Runhan Huang

I am a first-year Ph.D. student at Harvard University, supervised by Prof. Yilun Du. I received my B.Eng. in Computer Science from Tsinghua University, Yao Class, where I was fortunately supervised by Prof. Hang Zhao in MARS Lab. Before starting my PhD, I was a visiting researcher at Harvard's Kempner AI Institute working with Prof. Yilun Du and Prof. Heng Yang.

My research interests center on Robotics, Generative Models and Reinforcement Learning. My goal is to develop robotic systems that can interact robustly and plan effectively in real-world environments, achieving generalizable and adaptable behaviors.

Feel free to reach out if you want to chat about research or collaboration!

Email / CV / Scholar / Github / Linkedin / Twitter

Research Experience

Tsinghua University

Research Assistant, MARS Lab

Oct. 2025 - May 2026

Supervised by Prof. Hang Zhao, working on robot manipulation.

Harvard University

Visiting Researcher, Kempner AI Institute

Mar. 2025 - Oct. 2025

Supervised by Prof. Yilun Du and Prof. Heng Yang, working on diffusion planning.

Shanghai Qizhi Institute

Research Assistant, MARS Lab, Tsinghua & Shanghai Qizhi Institute

Jun. 2024 - Mar. 2025

Supervised by Prof. Hang Zhao, working on robot locomotion and robot manipulation.

Education

Harvard University

Ph.D. in Computer Science

2026 - Present

Supervised by Prof. Yilun Du.

Tsinghua University

B.Eng. in Computer Science, Yao Class

2022 - 2026

Supervised by Prof. Hang Zhao.

Research

* indicates equal contribution, † indicates equal advising, and highlight indicates representative papers.

	World Action Planner: Generalizable Robot Decision-Making with Action-Conditioned World Models Xiangcheng Zhang, Runhan Huang, Yilun Du arXiv, 2026 project page / paper / code World Action Planner is an agentic robot planning system that searches for and composes executable action plans through imagination with an action-conditioned world model, generalizing to compositional long-horizon tasks, novel layouts, and new real-robot tasks without expert demonstrations.
	SkillX: Unified Multi-Skill Policy Learning for Humanoid Soccer Zhangchen Ye, Enxuan Ruan, Yifei Bao, Runhan Huang*, Jiankun Yang, Jiakang Jin, Yixiao Huo, Pengyuan Wang, Yinan Han, Huaxing Huang, Wenhao Cui, Yiming Li†, Xiaoyu Tian† CoRL 2026 project page / arXiv SkillX learns and composes diverse humanoid soccer skills within a unified command-conditioned policy, enabling robust long-horizon execution and sim-to-real deployment.
	Generative Control as Optimization: Time-Unconditional Flow Matching for Adaptive and Robust Robotic Control Zunzhe Zhang, Runhan Huang, Yicheng Liu, Shaoting Zhu, Linzhan Mou, Hang Zhao NeurIPS 2026 project page / arXiv GeCO proposes a time-unconditional flow-matching framework that turns robotic action generation into iterative optimization, enabling adaptive inference, improved robustness, and intrinsic out-of-distribution awareness.
	MLS-Bench: A Holistic and Rigorous Assessment of AI Systems on Building Better AI Bohan Lyu, Yucheng Yang, Siqiao Huang, Jiaru Zhang, Qixin Xu, Xinghan Li, Xinyang Han, Yicheng Zhang, Huaqing Zhang, Runhan Huang*, Kaicheng Yang, Zitao Chen, Wentao Guo, Junlin Yang, Xinyue Ai, Wenhao Chai, Yadi Cao, Ziran Yang, Kun Wang, Dapeng Jiang, Huan-ang Gao, Shange Tang, Chengshuai Shi, Simon S. Du, Max Simchowitz, Jiantao Jiao, Dawn Song, Chi Jin NeurIPS 2026Oral · Top 0.36% project page / arXiv / code MLS-Bench evaluates whether AI systems can invent generalizable and scalable machine learning methods across diverse research tasks.
	TTT-Parkour: Rapid Test-Time Training for Perceptive Robot Parkour Shaoting Zhu, Baijun Ye, Jiaxuan Wang†, Jiakang Chen†, Ziwen Zhuang, Linzhan Mou, Runhan Huang, Hang Zhao CoRL 2026 project page / video / arXiv TTT-Parkour proposes a real-to-sim-to-real framework that leverages rapid test-time training (TTT) on novel terrains, significantly enhancing the robot's capability to traverse extremely difficult geometries.
	Flexible Locomotion Learning with Diffusion Model Predictive Control Runhan Huang, Haldun Balim, Heng Yang, Yilun Du ICRA 2026 project page / video / arXiv / code / media coverage Diffusion MPC introduces a test-time adaptable locomotion planner grounded in a diffusion-based generative prior. An interactive training procedure further improves the performance of diffusion-based planners.
	MoELoco: Mixture of Experts for Multitask Locomotion Runhan Huang, Shaoting Zhu, Yilun Du, Hang Zhao IROS 2025 project page / video / arXiv / code MoELoco introduces a multitask locomotion framework that employs a mixture-of-experts strategy to enhance reinforcement learning across diverse tasks while leveraging compositionality to generate new skills.
	VR-Robo: A Real-to-Sim-to-Real Framework for Visual Robot Navigation and Locomotion Shaoting Zhu, Linzhan Mou, Derun Li, Baijun Ye, Runhan Huang, Hang Zhao RA-L 2025 project page / video / arXiv / code VR-Robo introduces a digital twin framework using 3D Gaussian Splatting for photorealistic simulation, enabling RGB-based sim-to-real transfer for robot navigation and locomotion.
	Robust Robot Walker: Learning Agile Locomotion over Tiny Traps Shaoting Zhu, Runhan Huang, Linzhan Mou, Hang Zhao ICRA 2025 project page / video / arXiv / code RRW introduces a proprioception-only, two-stage training framework with goal command and a dedicated tiny trap benchmark, enabling quadruped robots to robustly traverse small obstacles.