Xiongfeng Peng
Xiongfeng Peng is a researcher at the Advanced Research Lab, Samsung R&D Institute China-Beijing (SRCB), where he works on Vision-Language-Action models for robotic manipulation. He is the lead author of DAM-VLA, a dynamic action model-based VLA framework integrating VLM reasoning with diffusion-based action models, and H-VLA, a hierarchical VLA framework that decouples high-level key-action reasoning from low-level motion generation. His work focuses on enabling robots to perform complex manipulation tasks in dynamic environments by bridging semantic visual-linguistic understanding with precise low-level control.
“H-VLA splits robotic control into two stages: a Key-Action Model that predicts the next manipulation subgoal (a 6-DoF end-effector pose + gripper state), and a Motion Planning Model that generates dense frame-by-frame actions conditioned on that subgoal.”
Source→“H-VLA achieves 98% vs. DAM-VLA's 78% on Google Robot VM (Table 1)”
Source→AI-extracted from podcast / newsletter / paper summaries. May contain errors.