M.S. @ Duke MEMS · Research Assistant @ General Robotics Lab

Hi, I'm Shucheng

I'm a robotics researcher, a curious builder, and someone who loves a good challenge.

Having grown up in China and studied in Japan and the United States, I enjoy exploring different perspectives and pushing my limits, whether I'm building robots in the lab or training at the gym.

  • Robot Learning
  • Sim-to-Real
  • Humanoid & Legged Robots
  • Safe Control

Bio

I'm pursuing my master's degree in Mechanical Engineering and Materials Science at Duke University, where I work as a research assistant in the General Robotics Lab under Prof. Boyuan Chen. My research focuses on reinforcement learning and sim-to-real robotics, from teaching a Unitree G1 humanoid to dance with a human partner, to building locomotion controllers for custom legged robots.

Before Duke, I earned my B.Eng. in Information Science and Systems Engineering from Ritsumeikan University in Japan. There, I worked with Prof. Wenbin Liu and Dr. Mikhail Svinin on safety-critical control for multi-robot navigation using control barrier functions, which led to a paper at IEEE ICRAS 2025.

News

  • Oct 2026CoDance is on arXiv and under review at ICRA 2027.
  • May 2026Joined the General Robotics Lab at Duke as a research assistant.
  • Sep 2025SE-CBF published at IEEE ICRAS 2025.
  • Aug 2025Started my M.S. at Duke University.
  • Feb 2025Graduated from Ritsumeikan University (Saionji Memorial Scholarship).

Publications

CoDance: Learning Reactive and Compliant Human-Humanoid Interaction from Video

Zhuoqun Chen, Shucheng Jia, Boyuan Chen

Under review at ICRA 2027

From a single video of two human dancers, a Unitree G1 humanoid learns partnered dancing: it follows a moving partner's footsteps while keeping two-hand contact and yielding compliantly to interaction forces, sustained over repeated forward and backward motion with a real person.
  • Built an asymmetric actor-critic PPO pipeline with a privileged critic and decoupled upper/lower-body rewards.
  • Designed domain randomization from measured sensing bias and communication latency.
  • Deployed the policy on Unitree G1 with real-time motion-capture observations.
  • PPO
  • Humanoid
  • Sim-to-Real
  • Unitree G1
  • Motion Capture

Super-Ellipse Control Barrier Functions for Enhanced Obstacle Avoidance in Complex Multi-Agent Environments

Wenbin Liu, Shucheng Jia, Mikhail Svinin

IEEE ICRAS 2025

Safe multi-robot navigation with control barrier functions that use super-ellipse boundaries instead of circles, so robots can pass rectangular obstacles without overly conservative detours. I built the overhead-camera perception pipeline (OpenCV, pixel-to-world mapping) and the CBF-QP controller in ROS. In a dense multi-robot intersection, task completion time dropped from 123 s to 62 s versus standard CBFs, in Gazebo and on real robots.
  • CBF
  • Quadratic Programming
  • ROS
  • Multi-Robot
  • OpenCV

Projects

Autonomous Dual-Arm Tic-Tac-Toe

Personal project, Duke · Fall 2025

Two Kinova Gen3 Lite arms see the board with an RGB-D camera, choose their moves, and take turns picking and placing pieces. I built the full pipeline: RGB-D perception, rule-based game logic in Python, ROS 2 integration, and MoveIt 2 pick-and-place. About 90% task success in Gazebo and 50% on the real arms.
  • ROS 2
  • MoveIt 2
  • RGB-D Perception
  • Gazebo
  • Kinova Gen3 Lite

Pneumatic Soft-Finger Robotic HandTeam of 3

MEMS Capstone, Duke · 2025

A three-finger silicone soft hand with closed-loop grasp-force control. I led the control design: a two-phase (approach / stabilize) dual-pump PWM controller on an Arduino Mega using fingertip force, bend and chamber-pressure feedback, with safety supervision for pressure limits and release.
  • Feedback Control
  • Soft Robotics
  • Arduino
  • Sensor Fusion

Ongoing research at GRL

Unpublished; details available upon request.

  • Robot design and sim-to-real RL locomotion for custom-built legged robots.
  • Perception-guided dual-arm manipulation for laboratory automation.

Education

Duke University, Durham, NC

M.S. in Mechanical Engineering & Materials Science · GPA 4.0/4.0

Robotics, Control Theory, Medical Robotics, Data Science & Machine Learning

Aug 2025 – May 2027 (expected)

Ritsumeikan University, Osaka, Japan

B.Eng. in Information Science & Systems Engineering · GPA 3.7/4.0

Saionji Memorial Scholarship for academic excellence

Apr 2021 – Feb 2025

Skills

Reinforcement learning
PPO, asymmetric actor-critic, reward design, domain randomization, sim-to-real transfer, legged locomotion, Isaac Gym, MuJoCo, URDF
Control & planning
Control theory, MPC, control barrier functions (CBF), quadratic programming, PD & closed-loop control, motion planning, robot navigation & obstacle avoidance
Robotics software
ROS / ROS 2, MoveIt 2, Gazebo, TF & coordinate calibration, motion-capture integration
Perception & ML
OpenCV, YOLOv8, RGB-D & point clouds, FoundationPose, SAM, AprilTag, PyTorch, TensorFlow / Keras, scikit-learn, CNNs
Robot platforms
Unitree G1 humanoid, Kinova Gen3 Lite, custom servo-driven biped & quadruped, mobile robots
Hardware & fabrication
SolidWorks, Fusion, 3D printing, laser cutting, silicone casting, Arduino, servos, IMUs, sensors & actuators, rapid prototyping
Programming
Python, C++, C, MATLAB / Simulink, Java, Linux
Languages
Chinese (native), Japanese, English