Yifan Han
Yifan Han is a master's student at the Institute of Automation, Chinese Academy of Sciences (CASIA), affiliated with the National Laboratory of Pattern Recognition and supervised by Professor Wenzhao Lian. His research focuses on robot manipulation, dexterous manipulation, and vision-language-action models for real-world robotic systems. He is best known as a co-first author of BORA, an offline-to-online reinforcement learning post-training framework for dexterous VLA models, and DexHiL, a human-in-the-loop post-training framework for dexterous manipulation, both developed in collaboration with Shanghai Jiao Tong University.
“BORA solves a critical bottleneck in deploying dexterous robot hands — the gap between a VLA model that 'understands' a task visually and one that can actually execute it reliably with 20+ fingers and joints in the real world. It achieves an 86% average success rate on five real-world dexterous tasks, up from ~53% with pure imitation learning.”
Source→“Designed the action-conditioned critic that 'fuses continuous action chunks with the VLM's cognition tokens... enables precise, action-conditioned value guidance evaluated on physical execution consequences rather than visual context alone.'”
Source→AI-extracted from podcast / newsletter / paper summaries. May contain errors.