IRON is not a cold machine — it is designed to coexist harmoniously with people. That is why we aim to build a robot that is profoundly human-like: a helper and companion for all of us.
To achieve this goal, the engineers at XPENG — one of Australia's leading Chinese EV brands — are constantly pushing the upper limits of humanoid robotics across three dimensions: design, hardware architecture and motion control.
Design
XPENG's robotics designers have proposed the first General-Purpose Humanoid Design Framework — an extensible design system that allows a robot to be built at different heights and proportions — the same systematic approach XPENG applies to electric vehicles in Australia.
Guided by this framework, the robot is tailored like a made-to-measure suit: it must look human on the outside while keeping its mechanical internals compact and efficient. The system connects the logic of form, structure and motion, allowing "aesthetics" and "performance" to grow in harmony. Through parametric design...
More ingeniously, it forms a complete static-dynamic closed loop — from design sketches and component layout to reinforcement learning (RL)-optimised motion. It is like a systematic course taking the robot "from form to movement", bringing it to life step by step.
Hardware Architecture
With the design direction of the new generation defined, the engineering team explored deeper bionic principles and engineering optimisation in the robot's hardware architecture.
Compared with the previous-generation IRON, new degrees of freedom in the shoulders and waist allow the robot to perform complex, nuanced movements more naturally — bending, twisting and shrugging. The redesigned waist structure gives IRON a flexible "spine", enabling human-like gait patterns.
At the same time, the optimised dual-arm structure expands the upper body's dynamic range, allowing the arms to perform inertia-driven "fling" movements that give the gait more rhythm and a more human feel.
Beyond that, hardware engineers "draw the blueprint against the human skeleton", planning the structure with reference to human anatomy. Through the degrees of freedom in the legs and waist, power transmission forms a coordinated mechanical chain. It is stable yet flexible — as if weaving the robot a skeleton that can bend and stretch.
Building a humanoid robot with such "human-like logic" is no easy task.
To achieve it, engineers redesigned a range of joint actuators, greatly reducing their size and arranging them within extremely compact spaces according to human anatomy — letting the robot's motion reveal a flowing mechanical beauty.
Motion Control
When it comes to demonstrating these capabilities, how the components coordinate to achieve a smooth gait depends on the design of motion control.
To help the robot learn a more natural, human-like way of walking, engineers captured gait data from real human models. The nuances of the "catwalk" — the sway of the centre of gravity, the rhythm of the shoulders — were precisely recorded. They became the robot's first lesson in "learning to walk".
During data processing, the motion-control engineers developed a new motion remapping algorithm that lets the robot imitate human gait while avoiding foot-slip and collisions.
With reinforcement learning and simulation training, the robot gradually mastered natural human gait. That greatly improved the success rate of on-machine testing and eased the transition from simulation to the real world. IRON has learned to walk with a light, rhythmic catwalk — as if a "given life".
In its pursuit of human likeness, XPENG Robotics keeps challenging the limits — making machines more human and more attuned to people — the same philosophy behind every battery electric car XPENG builds.
Perhaps you are the explorer who will join us in reaching the uncharted — shaping the future of mobility, from electric car models to humanoid robots.
Click to read more about our open positions — and help shape the future of electric cars in Australia.
