Dongxiu Liu

Beijing University of Posts and Telecommunications

Papers

Total Citations

H-Index

About

Dongxiu Liu is a leading researcher in embodied AI and foundation models, with a focus on bridging the gap between internet-scale data and physical-world agents. Her most impactful work, “Universal Actions for Enhanced Embodied Foundation Models” (2025, 6 citations), tackles a critical bottleneck in robotics: the incompatibility of diverse action spaces across crowd-sourced datasets. By proposing a universal action representation, Liu enables training on heterogeneous embodied data, dramatically improving generalization and sample efficiency for robotic agents. This contribution addresses a fundamental challenge in scaling embodied intelligence, akin to the breakthroughs seen in large language models. Her research sits at the intersection of computer vision, robotics, and representation learning, with implications for autonomous systems, human-robot interaction, and simulation-to-real transfer. Liu’s work is gaining rapid recognition for its practical impact, offering a pathway to more capable and adaptable embodied agents. Her approach promises to accelerate progress toward foundation models that can perceive, reason, and act in the physical world.