XPENG Motors has unveiled an upgrade to its second-generation VLA (Vision-Language-Action) large model. Leveraging the same foundational physical AI framework, this updated model bolsters its temporal comprehension and predictive prowess for future behaviors. It has been integrated into the edge computing aspect of XPENG's humanoid robot, IRON. IRON is now capable of real-time observation and judgment, constantly refining its movements. It re-evaluates its decisions every 0.2 seconds, drawing on visual data, and can foresee action changes within a 6-second window, allowing it to adapt nimbly to dynamic interaction scenarios. Slated for a November 2025 release, this humanoid robot boasts a graceful gait, a structure reminiscent of the human form, and a sophisticated blend of VLT (Vision-Language-Translation), VLA, and VLM (Vision-Language-Manipulation) capabilities. Outfitted with three Turing AI chips, it attains a high degree of intelligence, enabling functions such as conversation, locomotion, and interaction. XPENG has ambitious plans to commence mass production of the robot by the end of 2026, with deliveries expected in the first half of 2027. Initially, the robot will be deployed in commercial settings like stores and shopping malls, before gradually expanding its reach to industrial and domestic environments.