|
I am now an undergraduate student of Nanjing University (NJU), majoring in Computer Science and Technology (National Elite Program). I'm also an incoming Ph.D. student of Zhejiang University (ZJU). I'm now working as an intern in Shanghai Qi Zhi Institute. I have been working as a research intern at LAMDA since 2022, advised by Prof. Yang Yu. I was also a research intern in NUS , advised by Prof. Lin Shao . Email / CV / GitHub / Google Scholar |
|
|
My research focuses on reinforcement learning and robotics.
|
|
|
|
Macheng Shen, Jishen Peng, Zefang Huang In Submission Paper / Abstract
A fundamental challenge in imitation learning is the covariate shift problem. Existing methods to mitigate covariate shift often require additional expert interactions, access to environment dynamics, or complex adversarial training, which may not be practical in real-world applications. In this paper, we propose a simple yet effective method (DeCIL) to mitigate covariate shift by incorporating a denoising mechanism that enhances the contraction properties of the state transition mapping. Our approach involves training two neural networks: a dynamics model ( f ) that predicts the next state from the current state, and a joint state-action denoising policy network ( d ) that refines this state prediction via denoising and outputs the corresponding action. We provide theoretical analysis showing that the denoising network acts as a local contraction mapping, reducing the error propagation of the state transition and improving stability. Our method is straightforward to implement and can be easily integrated with existing imitation learning frameworks without requiring additional expert data or complex modifications to the training procedure. Empirical results demonstrate that our approach effectively improves success rate of various imitation learning tasks under noise perturbation.
|
|
Fan-Ming Luo, Zuolin Tu, Zefang Huang, Yang Yu Conference on Neural Information Processing Systems (NeurIPS), 2024
Real-world decision-making tasks are usually partially observable Markov decision processes (POMDPs), where the state is not fully observable. Recent progress has demonstrated that recurrent reinforcement learning (RL), which consists of a context encoder based on recurrent neural networks (RNNs) for unobservable state prediction and a multilayer perceptron (MLP) policy for decision making, can mitigate partial observability and serve as a robust baseline for POMDP tasks. However, previous recurrent RL methods face training stability issues due to the gradient instability of RNNs. In this paper, we propose Recurrent Off-policy RL with Context-Encoder-Specific Learning Rate (RESeL) to tackle this issue. Specifically, RESeL uses a lower learning rate for context encoder than other MLP layers to ensure the stability of the former while maintaining the training efficiency of the latter. We integrate this technique into existing off-policy RL methods, resulting in the RESeL algorithm. We evaluated RESeL in 18 POMDP tasks, including classic, meta-RL, and credit assignment scenarios, as well as five MDP locomotion tasks. The experiments demonstrate significant improvements in training stability with RESeL. Comparative results show that RESeL achieves notable performance improvements over previous recurrent RL baselines in POMDP tasks, and is competitive with or even surpasses state-of-the-art methods in MDP tasks. Further ablation studies highlight the necessity of applying a distinct learning rate for the context encoder.
|
|
Ruifeng Chen∗, Chengxing Jia∗, Zefang Huang, Tian-Shuo Liu, Xu-Hui Liu, Yang Yu The International Conference on Machine Learning (ICML), 2024
Learning a high-quality transition model is of great importance for sequential decision-making tasks, especially in offline settings. Nevertheless, the complex behaviors of transition dynamics in real-world environments pose challenges for the standard forward models because of their inductive bias towards smooth regressors, conflicting with the inherent nature of transitions such as discontinuity or large curvature. In this work, we propose to model the transition probability implicitly through a scalar-value energy function, which enables not only flexible distribution prediction but also capturing complex transition behaviors. The Energy-based Transition Models (ETM) are shown to accurately fit the discontinuous transition functions and better generalize to out-of-distribution transition data. Furthermore, we demonstrate that energy-based transition models improve the evaluation accuracy and significantly outperform other off-policy evaluation methods in DOPE benchmark. Finally, we show that energy-based transition models also benefit reinforcement learning and outperform prior offline RL algorithms in D4RL Gym-Mujoco tasks.
|
|
|
|
Gang Yang, Siyuan Luo, Yunhai Feng, Zhixin Sun, Chenrui Tie, Lin Shao IEEE International Conference on Robotics and Automation (ICRA), 2024 ArXiv / Website / Video / Abstract
We present Jade, a differentiable physics engine for articulated rigid bodies. Jade models contacts as the Linear Complementarity Problem (LCP). Compared to existing differentiable simulations, Jade offers features including intersection-free collision simulation and stable LCP solutions for multiple frictional contacts. We use continuous collision detection to detect the time of impact and adopt the backtracking strategy to prevent intersection between bodies with complex geometry shapes. We derive the gradient calculation to ensure the whole simulation process is differentiable under the backtracking mechanism. We modify the popular Dantzig algorithm to get valid solutions under multiple frictional contacts. We conduct extensive experiments to demonstrate the effectiveness of our differentiable physics simulation over a variety of contact-rich tasks.
I participated in part of the experimental work of the paper, mainly responsible for training the reinforcement learning algorithm to solve the LCP, but I did not signature.
|
|
|
|
|
 |
Language based: C/C++. This project faithfully recreates the well-known game BOMBIT, it can be played in the QT. |
|
|
 |
Language based: C/C++. This project simplifies and recreates the recently globally popular game Overcooked, and I designed manual and automatic these two modes of play for it. It can be run directly in Windows terminal. |
|
|
|
Zhejiang University, Hangzhou, China
Ph.D student in Control Science and Engineering • Sep. 2025 (expected) to - |
 |
|
Nanjing University, Nanjing, China
B.E. in Computer Science and Technology • Sep. 2021 to Jun. 2025 (expected) |
 |
|
|
|
LAMDA Lab (led by Zhi-Hua Zhou), Nanjing University
Research Intern • Sept. 2022 to May. 2024 Advisor: Prof. Yang Yu Research Topics: Reinforcement Learning |
 |
|
National University of Singapore
Research Intern • Mar. 2023 to Oct. 2023 Advisor: Prof. Lin Shao Research Topics: Robotics |
 |
|
|
|
Shanghai Qi Zhi Institute
Intern • Aug. 2024 to Now Research Topics: Robotics |
 |
|
|
|
|
|
Before university, I had a very rich experience in studying for mathematical Olympiads, spanning ten years. During this period, I achieved many honors, including a bronze medal in the China Mathematical Olympiad (CMO 2020), and consistently placed in the top three in all provincial mathematics competitions. My extensive experience in mathematical Olympiads has greatly helped my growth, cultivating many habits, especially how to independently think when encountering difficult problems, and actively communicate and discuss with teachers and classmates afterwards. I am very grateful for this unforgettable and meaningful learning experience!
I am very passionate about sports, especially volleyball and swimming. I served as a referee for the men's volleyball department cup at Nanjing University in the autumn semester of 2022-2023. |
|
|
 |
Language based: MicroPython. Under the guidance of teachers from the Medical School of Nanjing University, I independently completed the welding and connection of hardware devices, as well as the design of the code related to the blood oxygen meter. It can measure heart rate (HR), blood oxygen saturation (SPO2), and body temperature (Temp) relatively accurately. Although the device is relatively rudimentary, I hope it can fulfill its role. |