Pingcheng Jian 简平诚

I am a final year PhD student in the ECE department at Duke University. I'm fortunate to be co-advised by Professor Michael Zavlanos and Professor Boyuan Chen.

Before joining Duke, I obtained my bachelor of engineering from the Department of Automation at Tsinghua University.

Email  /  Resume  /  Research Statement  /  Transcript  /  Google Scholar  /  Github

profile photo
Research

My area of research is robot learning, and my previous publications cover the 3 topics:
1) Applying foundation models for the preference feedback in reinforcement learning,
2) Modularity for robot learning,
3) Adversarial learning for robust control.

Journal Papers
Perception Stitching: Zero-Shot Perception Encoder Transfer for Visuomotor Robot Policies
Pingcheng Jian, Easop Lee, Zachary I. Bell, Michael M. Zavlanos, Boyuan Chen
Transactions on Machine Learning Research (TMLR), 2024
project website / arXiv / OpenReview / video /

We present Perception Stitching that enables strong zero-shot adaptation to large visual changes by directly stitching novel combinations of visual encoders.

Conference Papers
LAPP: Large Language Model Feedback for Preference-Driven Reinforcement Learning
Pingcheng Jian, Xiao Wei, Yanbaihui Liu, Samuel A. Moore, Michael M. Zavlanos, Boyuan Chen
In submission to Robotics: Science and Systems (RSS), 2025
pdf /

We present LAPP (LLM-Assisted Preference Prediction), a novel framework that leverages LLMs to generate preference feedback from raw state-action trajectories that is used to guide reinforcement learning (RL) agents.

Policy Stitching: Learning Transferable Robot Policies
Pingcheng Jian, Easop Lee, Zachary Bell, Michael M. Zavlanos, Boyuan Chen
Conference on Robot Learning (CoRL), 2023
project website / arXiv / video / CoRL talk /

We propose Policy Stitching, a novel framework to facilitate multi-task and multi-robot transfer.

Adversarial Skill Learning for Robust Manipulation
Pingcheng Jian*, Chao Yang*, Di Guo, Huaping Liu, Fuchun Sun
International Conference on Robotics and Automation (ICRA), 2021
arXiv / video / code /

Using adversarial reinforcement learning to imrove the robustness of robotic manipulation.

Physics-Guided Active Learning of Environmental Flow Fields
Reza Khodayi-mehr, Pingcheng Jian, Michael M. Zavlanos
Learning for Dynamics and Control (L4DC), 2023
arXiv /

We propose a physics-based method to learn environmental fields (EFs) using a mobile robot.

Workshop Papers
DAIR: Disentangled Attention Intrinsic Regularization for Safe and Efficient Bimanual Manipulation
Minghao Zhang*, Pingcheng Jian*, Yi Wu, Huazhe Xu, Xiaolong Wang
ICML Workshop: Reinforcement Learning for Real Life, 2021
project website / arXiv / video /

Solve complex bimanual robot manipulation tasks on multiple objects with disentangled attention, which provides an intrinsic regularization for two robots to focus on separate sub-tasks and objects.

Projects

Some course projects show my engineering capability.

Quadruped Spider Robot
/ video /

I built all the hardware and software of this spider robot from scratch. I drew the parts of this robot with SOLIDWORKS and then printed them with 3D printer. The controller of this robot is a Raspberry Pi, and the control algorithm is written in Python. Then I generated the .urdf files of this robot and built the simulator of this robot with PyBullet.

Teaching Experiences

Teaching Assistant at Duke: ECE 382L/ME 344L - Control of Dynamic Systems / lecture / recitation /
Teaching Assistant at Duke: ECE 689/COMPSCI 676 - Advanced Topics in Deep Learning