XPeng’s IRON Robot Responds to Gestures Without Voice Commands in New Demo

XPeng has published a video from its “XPeng Robot Anthropomorphic Lab” that tests IRON’s gesture-led interaction. In the demonstration, a tester switches repeatedly among a handshake, a fist bump and a high-five, while the humanoid robot adjusts its own motion in response rather than waiting for a voice command.

According to XPeng, the sequence is designed to show close-range hand coordination. When the position of the high-five changes, IRON switches hands to meet the tester, a correction intended to demonstrate continuous visual perception and motion adjustment during an ongoing interaction. The clip focuses on hand-level responsiveness rather than locomotion or object manipulation.

The company says IRON re-infers from visual frames every 0.2 seconds and predicts how the action may change over the following six seconds. This perception-and-adjustment loop is meant to let the robot keep pace with rapid human gesture changes, instead of executing a fixed, pre-programmed routine.

XPeng frames the capability as part of its “One Infra, Two Products” physical AI system, in which a shared AI foundation supports both intelligent vehicles and humanoid robots. The video is a short interaction test rather than a full product launch, but it highlights XPeng’s effort to connect automotive AI infrastructure with embodied robotics.