ZTE

“NebulaVLA: A Dual-Frequency Vision-Language-Action Model With Guide Action for Robotic Manipulation”

“Deliberative reasoning (System 2, 10 Hz VLM) is decoupled from reactive control (System 1, 20 Hz DiT generator), optimizing the compute-latency and enabling distributed deployment.”

“We conduct experiments to validate real-world deployment performance—particularly the effectiveness of the heterogeneous-frequency architecture and the Guide Action mechanism on AgiBot A2 robot platform”

“Yongcheng Wang, Lab/Institution: ZTE Corporation, The corresponding author of the paper”

“Cong Zhao, Lab/Institution: ZTE Corporation, The lead author of the paper”