Yanjiang Guo

Hi! I am a final year PhD student in Computer Science at Tsinghua University, advised by Prof. Jianyu Chen. I received my bachelor's degree in EE also from Tsinghua University in 2022 . From March 2025 to Jan. 2026, I was fortunate to visit Stanford University and work with Prof. Chelsea Finn, founder of Physical Intelligence.

My research focuses on building Physical AI that can learn and perform a wide range of tasks in the real world. I explore how pretrained foundation models can enable scalable learning and generalization, with a focus on vision-language-action policies and video-model-based policies. Currently, I also worked on agentic robot policies.

Besides research, I enjoy watching and playing soccer⚽. I also love physics👾 — “Everything should be made as simple as possible, but not simpler.”

Email / Scholar / Twitter / Github

Awards:
[2024.07] Best Paper Award Finalists in RSS 2024.
[2022.06] Outstanding Graduates Award, Top 10% Tsinghua undergraduate students.
[2017.11] Silver Medal in 34th National Physics Olympiad (CPhO).

Selected Research (* indicates equal contribution)

	Video Prediction Policy 2: Predict Better, Act Better Yanjiang Guo, Haodong Yan, Zhide Zhong, Zhongru Zhang, Qingyuan Yang, Qingzhou Lu, Xiaoyu Chen, Yen-Jen Wang, Shuying Deng, Chenghan Yang, Puzhen Yuan, Chenxin Liu, Tun Ban, Xiang Zhu, Yichen Liu, Kun Feng, Haoang Li, Jianyu Chen Preprint, 2026* project page / code / arXiv / Hugging Face / 量子位 We introduce VPP2, a world action model with strong zero-shot generalization in video prediction and action generation, combining event-level video pretraining, single-step visual planning, and a mixture-of-transformers action module.
	VLAW: Iterative Co-Improvement of Vision-Language-Action Policy and World Model Yanjiang Guo, Tony Lee, Lucy Xiaoyang Shi, Jianyu Chen, Percy Liang, Chelsea Finn International Conference on Machine Learning (ICML), 2026* project page / code / arxiv / twitter We iteratively improve VLA and action-conditioned world model.
	Ctrl-World: A Controllable Generative World Model for Robot Manipulation Yanjiang Guo, Lucy Xiaoyang Shi, Jianyu Chen, Chelsea Finn International Conference on Learning Representations (ICLR), 2026 project page / code / arxiv / twitter We train a controllable generative world model that can be used to evaluate and improve SOTA generalist robot policy.
	Video Prediction Policy: A Generalist Robot Policy with Predictive Visual Representations Yucheng Hu, Yanjiang Guo, Pengchao Wang, Xiaoyu Chen, Yen-Jen Wang, Jianke Zhang, Koushil Sreenath, Chaochao Lu, Jianyu Chen International Conference on Machine Learning (ICML), 2025 (Spotlight) project page / code / arXiv / twitter / 机器之心 / 量子位 We finetune a general-purpose video diffusion model into manipulation-focused video prediction model to guide policy learning.
	Prediction with Action: Visual Policy Learning via Joint Denoising Process Yanjiang Guo, Yucheng Hu, Jianke Zhang, Yen-Jen Wang, Xiaoyu Chen, Chaochao Lu#, Jianyu Chen# Advances in Neural Information Processing Systems (NeurIPS), 2024 project page / code / arXiv We jointly predict future images and robot actions in a unified DiT network, transfering physical knowledge from internet video data to robots.

	UniJEPA: Enhancing Robot Policy via Unified Continuous and Discrete Representation Learning Jianke Zhang, Yucheng Hu, Yanjiang Guo, Xiaoyu Chen, Yichen Liu, Wenna Chen, Chaochao Lu, Jianyu Chen International Conference on Machine Learning (ICML), 2026 project page / arXiv We predict latent future states to help the robot learn better policies.
	BagelVLA: Enhancing Long-Horizon Manipulation via Interleaved Vision-Language-Action Generation Yucheng Hu, Jianke Zhang, Yuanfei Luo, Yanjiang Guo, Xiaoyu Chen, Xinshu Sun, Kun Feng, Qingzhou Lu, Sheng Chen, Yangang Zhang, Wei Li, Jianyu Chen Robotics: Science and Systems (RSS), 2026* project page / arXiv We build a unified VLA model based on the Bagel model with unified understanding and generation capabilities.
	VLM4VLA: Revisiting Vision-Language-Models in Vision-Language-Action Models Jianke Zhang, Xiaoyu Chen, Yanjiang Guo, Yucheng Hu, Jianyu Chen International Conference on Learning Representations (ICLR), 2026 project page / arXiv / code We systematically evaluate how the base VLM affect the performance of VLA.
	Villa-x: enhancing latent action modeling in vision-language-action models Xiaoyu Chen, Hangxing Wei, Pushi Zhang, Chuheng Zhang, Kaixin Wang, Yanjiang Guo, Rushuai Yang, Yucen Wang, Xinquan Xiao, Li Zhao, Jianyu Chen, Jiang Bian International Conference on Learning Representations (ICLR), 2026* project page / arXiv / code We incoperate latent action modeling into VLA model.
	UP-VLA: A Unified Understanding and Prediction Model for Embodied Agent Jianke Zhang, Yanjiang Guo, Yucheng Hu, Xiaoyu Chen, Jianyu Chen International Conference on Machine Learning (ICML), 2025 arXiv / code We incoperate both multi-modal understanding (MMU) and future prediction into VLA model, enhancing both high-level semantic knowledge and low-level visual dynamics.
	Improving Vision-Language-Action Model with Online Reinforcement Learning Yanjiang Guo, Jianke Zhang, Xiaoyu Chen, Xiang Ji, Yen-Jen Wang, Yucheng Hu, Jianyu Chen International Conference on Robotics and Automation (ICRA), 2025* arXiv / twitter1 / twitter2 We make some initial exploration on leveraging online RL to improve the VLA model! We notice that online RL for VLA can be extremely unstable and thus we adopted a iterative approach.
	HiRT: Enhancing Robotic Control with Hierarchical Robot Transformers Jianke Zhang, Yanjiang Guo, Xiaoyu Chen, Yen-Jen Wang, Yucheng Hu, Chengming Shi, Jianyu Chen Conference on Robot Learning (CoRL), 2024 arXiv / twitter / 机器之心 We finetune pretrained VLM into VLA models with hierarchical transformers, keeping the generalization ability but also much higher control frequency.

ProCeedRL: Process Critic with Exploratory Demonstration Reinforcement Learning for LLM Agentic Reasoning
Jingyue Gao, Yanjiang Guo , Xiaoshuai Chen, Jianyu Chen
Annual Meeting of the Association for Computational Linguistics (ACL), 2026
arXiv

Guided RL for LLM agentic reasoning.

DoReMi: Grounding Language Model by Detecting and Recovering from Plan-Execution Misalignment
Yanjiang Guo* , Yen-Jen Wang*, Lihan Zha*, Jianyu Chen
International Conference on Intelligent Robots and Systems (IROS), 2024
project page / arXiv

We leverage LLM to pefrom both planning and monitoring, with a fine-tuned VLM as detector.

Advancing Humanoid Locomotion: Mastering Challenging Terrains with Denoising World Model Learning
Xinyang Gu*, Yen-Jen Wang*, Xiang Zhu*, Chengming Shi*, Yanjiang Guo, Yichen Liu, Jianyu Chen
Robotics: Science and Systems (RSS), 2024 (Best Paper Award Finalists)
project page / code / arXiv / 机器之心

We train humanoid robot to master challenging terrains such as stairs, slopes, and snow grounds with zero-shot sim2real transfer.

Source code from Jon Barron.