CV

Contact Information

Name Zewen Yang
Professional Title Senior Research Scientist
Email zewenreal@gmail.com

Professional Summary

My mission is to turn advances in machine learning into safe, reliable systems that solve real-world problems. I am a Senior Research Scientist at Agile Robots in Munich, working at the intersection of machine learning, robotics, and control. I develop Vision-Language-Action models, World Action Models, generative motion-planning methods, and agentic AI frameworks that enable robots to perceive, reason, plan, and act. My research has been published in leading venues across machine learning and multi-agent AI (IEEE TNNLS, AAAI, AAMAS), robotics (IEEE T-RO, ICRA, IROS), and control (IEEE TAC, CDC, ACC, L4DC). Before joining Agile Robots, I worked on robot learning and teleoperation at the Munich Institute of Robotics and Machine Intelligence at TUM, and applied machine learning to public health research at the Robert Koch Institute. From public health to autonomous robots, the common thread in my work is turning promising machine learning methods into solutions for consequential real-world problems. These experiences reinforced my belief that meaningful innovation requires both scientific rigor and engineering depth. I enjoy working across the full path from framing research questions and developing rigorous methods to building and evaluating systems in practice. I also care deeply about education, mentoring, and exchanging ideas with passionate researchers, engineers, and students.

Experience

  • 2026 - present

    Munich, Germany

    Senior Research Scientist
    Agile Robots SE
    • Own end-to-end development of expert VLA policies for an industrial dust-cup removal task on semi-humanoid robots with multi-camera perception, from data collection through on-robot deployment
    • Built the post-processing and cleaning pipeline that turns several thousand real-robot demonstrations into training and benchmark datasets, with training on a GPU cluster and experiment tracking in Weights & Biases
    • Train, benchmark, and deploy in-house and state-of-the-art VLA policies on robot hardware; the final policy reached 99% task success over 100 real-robot trials
    • Led a 4-person cross-functional team to deliver a humanoid interaction system shown live at NVIDIA GTC 2026, owning the overall pipeline and motion planning with VLM agents decomposing high-level goals
    • Defined the research roadmap on distributed flow-matching policies for robot motion generation and mentor 2 student researchers on an agentic framework for long-horizon manipulation tasks
  • 2025 - 2025

    Munich, Germany

    Postdoctoral Researcher
    Munich Institute of Robotics and Machine Intelligence (MIRMI), TUM
    • Teaching Curriculum: Advanced Robot Control and Learning
    • Federal Ministry of Education and Research, Germany — 6G-life: 6G Communication Networks with a Focus on Human-Machine Collaboration (€70 Million)
      • Proposed safe motion planning of robots with generative models (Diffusion/Flow Matching models)
      • Developed algorithms of cloud-based high-frequency control for bilateral teleoperation
      • Designed novel one-shot imitation learning algorithms for robotic systems
  • 2023 - 2024

    Berlin, Germany

    Postdoctoral Researcher
    Robert Koch Institute — Center for Artificial Intelligence in Public Health Research
    • Teaching Curriculum: Advanced Data Visualization for (Explainable) AI
    • Federal Ministry of Health, Germany — AI-DAVis-PANDEMICS: AI-supported Data Analysis & Visualization of Pandemics (€1.78 Million)
      • Built a real-time audience-centric web app for pandemic surveillance with ML-based data analysis
      • Proposed a nested framework for ML models’ design and validation considering AI regulations
      • Designed interpretable molecular encodings and representations for machine learning tasks
  • 2022 - 2023

    Berlin, Germany

    Data Scientist
    Data Science Retreat
    • Statistical Learning: evaluated time series models via MLflow for forecasting the spread of disease; built a churn prediction model with data pre-processing and feature engineering
    • Deep Learning: real-time object detection, action detection and gesture analysis for sign language recognition; trained image classification models using vision transformers
    • Natural Language Processing: extracted and visualized character relationships via named entity recognition; used node and word embeddings to recommend research collaborators; fine-tuned language models for segmentation and text-generation tasks with GPT
  • 2019 - 2022

    Munich, Germany

    Research Associate
    Technical University of Munich — Chair of Information-oriented Control
    • Supervised 10+ student theses, several achieving the highest German grade (1.0)
    • Teaching Curriculum: Advanced Control and Robotics
    • European Union’s Horizon 2020 — SeaClear: Search, Identification and Collection of Marine Litter with Autonomous Robots (1.0: €5 Million, 2.0: €9 Million)
      • Developed a data-driven model for Autonomous Underwater Vehicles (AUVs) with uncertainty quantification
      • Created a simulation platform for conducting control tests of AUVs
      • Contributed to hardware tests at Marseilles and Hamburg ports (e.g., bottle grabbing and vehicle recycling)
    • European Union’s Horizon 2020 — REHYB: Rehabilitation Based on Hybrid Neuroprosthesis (€7 Million)
      • Built a digital twin for real-time monitoring of human motion
      • Detected abnormal movements using deep neural networks with multiple sensors
      • Investigated the safe control of human muscles by FES to achieve functional movements
  • 2017 - 2022

    Harbin, China

    Principal Investigator / Doctoral Researcher
    Harbin Engineering University — College of Intelligent Systems Science and Engineering
    • Ph.D. Student Research & Innovation Fund of Fundamental Research Funds for Central Universities: Research on the Coordinated Control Method of AUV Swarm (¥70k)
      • Led a research team and completed the project with top-tier publications
      • Proposed novel formation control methods for AUV swarm using the consensus algorithm
    • National Natural Science Foundation of China & National Key Innovation Project: Environmental Prediction & Task Reconfiguration of AUV for Covert Operations in the Coastal Sea (¥600k); Overall Technology Scheme of the Multiple AUVs System (¥50 Million)
      • Developed advanced control algorithms for underactuated AUVs
      • Investigated obstacle avoidance and recovery docking approaches for AUVs
      • Authored a chapter titled “Reliability Analysis & Design Technology of AUV” in Control Technology of AUV
  • 2016 - 2017

    Suzhou, China

    Test & Analysis Engineer
    Suzhou Electrical Apparatus Science Academy Co., Ltd.
    • Implemented experiment tests to analyze various characteristics of fiber-optic current transformers
  • 2014 - 2014

    Harbin, China

    Automation Engineer
    Harbin Turbine Co., Ltd.
    • CNC lathe programming, quality inspection

Education

  • 2019 - 2022

    Munich, Germany

    Ph.D.
    Technical University of Munich (TUM)
    Machine Learning and Control Theory
  • 2017 - 2019

    Harbin, China

    Ph.D.
    Harbin Engineering University
    Control Science & Engineering
  • 2014 - 2017

    Harbin, China

    Master of Engineering
    Northeast Forestry University
    Mechanical and Electrical Engineering

Skills

AI Tools: Pi, Cursor, Claude Code, Codex, Perplexity
Programming & Markup Languages: Python, Matlab & Simulink, Julia, C++, JavaScript, TypeScript, SQL, HTML, CSS, LaTeX, Markdown
Machine Learning: PyTorch, JAX, TensorFlow, Keras, Hugging Face, scikit-learn, statsmodels, Unsloth, Transformers/Tokenizers/PEFT/TRL (Hugging Face), openai, spaCy, NLTK, Gensim
Robotics: Isaac Sim/Lab, MuJoCo, ROS, Gazebo, LeRobot, OpenCV, Qualisys
Web & Front-end: Next.js, Vite, React, Tailwind CSS, shadcn/ui, Aceternity UI, Figma
Back-end & Database: Django, Flask, FastAPI, Postman, Selenium, MySQL, PostgreSQL
Visualization: D3.js, Vega-Lite/-Altair, Observable, Plotly, Matplotlib, seaborn, GeoPandas, NumPy, Pandas
Cloud & Hosting Tools: Git, GitHub, GitLab, Docker, AWS, Streamlit, Render, Heroku, Colab, Overleaf, Notion, Obsidian, Jira

Languages

English : Fluent
Chinese : Native
German : Basic

Publications

Awards

Interests

Fitness: Strength Training, Running, Hiking
Movies: Sci-Fi, Documentary, Action, Comedy, Drama
Music: R&B, Hip-Hop, Electronic