Xianyuan Zhan

Papers

Total Citations

H-Index

About

Xianyuan Zhan is an artificial intelligence researcher whose work sits at the intersection of embodied AI, reinforcement learning, and foundation models. His research tackles some of the most challenging problems in building intelligent agents capable of learning from diverse, real-world data and translating that knowledge into effective action. Among his notable contributions, Zhan has explored the development of universal action frameworks for embodied foundation models, addressing a fundamental bottleneck in the field: the incompatibility of action spaces across heterogeneous embodied datasets. This work, already accumulating citations shortly after its 2025 publication, pushes toward scalable, internet-scale training for physical agents — a critical step in realizing truly general-purpose robots. His reinforcement learning research further demonstrates methodological depth. His work on off-policy actor-critic methods introduces novel strategies for exploiting historically successful experiences to improve Q-value estimation, moving beyond the conventional preoccupation with overestimation bias to unlock more robust policy learning. Though still building his citation profile, Zhan's focus on bridging large-scale learning paradigms with embodied intelligence positions him as a timely and forward-thinking contributor to one of AI's most consequential frontiers.