About Jing-Cheng Pang (庞竟成):

[Research Overview] [Recent News] [Selected Publications] [Projects]

I am currently an AI Researcher at Huawei, focusing on building Domain Agent through reinforcement learning and large language models. I joined Huawei in July 2025 through the TopMinds Talent Program. Previously, I received my PhD in June 2025 from Nanjing University (LAMDA Group), where I was very fortunate to be supervised by Prof. Yang Yu to conduct reinforcement learning research. In 2024, I was a visiting scholar with Professor Masashi Sugiyama’s team at RIKEN-AIP in Tokyo, Japan (July - October). Prior to that, I obtained my BSc from the University of Electronic Science and Technology of China (UESTC) in June 2019.

Research Overview

During the PhD phase, my research focuses on connecting human and intelligent agent through natural language. By integrating reinforcement learning (RL) and large language models (LLMs), I aim to develop systems that not only interpret human intent but also act autonomously and learn iteratively in dynamic environments. Particularly, my study includes:

• Reinforcement Learning: language-conditioned RL, optimization algorithm, imitation learning, generalist agent;
• Large Language Models: model training, inference-time optimization, LLM-based agent;
• Embodied AI: home-service robot, sim2real policy learning.

Free free to contact/follow me if you are interested in my work.

[Back to top]

Recent News

• 2025.07: I join Huawei (AI Data Department) as an AI Researcher, focusing on building Domain Agent with RL and LLMs.
• 2025.06: ASRE (for RL optimization) and LaMSeI (for LLM application) are accepted by TNNLS and TMLR, respectively.
• 2025.03: Invited talk at HUAWEI Jiaxian Community (稼先社区). Topic: Exploration on Intelligent Agent System based on RL and LLMs [slides].
• 2025.01: ReViWo for robotics manipulation under viewpoint disturbance, is accepted by ICLR 2025.
• 2025.01: Invited talk at HUAWEI NAIE Group. Topic: RL-driven LLM Optimization and Recent Progress [slides].
• 2024.12: InCLET is accepted by AAMAS 2025 as a full paper. We will give an oral presentation at Detriot (It’s a pity that we can not get the visa)!
• 2024.09: Our work, KALM, is accepted by NeurIPS 2024!
• 2024.07: I start my visiting to Prof. Masashi Sugiyama’s team at RIKEN-AIP, Tokyo, Japan, doing RL research
• 2024.01: RLC for LLM self-improvement is accepted by ICLR 2024
• 2023.11: Awarded as Top Reviewer (8%) of NeurIPS 2023
• 2023.09: TALAR, for instruction-following agents, gets accepted by NeurIPS 2023

[Back to top]

Selected Publications

1. Jing-Cheng Pang, Nan Tang, Kaiyuan Li, Yuting Tang, Xin-Qiang Cai, Zhen-Yu Zhang, Gang Niu, Masashi Sugiyama and Yang Yu. Learning View-invariant World Models for Visual Robotic Manipulation. In: ICLR, 2025. [paper]
2. Peng-Yuan Wang, Jing-Cheng Pang, Chen-Yang Wang, Xu-Hui Liu, Tian-Shuo Liu, Si-Hang Yang, Hong Qian and Yang Yu. InCLET: In-context Learning from Language Models can Improve Embodied Instruction-following. In: AAMAS (Oral), 2025. [paper]
3. Jing-Cheng Pang, Si-Hang Yang, Kaiyuan Li, Jiaji Zhang, Xiong-Hui Chen, Nan Tang and Yang Yu. KALM: Knowledgeable Agents by Offline Reinforcement Learning from Large Language Model Rollouts. In: NeurIPS, 2024. [paper]
4. Jing-Cheng Pang, Peng-Yuan Wang, Kaiyuan Li, Xiong-Hui Chen, Jiacheng Xu, Zongzhang Zhang and Yang Yu. Language Model Self-improvement by Reinforcement Learning Contemplation. In: ICLR, 2024. [paper]
5. Jing-Cheng Pang, Xinyu Yang, Si-Hang Yang, Xiong-Hui Chen and Yang Yu. Natural Language Instruction-following with Task-related Language Development and Translation. In: NeurIPS, 2023. [paper]
6. Jing-Cheng Pang, Tian Xu, Shengyi Jiang, Yu-Ren Liu and Yang Yu. Reinforcement Learning With Sparse-Executing Actions via Sparsity Regularization. TNNLS, to appear. [paper]
7. Shengyi Jiang, Jing-Cheng Pang and Yang Yu. Offline Imitation Learning with a Misspecified Simulator. In: NeurIPS, 2020. [paper]

[Full publication list][Google Scholar]

[Back to top]

Projects

• Awesome-Papers-Autonomous-Agent  
  A collection of recent papers on building autonomous agent. Two topics included: RL-based / LLM-based agents.
• ImagineBench  
  A benchmark for evaluating RL algorithms that train the policies using both real data and imaginary rollouts from LLMs.

[Back to top]