Zeying Gong

I am a second-year PhD student at Hong Kong University of Science and Technology, Guangzhou, under the primary supervision of Prof. Junwei Liang, and co-supervision of Prof. Yingcong Chen. During my master's studies, I was co-supervised by Prof. Liuqing Yang.

Prior to this, I have spent over a year as a 5G network engineer at Huawei. There, I served as one of the technical leaders for the delivery of High-speed Railway 4G & 5G base station project in Jiangxi Province. I earned the bachelor's degree in Electronic Information Science and Technology from Sun Yat-sen University, supervised by Prof. Bin Rao.

My research interest is Embodied AI, especially Navigation.

Email  /  Scholar  /  Github  /  Resume

News

Publications

My research aims to develop generalist embodied agents that can understand and interact with complex real-world environments. My current interests lie in learning world and action representations from large-scale egocentric experience, enabling scalable and generalizable robot navigation and manipulation.

The following are my representative papers.

A unified navigation policy trained via rectified flow imitation learning and GRPO reinforcement learning, validated on humanoid, quadruped, and wheeled robots.

An LLM-driven hierarchical navigation framework that enables robots to navigate multi-floor buildings and find target objects through coarse-to-fine exploration strategy.

A future-aware social navigation framework that predicts human intentions and proactively plans collision-free paths in crowded environments.

A future-aware framework with a action-conditioned world model for single/multi-robot social navigation, validated on the Unitree Go2 quadruped robot.

FutureNav is a VLM‑based unified world‑action modeling framework that optimizes action policy, inverse/forward dynamics, and future generation objectives.

An agent-interface-aligned supervision framework for open-world mobile manipulation, enabling embodied agents to anticipate interaction outcomes before execution.

A unified VLA framework that bridges vision, language, and action through structured affordance forecasting, enabling robots to reason about what to manipulate, where to interact, and how to act.

A large-scale international robotics challenge focusing on generalizable perception and navigation. Co-organized and hosted Track 2 on social navigation at IROS 2025.

A comprehensive survey on vision-language-action models for autonomous driving, covering past developments, current state-of-the-art, and future directions.

A dynamic mapping framework that integrates human trajectory prediction for zero-shot social navigation in crowded environments.

A boundary-aware active learning method that enhances sparse-supervised point cloud semantic segmentation via geometric feature and model uncertainty fusion.

An efficient CNN framework for long-term time series forecasting that has been widely adopted in the community with over 100 citations.

A comprehensive benchmark and novel model for precipitation forecasting, advancing weather prediction capabilities.

Research on electromagnetic calculations for clustered UAV systems with practical applications in unmanned aerial vehicle operations.

Awards