Ken-Joel Simmoteit

I’m a Ph.D. Student for Robot Learning at Tu Darmstadt, advised by Jan Peters. I will conduct research on reinforcement learning for vision–language–action (VLA) models in the context of dexterous manipulation. In addition, I aim to investigate the use of generative models for learning dynamic systems, with the objective of ensuring that predicted motions are both stable and safe.

I truly believe that autonomous robots will change our modern world and therefore I want to enable multimodal, context-aware and safe robot skills. I am also interested in generative models, reinforcement learning, vision–language–action models, learning complex dynamics and Gaussian Splatting.

I did my Master studies at the Karlsruhe Institute of Technology (KIT), studying mechatronics and information technology with a focus on robotics and control engineering. During my studies, I conducted research at the Bosch Center for Artificial Intelligence (BCAI), focusing on learning contractive robot skills with obstacle avoidance using implicitly learned surface representations. There I also developed generative models for injective mapping between task and joint space in the robot. Prior to that, I worked at the Research Center for Information Technology (FZI), investigating locomotion behavior in a six-legged robot. During this period, I also co-founded the tech startup Desoltik, aiming to enable AI- & robot-driven semiconductor reconditioning.
In addition, I applied my knowledge of control and system dynamics through voluntary work with the university group mu-uero HYPERLOOP. There, I was responsible for laying the groundwork in the non-linear levitation control, modeling the electromagnetic system and development of a sensor fusion for state estimation.