Short biography

I started my bachelor’s degree in Engineering Science at the Technical University of Munich in 2013 where I gained broad interdisciplinary competencies with particular focus on mechanical, electrical and software engineering based on a strong mathematical foundation. In my bachelor’s thesis, I worked on improving 3D vision in challenging light scenarios using high dynamic range imaging. I continued at the TUM with the master degree in Robotics, Cognition, Intelligence, where I focused on learning about robotic dynamics and motion planning, on computer vision and on deep learning. My master’s thesis investigated an efficient learning-based method to generate adaptable collision-free trajectories for robotic task and motion planning. The results were published in the 2020 IEEE-RAS International Conference on Humanoid Robots, where I also presented an interactive iPoster. During my studies I was actively seeking a connection to the industry by working part-time at Infineon and Accenture and full-time at BMW during a six-month internship. Additionally, I enjoyed getting to know other cultures during my studies abroad at Sun Yat-Sen University in Guangzhou, China and at the Queensland University of Technology in Brisbane, Australia.

Short description of project objectives

This PhD project aims at connecting the research areas of robotic task and motion planning with computer vision and 5G communications. In a development towards shorter product life cycles and higher product customizations, Industry 4.0 pursues an efficient production down to a batch size of one. However, current industrial robots are configured in contained spaces separated from human operators to produce high-volume outputs under minimal uncertain conditions. More flexible robotic systems, such as autonomous mobile robots (AMRs), will be required to handle uncertain environments that are hard to model. For instance, the behavior of human operators or contact-rich interactions in assembly tasks. Essential for these AMRs is the ability to quickly process sensor inputs and compute reasonable actions in a closed-loop fashion.

To this end, 5G is expected to provide powerful wireless communication to enable fast and reliable data transmissions at high sample rates. This ultra-reliable low-latency communication (URLLC) ability of 5G networks also introduces the opportunity to offload heavy computations from an AMR to an edge server, called multi-access edge computing (MEC). It addresses the limited computational resources onboard an AMR. Complex tasks demand performing computationally expensive operations, such as image processing or deep learning algorithms, which are faster computed on a more powerful edge server. One main objective of this work is to compare onboard and edge computing solutions, as well as edge computing solutions using 5G, 4G and wifi networks and analyse their performance in autonomous and collaborative robotic experiments which will result in a quantified assessment of the performance differences on productivity and human safety in industrial scenarios. Afterwards, this work aims at contributing to robotics methods that benefit from 5G-based edge computing.