute positioning accuracy. While a number calibration methods exist to address these issues for articulated robots, they do not extend to the more intricate case of large prismatic joints. In this project, we develop a robot model for gantry systems enabling us to perform effective end-to-end calibration.

Plasma coating is a process in which a protective layer is produced by spraying metal or ceramic powder onto a workpiece at immense temperature and velocity, typically through a robot. To achieve a high coating quality, the process puts strict requirements on the tool path, leading to very fast robot movements. The goal of this project to develop a path planning algorithm that finds an optimal toolpath, ensuring uniform coating thickness while respecting any constraints on the process parameters and robot acceleration.

This research introduces a novel robot-agnostic calibration method to enhance the positioning accuracy of articulated robots. The proposed robot model is based on the product of exponentials formula and allows to compensate for geometric and backlash errors as well as for thermal drift. It reduces the average positioning error to 0.15mm.

The success of robotic milling is dependent on the robot path accuracy. This work explores an iterative learning control approach for simple robotic milling operations, leveraging the high repeatability of the robot. A surface flatness better than +-50um and holes with a positional error below 25um were obtained during the processing of an aluminium workpiece.

This research addresses the challenge of time-consuming robot teaching methods by introducing a CAD 6D space mouse combined with an IMU as an input device. Implemented in Python, the method offers robot-agnostic, intuitive, and real-time control at the joint level, effectively managing singularities and joint limits. User studies demonstrate a twofold improvement in accuracy and speed compared to existing solutions.

In this work we examined teaching dynamic and complex time dependent movements found for example in painting or disassembling applications using the wand teaching approach and a Creaform C-Track industrial tracker. The robot-agnostic method, implemented in Python controlling the robot at the joint level, demonstrates promising performance in various benchmarks, identifying core challenges and proposing novel solutions for future exploration.

In wire arc additive manufacturing, achieving the desired final workpiece geometry relies on depositing layers of weld beads, with each bead altering the current geometry and setting the boundary for subsequent depositions. We devised Deep Learning and numerical optimization models. These models predict the evolution of the workpiece based on the current geometry and process parameters, offering valuable insights for simulation, planning, and control.