About
I'm a robotics software engineer based in Munich, working at the intersection of reinforcement learning, dexterous manipulation and physical simulation.
My focus is closing the loop between high-fidelity simulation and the messy real world — the perception, control and policy stacks that let humanoid and manipulator systems act with intent.
Selected work
Weakly-supervised Sim2Real for roadside 3D detection
A weakly-supervised Sim2Real domain adaptation framework for monocular 3D object detection on roadside infrastructure. Closes the simulation-to-reality gap with synthetic supervision and a domain-adaptive teacher–student loop.
Inspect, visualize and author URDF / xacro models inside VS Code
A VS Code extension that brings interactive URDF and xacro visualization, joint control, semantic authoring (SRDF) and reachability analysis directly into the editor.
A keyboard-driven TUI cockpit for ROS 2
Rosight = ROS + sight. A terminal cockpit that lets you browse the live ROS 2 graph — messages, nodes, services, actions, parameters, TF, bags, interfaces — and live-plot any numeric field, all from one screen without a mouse.
Experience
Robotics & AI Engineer
Agile Robots SE · Munich, Germany
Robot learning across reinforcement learning, dexterous manipulation and physical simulation. Building the policy and infrastructure layers behind the next generation of humanoid systems.
Working Student · Robotics & AI
Agile Robots SE · Munich, Germany
Robot manipulation and motion planning. Bridged research prototypes and production-bound robot stacks.
Technical University of Munich
M.Sc. · Robotics, Cognition, Intelligence
Top of class · Deutschlandstipendium
Tongji University
B.Eng. · Vehicle Engineering
Distinguished Graduate
Publications
Capabilities
Reinforcement learning, imitation learning and policy training for manipulation and locomotion.
Dexterous manipulation, motion planning and low-level control on real and simulated platforms.
3D object detection, Sim2Real domain adaptation, Visual SLAM and scene understanding.
High-fidelity physical simulation pipelines for training, validation and synthetic data.
Deep learning frameworks and the model-training stack behind modern robotic perception and policy learning.
Production-grade C++/Python, Linux toolchains, and the glue between research and product.
Let's talk
