I am a PhD candidate at Prof. Schoellig’s lab at Technical University of Munich. My main interests include control for complex systems, reinforcement learning, end-to-end learning, and robotics. Previously, I completed a M.Sc. degree in Robotics at TUM, and wrote my master’s thesis with Prof. Ijspeert. I was at Robert Bosch GmbH and in the Formula Student. Member of AI Grid (BMFTR).
Tutorial marking ten years of CDC Learning-Based Control sessions, covering stability and safety foundations, dynamics learning, data-driven control, and RL/IL/BO, with open-source examples.
Multi-view latent action pretraining recovers 6D actions that single views miss, lifting hardware success rate from 21% to 46% and enabling transfer from human videos.
55 scene-aligned 4D human motion sequence dataset from egocentric wearables for training perceptive humanoid locomotion policies on challenging terrain.
Expensive robot data for the contact-rich segment combined with robot-data-free methods for free-space motion enables success rates of >95% and robustness to OOD.
Functionally decomposing locomotion into constraints, energy-efficiency, task reward and perception enables efficient locomotion without explicit gait priors.
C++ Controllers, Python Interfaces, and Gymnasium environments for teleoperation and deploying learning-based policies on manipulators.
A single gait cycle of expert demonstration significantly improves achieved reward and visual appearance of the learned gait in locomotion tasks - even for transfer tasks.
Inter-GPU communication compression for Large Language Models leads to up to 2x reduction of time-to-first-token with negligible model performance degradation.
Lampreys can locomote using only based on local sensory feedback loops. Similar neural network designs show favorable properties for robot learning.
3D-printed chip for mechanoactivation at single-cell level to study intracellular calcium signaling and translocation.
Art history diagrams that organize the historic movements and trends in art, from Barr’s 1936 chart to a data-driven map of 375 movements and 55,000 artists.
How can humans control robots more intuitively? We explore seven projects in which artists use robots through various AI-enabled human-machine interfaces.
First Scores and Tabs for the Guitars in Sexual Healing on the Internet.
A Franka Emika FR3 robotic manipulator builds lego structures.
Teaching and studying tool for TU Dresden with 3500 users. Awarded with the german Digital Changemaker Award.
Building a robot for autonomous outdoor exploration and artifact detection
Presented at KIRO 2024, the first German AI-in-robotics conference. Awarded for the work of my master’s thesis.
Check out the Huberman Lab podcast for a healthy lifestyle and science-based tools for everyday life. The Lex Fridman podcast for broadening your worldview and socially concerning topics. I like to look at DailyArt.
