I work on robot learning: reinforcement learning that transfers from simulation to hardware, vision-language-action models, and the data systems that carry learned skills onto real robots.
Research
One question runs through all of it: how do robots acquire skills that survive contact with the real world? I approach it from four directions.
Training dexterous, contact-rich manipulation policies in massively parallel simulation, then closing the reality gap with domain randomisation and distillation into vision-based policies that run on real hardware.
Generalist policies that map pixels and language to actions — and the infrastructure behind them: synchronised multi-camera capture, robot-state logging and dataset tooling for training at scale.
Intuitive interfaces, from hand tracking to XR, that let people demonstrate dexterous skills, with retargeting from human hands to multi-fingered robot hands.
High-fidelity twins of real work cells with real-to-sim calibration of dynamics and sensors, so learned policies meet the precision and reliability that production demands.
The loop
Simulation and reality, feeding each other.
Simulate
Calibrated digital twins and parallel environments
Learn
RL at scale, distilled into deployable policies
Demonstrate
Human teleoperation for skills simulation can't capture
Deploy
Policies on real hardware — and data back into the loop
Work
Published research and the tools I build in the open, for myself and for other roboticists.
Paper · ITSC 2024Weakly-supervised Sim2Real for roadside 3D detection
Inspect, visualize and author URDF / xacro models inside VS Code
A keyboard-driven TUI cockpit for ROS 2
A bilingual editorial chronicle — from Greek myth to embodied AI
About
I'm a Senior AI Engineer at Agile Robots in Munich, where I work on robot learning end to end — from simulation and training to data collection and deployment on real hardware.
I studied Robotics, Cognition, Intelligence at TUM, after a degree in vehicle engineering at Tongji. My master's research on weakly-supervised sim-to-real perception was published at IEEE ITSC. In my own time I build open-source tools for roboticists.
Senior AI Engineer
Agile Robots SE
Robot learning across sim-to-real reinforcement learning, vision-language-action models and teleoperation — from data infrastructure to policies on real hardware.
Robotics Software Engineer
Agile Robots SE
Manipulation, motion planning and simulation tooling for industrial robot systems.
Technical University of Munich
M.Sc. · Robotics, Cognition, Intelligence
Deutschlandstipendium
Tongji University
B.Eng. · Vehicle Engineering
Excellent Graduate
Toolkit
Contact
I'm happy to talk about robot learning, sim-to-real, research collaborations or open-source tooling.