W.S.G. Isaac Newton

Sioux Technologies is a global high-tech company with its headquarters located in Eindhoven. They develop, innovate and assemble high-tech systems with advanced software, mathware, electronics and mechatronics.

One of the many things Sioux works on is extreme positioning. Typically,

state-of-the-art systems mainly rely on high-quality mechatronic designs that aim to maximize stiffness and minimize non-linearity. While this approach simplifies control strategies, this comes at the expense of extremely high hardware costs. Reducing the costs of these kinds of systems while maintaining performance often means that complex controller design is the alternative. However, some of the ideal mechatronic properties like high stiffness will have to be sacrificed. The identification of the parameters of a low-cost system can be very difficult because the dynamics of the system are not fully understood. During my internship, I worked on using Residual Reinforcement Learning to further improve the performance of a low-cost positioning system. Reinforcement learning was used on top of already existing feedback and feedforward controllers, meaning that reinforcement learning only had to compensate for the residual undesired behaviour of the system. This project was carried out in the Mathware department in collaboration with the Mechatronics department.

While the project was mainly carried out individually, a supervisor and other interested people were always eager to help. Working on a multidisciplinary project that involved two departments made it possible to gain new insights into the fields of reinforcement learning and control. Furthermore, I got the opportunity to share my findings in both fields and to introduce people from both departments to each other’s expertise, which was a valuable experience.