Zihan Yang's Homepage

Zihan (Kevin) Yang

I am currently a master's student at School of Aeronautic Science and Engineering, Beihang University, Beijing, China, supervised by Prof. Kexin Guo, Prof. Xiang Yu and Prof. Lei Guo.

In 2023, I received my B.Eng. in Material Science and Engineering from Northwestern Polytechnical University, Xi'an, China, and Queen Mary University of London, London, UK, under a Sino-foreign cooperative education program. I have a multidisciplinary perspective, which often inspires my research.

My research interest lies in the area of learning and robotics. From 2023-2025, I spent my time working on deep learning for planning and control, with extensions toward how control theory can be utilized for machine learning. I am working on generative models and reinforcement learning approaches for more complex robotic tasks. I am also seeking for a PhD position in the field of robotics and AI.

Humankind are working towards a better future, which can be seen as an inverse problem in the sense that we are endeavoring to transform our world towards some optimal conditions. My goal is trying to solve such inverse problems that encourage human progress.

CV / Github / ORCID / Research Statement

Email: snrt_zzhan@buaa.edu.cn

profile photo Zihan Yang 杨子涵

Research

My current researches include deep learning, online learning, motion planning and control for robots. I also encountered some interesting problems where machine learning meets control theory.
Some papers are highlighted in title. ^*Equal contribution.

	Towards Non-Structural Disturbance Prediction: Feedback-Calibrated Meta-Adaptation Zihan Yang Jindou Jia, Meng Wang, Yuhang Liu, Kexin Guo, Xiang Yu. Under Review. project page A meta-adaptation framework for predicting general non-structural disturbances, followed by feedback-calibrated online adaptation to estimate model parameters with attenuation on learning residuals.
	Feedback Favors the Generalization of Neural ODEs Jindou Jia^, Zihan Yang^, Meng Wang, Kexin Guo, Xiang Yu. International Conference on Learning and Representation (ICLR) 2025 Oral Presentation project page / arXiv / code Feedback neural network improves the generalization of neural ODE through real-time state-feedbacks in continuous-time tasks.
	Learning-based Observer for Coupled Disturbances Jindou Jia^, Meng Wang^, Zihan Yang, Bin Yang, Yuhang Liu, Kexin Guo, Xiang Yu. arXiv Decoupling the disturbance into a state-coupled part and an external-input-related part using Chebyshev polynomial approximation. Online learning of the disturbance model by estimating the external-input-related part.
	Optimizing Control-Friendly Trajectories with Self-Supervised Residual Learning Kexin Guo^, Zihan Yang^, Yuhang Liu, Jindou Jia, Xiang Yu. Under Review. An approach for taming model uncertainties through minimizing the residual dynamics via trajectory optimization.
	TRACE: Trajectory Refinement with Control Error Enables Safe and Accurate Maneuvers Zihan Yang, Jindou Jia, Yuhang Liu, Kexin Guo, Xiang Yu, Lei Guo. IEEE International Conference on Control and Automation (ICCA) 2024 Best Student Paper Award IEEE Xplore A closed-loop trajectory refinement method for quadrotors utilizing the differential-flatness property, enabling safe and accurate maneuvers without controller fine-tuning.
	Co-Optimization of Motion and Energy Domain for Hydrogen-Powered Hybrid UAVs: A Bi-Directional Coupling Architecture Xiaowei Song, Xiaoyu Guo, Guowei Liu, Zihan Yang, Lu Liu. IEEE International Conference on Control and Automation (ICCA) 2025 A Motion-Energy co-optimization architecture for hydrogen-powered hybrid UAVs.

Misc.

In spare time, I enjoy travelling ✈, playing electric guitar 🎸 and a good cup of coffee ☕.
I keep moving on towards interesting ideas and a chill lifestyle.

Credits to the website design from Jon Barron.