Lukas

Hi!
I'm Lukas Chen,

I'm an MSc Robotics student at the University of Manchester. Before that, I graduated from the University of Edinburgh with a First Class BSc in Artificial Intelligence & Computer Science, where my dissertation looked at the Sim2Real gap in reinforcement learning for contact-rich robotic manipulation. For four years I was a software developer on Edinburgh University Formula Student's autonomous driving team, which won the AI division at Formula Student UK three years in a row.

View my CV

About me

Education

  • MSc Robotics University of Manchester · Sep 2026 – Sep 2027
  • BSc Artificial Intelligence & Computer Science University of Edinburgh · Sep 2022 – Jul 2026 1st Class with Honours Dissertation: Sim2Real Gap in Reinforcement Learning for Robotic Contact-rich Manipulation Read my dissertation

Experience

  • The University of Manchester
    Research Intern, University of Manchester VLA + tactile sensing for humanoid manipulation · 2026 – present
  • Edinburgh University Formula Student
    Software Developer, Edinburgh University Formula Student (EUFS) Sep 2022 – Jul 2026
  • The University of Edinburgh
    Teaching Assistant, University of Edinburgh Jan – Apr 2026
  • The University of Edinburgh
    Summer Research Intern, IPAB, University of Edinburgh Jun – Sep 2025

Skills

Programming languages

  • Python
  • C++
  • C
  • Java
  • JavaScript
  • SQL
  • MIPS
  • PDDL
  • Haskell

Robotics & machine learning

  • ROS 2
  • MuJoCo
  • Isaac Sim
  • PyBullet
  • PyTorch
  • TensorFlow
  • SKRL
  • NumPy
  • MATLAB

Software & tools

  • Linux
  • Docker
  • Git
  • Fusion
  • React
  • Spring
  • Redis
  • Kafka

Languages

  • English (Fluent)
  • Chinese (Native)
  • Cantonese (Native)

Research interests

  • Reinforcement learning for robotics
  • Sim2Real transfer
  • Contact-rich manipulation
  • State estimation & SLAM
  • LLM agents for robotics

My projects

  • VLA + Tactile for Humanoid Manipulation

    Current Research internship · University of Manchester

    Building a vision-language-action (VLA) model that gives a humanoid robot a sense of touch. Using the iCub's stereo vision and tactile signals, the work targets tabletop manipulation and learning actions from human demonstrations, aiming for robots that perceive and act more like humans.

    • VLA models
    • Tactile sensing
    • Humanoid robots
    • iCub

    Supervised by Dr. Wenjie Huang and Dr. Fangjun Li

    View project page

  • Robotic Sim2Real for Non-Prehensile Manipulation

    Robotic Sim2Real for Non-Prehensile Manipulation

    BSc dissertation · supervised by Prof. Sethu Vijayakumar

    Trained reinforcement learning agents for non-prehensile manipulation in MuJoCo and Isaac Sim, and built pipelines for training across multiple simulators. The research looks at how the structure of the dynamics model affects Sim2Real transfer.

    • Reinforcement learning
    • MuJoCo
    • Isaac Sim
    • Sim2Real
    Read my dissertation

  • Autonomous Driving System for EUFS

    Edinburgh University Formula Student · 2022 – 2026

    Core developer on the ROS-based software that drives the team's autonomous race car, which won the AI division at Formula Student UK three years running (2023–2025). Led the 8-person Localisation & Mapping sub-team, implemented real-time state estimation (FastSLAM, EKF, particle filter), and built CI/CD pipelines that halved code review time.

    • ROS
    • SLAM
    • State estimation
    • CI/CD

  • MuJoCo MCP

    MuJoCo MCP

    An MCP server that exposes a gym-like API as skills, so LLM agents can interact with physics simulations.

    • MuJoCo
    • MCP
    • LLM agents
    • Python
    View on GitHub

  • Dual-Arm Motion Planning & Control

    Dual-Arm Motion Planning & Control

    Advanced Robotics coursework · University of Edinburgh

    Got a dual-arm robot to grip a cube with both hands and carry it to a target without hitting obstacles. Grasp poses come from optimisation-based inverse geometry, the route from a bi-directional RRT (RRT-Connect), and the motion is smoothed into a Bézier trajectory that an inverse-dynamics controller follows in PyBullet.

    • Motion planning
    • RRT*
    • Inverse Kinemetic
    • Pinocchio
    • PyBullet

    Pair project

    View on GitHub

Contact me