← Robot list · Robots in the real world course
Turn the robot with the mouse or your finger. Click an axis (a coloured circle or a card) to light up its ring and its axis and to read its description, press X-ray to look inside at the motors, gearboxes, controller and cables, and move each joint with its slider.
Move the sliders to crouch, twist the waist, raise the arms or turn the head. Press Walk to make it walk; the pace slider sets the step length and the rhythm.
Each leg has about 6 joints: 3 at the hip, 1 at the knee and 2 at the ankle. The strongest motors are here, because the legs carry the whole weight and absorb the impact of the steps. Seen in X-ray: motors, ball screws and linear actuators.
A joint that twists and leans the upper body. Twisting the waist against the legs cancels the spin that each step gives to the body, like a person swinging their arms.
Three joints at the shoulder, one at the elbow (and often two at the wrist). The arms balance the walk and do the work: carrying, pushing, opening doors.
Five fingers (a thumb and four fingers) moved by small motors placed in the forearm and pulled by tendons (seen in X-ray). Dexterous hands have 12 to more than 20 degrees of freedom and tactile sensors.
Stereo cameras (depth), microphones and often a lidar. The neck turns the head so that the sensors can look around while the body walks.
The battery sits in the torso, in the middle of the body, and gives 1 to 2 hours of autonomy. The computer runs the balance loop (1 kHz) and, on a second board, vision and planning. Seen in X-ray.
The humanoid really walks: a gait generator moves each foot (backwards under the body while it is on the ground, forwards in the air) and the two-link inverse kinematics of the leg finds the hip and knee angles that put the ankle there; the ankle then keeps the foot flat. When it stands, the hips move so that the centre of mass (computed from the mass of every part) stays over the feet: crouch and raise an arm to see it. When it walks, the centre of mass leaves the support area at each step, so the balance is dynamic, a controlled fall; real humanoids add feedback from the IMU and the force sensors (ZMP or model predictive control). The model is simplified: the real inside depends on the make.