Yuze Zhu 朱昱泽

I am a fourth-year undergraduate student at the School of Artificial Intelligence, University of Chinese Academy of Sciences (UCAS). I am currently a visiting student at UC Berkeley, working at Berkeley AI Research (BAIR), advised by Roei Herzig.

My long-term goal is to build general-purpose robots that understand and interact with the physical world. I am particularly interested in representation learning and spatial intelligence, studying how robots can develop an intuitive understanding of 3D space to support robust generalization across tasks and environments. I am currently seeking a PhD position starting in Fall 2027, and am happy to discuss potential research fit and collaboration opportunities.

Beyond research, I am a passionate photographer. Born and raised in Hangzhou, I spent three years in Beijing for my studies and am now based in Berkeley, California, and I always enjoy meeting friends from around the world. If you would like to chat about research, daily life, or anything culturally Chinese, please feel free to reach out. Emails from anywhere are warmly welcome!

Berkeley AI Research (BAIR) · UC Berkeley

Aug 2026 - Present

Berkeley, California · Visiting Student · Advised by Roei Herzig (Darrell Group)

Investigating learned representations in World Action Models (WAMs), with a focus on how task-relevant information emerges and supports generalization and adaptation in robotic manipulation. Also exploring human-video in-context learning, studying how representations from demonstrations can be aligned with the robot's current state to guide future video and action generation.

FiveAges (Embodied AI Startup) & CASIA

Mar 2026 - Aug 2026

Beijing, China · Algorithm Intern · Advised by Yan Huang and Liang Wang

Co-developed BridgeVLA++, a memory-augmented Vision-Language-Action framework for generalizable 3D manipulation, and led its simulation and real-robot evaluation; co-first-authored the manuscript submitted to IEEE TPAMI.

State Key Laboratory of Multimodal AI Systems, CASIA

Jul 2025 - Jan 2026

Beijing, China · Research Intern · Advised by Junyu Gao and Changsheng Xu

Developed a closed-loop self-correction framework for continuous Vision-Language Navigation (VLN) using world-model foresight and environmental feedback, and led the experimental implementation and evaluation on R2R-CE and RxR-CE; resulted in a second-author paper accepted at ICML 2026.