Arnold’s academic background and research mainly cover the areas of human pose estimation, motion analysis, and motion capture. He began with his bachelor’s thesis on the transfer of human movements to non-human models. He continued his work in the field of motion capture in his master’s thesis, in which he worked on improving a model for estimating human pose with a single RGB-D camera by using data from a second camera as additional input. In his past research, he also investigated the effects of faulty labels in state-of-the-art datasets and benchmarks for estimating human posture from RGB images, and developed a markerless real-time multi-view motion capture system.
Currently, his doctoral research focuses on hand tracking as well as contact and force estimation. A central focus of his work is enabling humanoid robotic hands to learn by demonstration. To achieve this, he works closely with the robotics department and investigates models and strategies for capturing the required data.