Jinliang Zheng

Papers

Total Citations

H-Index

About

Jinliang Zheng is a rising star in embodied AI, whose work tackles one of the field’s most persistent bottlenecks: the fragmentation of action spaces across diverse robotic datasets. His landmark paper, “Universal Actions for Enhanced Embodied Foundation Models” (2025, 6 citations), proposes a novel framework to unify heterogeneous action representations, enabling foundation models to train on internet-scale, crowd-sourced embodied data without costly retargeting. This contribution directly addresses why large-scale pretraining—so successful in vision and language—has struggled to transfer to robotics. By designing a universal action space, Zheng’s approach allows embodied agents to learn more generalizable manipulation and navigation skills from previously incompatible datasets, significantly expanding the pool of usable training data. Though early in his career, his work has already garnered attention for its potential to democratize robot learning, moving beyond lab-specific action formats. Zheng’s research sits at the intersection of robot learning, representation learning, and foundation models, with a clear trajectory toward building more capable, data-efficient embodied agents. His insights are shaping how the community thinks about scaling robot learning, making him a researcher to watch in the push toward general-purpose embodied intelligence.