Yuan Xu
Yuan Xu is a researcher affiliated with Peking University whose work focuses on Vision-Language-Action models and robotic action representations. He is a co-author of the Semantic Anchoring method, which addresses the degradation of semantic structure during VLA fine-tuning by anchoring action representations to a semantic manifold through contrastive alignment and shared/private decomposition. His research spans AI and robotics, with contributions to improving both in-distribution task performance and out-of-distribution generalization in robotic systems.
“We introduce XEWorld, a controlled cross-embodiment testbed for world models that isolates embodiments by evaluating held-out robots within physically identical scenes”
Source→“We propose StaKe, a plug-in structured supervision framework for VLA fine-tuning that introduces two complementary signals—Stage Supervision (SS) and Keyframe Supervision (KS)—to improve fine-tuning performance while leaving the inference loop unchanged.”
Source→AI-extracted from podcast / newsletter / paper summaries. May contain errors.