XPENG IRON demonstrates adaptive gestures in humanoid test

XPENG IRON adjusts its hand movements as a person switches between handshakes, fist bumps and high fives in a laboratory demonstration published on XPENG’s YouTube channel. The company attributes the robot’s responses to its VLA 2.0 system, which it says continually interprets visual input and updates motion plans rather than waiting for a gesture routine to finish. The tester interacts without spoken commands.

The sequence tests changes during an interaction, not just recognition of an initial pose. A tester offers a handshake, then closes the hand into a fist before changing to an open palm. IRON adjusts its response as the gestures change. When the tester moves a high five beyond the responding arm’s range, the robot uses its other hand to meet it. Onscreen tracking labels identify changing gestures, including a fist and a moving palm.

According to XPENG’s presenter, IRON reassesses the scene every 0.2 seconds, linking visual perception with motion planning and execution in a continuous loop. That figure describes the claimed reasoning interval, rather than a measured completion time for each gesture. XPENG says the upgraded VLA 2.0 model has stronger temporal understanding and future behavior prediction, and that this capability runs on the robot itself. The demonstration focuses on adapting an action already under way as the person’s intended interaction changes.

Source: youtube.com – XPENG