Iron shakes hands, fist-bumps or high-fives. If an outstretched hand suddenly closes into a fist, it adapts its response at once. If the raised hand moves away from its own for a high five, it switches hands to follow.
According to IT之家, the robot does not simply recognise a gesture and then run a fixed sequence of movements. Every 0.2 second, it re-analyses the image in front of it and predicts changes in movement over the next 6 seconds.
These advances come from the second generation of VLA, Xpeng's large artificial intelligence model. According to the announcement, the model has a better grasp of how actions follow one another and predicts upcoming behaviour more accurately. It runs directly on board Iron.
According to Xpeng, Iron has a spine modelled on the human one, artificial muscles and a soft skin covering the whole body. Its head is a curved 3D screen, and its dexterous hands have 22 degrees of freedom.
The company plans to start mass production of Iron at the end of 2026 and begin deliveries in the first half of 2027. The robot will go first to shops and shopping centres, then to industry and private homes.
Update of 24 October: In the video, the person testing the robot gives it no spoken command. The image re-analysis rate announced by Xpeng matches the reasoning interval the company claims. It is not a measured time to complete each gesture.
Sources
- IT之家 ITHome · ithome.com
- Humanoid.guide · humanoid.guide