Mingleyang Li | 李铭乐洋

Mingleyang Li 李铭乐洋

I am a second-year undergraduate student at Peking University, majoring in Computer Science and Technology in the Turing Class. Before university, I focused on competitive programming and earned direct university admission eligibility through the National Olympiad in Informatics in 2023.

My current research centers on world action models, with the goal of learning predictive representations that connect visual observations, robot actions, and future physical dynamics. I was fortunate to first encounter the field of robotics through Prof. Hao Dong. I was also fortunate to get to know Yuran Wang, and together, we have worked on quite a few interesting projects :>

Chinese version (中文版本)

我是北京大学计算机科学与技术专业（图灵班）的大二本科生。在这之前，我主要学习信息学竞赛，并于 2023 年通过全国青少年信息学奥林匹克竞赛获得保送资格。

目前，我的研究重心是世界动作模型，希望学习能够连接视觉观测、机器人动作与未来物理动态的预测表征。我有幸由董豪教授接触到 robotics 这个领域。也很幸运地认识到了王昱然学长，我们一起合作完成了不少有意思的工作 :>

Email Google Scholar GitHub 小红书

登高峯一秒得獎一秒再破紀錄的一秒港灣晚燈山頂破曉摘下懷念記住美妙
— Eason Chan，《沙龍》

News

[2026/06] 🎉 I received the SenseTime Scholarship as the youngest recipient, one of 30 undergraduates selected nationwide.
[2026/01] 🎉 One paper was accepted to ICRA 2026.

Research

My research currently centers on World Action Models (WAMs): predictive models that learn how the physical world evolves under robot actions. I am particularly interested in action-conditioned representations, model architectures, scalable training, and systematic evaluation, with the broader goal of building robot agents that can anticipate the consequences of their actions and interact more effectively with the real world. Representative works are highlighted.
(*equal contribution, †corresponding author, ‡project lead)

	OpenWAM: An Open, Modular Exploration Towards Systematic World–Action Model Pretraining Yuran Wang‡, Siqiao Huang‡, *Mingleyang Li,** Chenhao Zhang, Jiaqi Liang, Weiyang Jin, Yue Chen, Xuemin Chi, Donghao Zhou, Qize Yu, Yu-Kai Wang, Yuhan Rui, Shenzhe Yao, Zhen Yuan, Zhenhao Shen, Kefei Zhu, Zijie Zhu, Ning Gao, Xiaowei Chi, Guanqi He, Shanghang Zhang, Hao Dong, Lin Shao†, Hang Zhao† Paper Project Page Code Hugging Face arXiv 2026 More details OpenWAM is an open research stack for systematic World–Action Model pretraining, bringing together modular infrastructure (OpenWAM-Infra), controlled studies of model design and training (OpenWAM-Study), and an open pretrained model (OpenWAM-α) for robot manipulation.
	XPolicyLab: A Unified Standard and Open Ecosystem for Robot Policy Evaluation and Deployment Tianxing Chen, Yue Chen, …, Mingleyang Li, … Paper Project Page Code *IROS 2026 @ ScaleInfra, Oral Presentation* More details XPolicyLab is a unified standard and open ecosystem that reduces N-policy-to-M-environment integration from O(NM) to O(N+M), standardizing installation, serving, and evaluation for 42 robot policies across simulation and real-robot platforms.
	RMBench: Memory-Dependent Robotic Manipulation Benchmark with Insights into Policy Design Tianxing Chen, Yuran Wang, *Mingleyang Li,** Yan Qin*, Hao Shi, Zixuan Li, Yifan Hu, Yingsheng Zhang, Kaixuan Wang, Yue Chen, Hongcheng Wang, Renjing Xu, Ruihai Wu, Yao Mu, Yaodong Yang, Hao Dong†, Ping Luo† Paper Project Page Code Under Review More details RMBench is a simulation benchmark comprising 9 manipulation tasks that span multiple levels of memory complexity, enabling systematic evaluation of policy memory capabilities in robotic manipulation.
	HeteroGenManip: Generalizable Manipulation for Heterogeneous Object Interactions Zhenhao Shen, Zeming Yang, Yue Chen, Yuran Wang, Shengqiang Xu, Mingleyang Li, Hao Dong, Ruihai Wu† Paper Project Page Code Under Review More details HeteroGenManip is a task-conditioned two-stage manipulation framework that decouples grasping from interaction, routing each object to a category-specialized foundation model via a Multi-Foundation-Model Diffusion Policy to generalize across heterogeneous object interactions.
	GarmentPile++: Affordance-Driven Cluttered Garments Retrieval with Vision-Language Reasoning *Mingleyang Li,** Yuran Wang, Yue Chen, Tianxing Chen, Jiaqi Liang, Zishun Shen, Haoran Lu, Ruihai Wu†, Hao Dong† Paper Project Page Code ICRA 2026* More details GarmentPile++ is a novel garment retrieval pipeline that can not only follow language instruction to execute safe and clean retrieval but also guarantee exactly one garment is retrieved per attempt.

Education

Peking University (PKU) 2025.09 - Present
Undergraduate Student, Turing Class

Selected Awards and Honors

[2026] SenseTime Scholarship (youngest recipient; 30 undergraduate awardees nationwide)
[2025] Peking University Freshman Scholarship, Third Prize
[2025] Dean's Scholarship for Freshmen, School of Electronics Engineering and Computer Science, Peking University

Algorithm Competitions

[2026] The 24th Peking University "Jiukun Cup" Programming Contest (PKU-CPC), First Prize
[2025] The 2025 ICPC Asia Chengdu Regional Contest, Gold Medal
[2025] The 23rd Peking University "Jiukun Cup" Programming Contest (PKU-CPC), First Prize
[2024] China Team Selection (CTS/CTSC), Rank 7/30
[2023] National Olympiad in Informatics (NOI) 2023, Gold Medal, National Training Team, Rank 33/50
[2022] National Olympiad in Informatics (NOI) 2022, Category E (junior-high participation, not counted as official), reached the National Training Team cutoff, Rank 2/2

This homepage is designed based on Jon Barron's website and deployed on Github Pages.
Last updated: Sep. 9, 2026
© 2026 Mingleyang Li

	OpenWAM: An Open, Modular Exploration Towards Systematic World–Action Model Pretraining Yuran Wang‡, Siqiao Huang‡, *Mingleyang Li,** Chenhao Zhang, Jiaqi Liang, Weiyang Jin, Yue Chen, Xuemin Chi, Donghao Zhou, Qize Yu, Yu-Kai Wang, Yuhan Rui, Shenzhe Yao, Zhen Yuan, Zhenhao Shen, Kefei Zhu, Zijie Zhu, Ning Gao, Xiaowei Chi, Guanqi He, Shanghang Zhang, Hao Dong, Lin Shao†, Hang Zhao† Paper Project Page Code Hugging Face arXiv 2026 More details OpenWAM is an open research stack for systematic World–Action Model pretraining, bringing together modular infrastructure (OpenWAM-Infra), controlled studies of model design and training (OpenWAM-Study), and an open pretrained model (OpenWAM-α) for robot manipulation.
	XPolicyLab: A Unified Standard and Open Ecosystem for Robot Policy Evaluation and Deployment Tianxing Chen, Yue Chen, …, Mingleyang Li, … Paper Project Page Code *IROS 2026 @ ScaleInfra, Oral Presentation* More details XPolicyLab is a unified standard and open ecosystem that reduces N-policy-to-M-environment integration from O(NM) to O(N+M), standardizing installation, serving, and evaluation for 42 robot policies across simulation and real-robot platforms.
	RMBench: Memory-Dependent Robotic Manipulation Benchmark with Insights into Policy Design Tianxing Chen, Yuran Wang, *Mingleyang Li,** Yan Qin*, Hao Shi, Zixuan Li, Yifan Hu, Yingsheng Zhang, Kaixuan Wang, Yue Chen, Hongcheng Wang, Renjing Xu, Ruihai Wu, Yao Mu, Yaodong Yang, Hao Dong†, Ping Luo† Paper Project Page Code Under Review More details RMBench is a simulation benchmark comprising 9 manipulation tasks that span multiple levels of memory complexity, enabling systematic evaluation of policy memory capabilities in robotic manipulation.
	HeteroGenManip: Generalizable Manipulation for Heterogeneous Object Interactions Zhenhao Shen, Zeming Yang, Yue Chen, Yuran Wang, Shengqiang Xu, Mingleyang Li, Hao Dong, Ruihai Wu† Paper Project Page Code Under Review More details HeteroGenManip is a task-conditioned two-stage manipulation framework that decouples grasping from interaction, routing each object to a category-specialized foundation model via a Multi-Foundation-Model Diffusion Policy to generalize across heterogeneous object interactions.
	GarmentPile++: Affordance-Driven Cluttered Garments Retrieval with Vision-Language Reasoning *Mingleyang Li,** Yuran Wang, Yue Chen, Tianxing Chen, Jiaqi Liang, Zishun Shen, Haoran Lu, Ruihai Wu†, Hao Dong† Paper Project Page Code ICRA 2026* More details GarmentPile++ is a novel garment retrieval pipeline that can not only follow language instruction to execute safe and clean retrieval but also guarantee exactly one garment is retrieved per attempt.