Zhixuan Liang

University of Hong Kong, Hong Kong Polytechnic University

Papers

Total Citations

H-Index

About

Zhixuan Liang is a researcher at the forefront of robotic planning and multi-agent systems, with a focus on integrating generative AI with hierarchical decision-making. Their work primarily addresses the challenge of enabling robots to execute complex, long-horizon tasks from high-level instructions. Liang’s major contribution is the development of SkillDiffuser, an end-to-end framework that bridges diffusion-based trajectory generation with interpretable skill abstractions, allowing for coherent, long-range task execution. This work has garnered 18 citations since its 2024 publication, highlighting its immediate impact on the field. Earlier, Liang pioneered AdaptDiffuser, a self-evolving planner that uses diffusion models to overcome the limitations of offline reinforcement learning by adaptively generating diverse training data. With 8 citations, this work established a new paradigm for adaptive planning. Liang’s foundational research in hierarchical deep reinforcement learning for multi-robot cooperation in partially observable environments (6 citations) addressed critical communication bottlenecks in real-world applications like search and rescue. By combining theoretical rigor with practical implementations, Liang is shaping the next generation of intelligent, autonomous systems capable of reasoning and acting in complex, dynamic environments.