Technology explainer
How Do Humanoid Robots Keep Their Balance While Running?
Running robots combine motion sensors, foot-force measurements and rapid control decisions to keep catching their moving bodies. Speed alone is not enough: safe stopping, recovery from slips and repeatable performance reveal whether the machine can work beyond a prepared track.
A humanoid robot can run only if it repeatedly catches itself before it falls. Motors move the legs, but the harder job belongs to the control system, which must estimate the body's motion and choose the next foot placement within milliseconds.
What does balance mean for a moving robot?
A standing machine tries to keep its centre of mass above the area supported by its feet. Running removes that comfortable condition because both feet may leave the ground. The controller instead manages momentum, plans where the next foot should land and prepares for the impact that follows.
Engineers often describe this as dynamic balance. The robot is allowed to lean or fall briefly, provided the next step redirects its motion before the body tips beyond recovery.
How does the robot know that it is tipping?
An inertial measurement unit tracks acceleration and rotation. Joint encoders report the angle of each hip, knee and ankle, while force sensors in the feet show when and where the ground is pushing back.
The computer combines these imperfect measurements into an estimate of the robot's position, speed and orientation. A delayed or noisy estimate can make the machine place a foot too late, so sensor timing matters as much as raw motor power.
How does the controller choose the next movement?
One approach uses a simplified model of the body to predict how different foot placements will change momentum. Another trains a policy in simulation, where the robot can experience millions of slips and pushes without damaging hardware. Many systems combine model-based planning with learned corrections.
The controller then sends torque or position commands to the joints. Fast feedback adjusts those commands as the foot hits the floor, the surface compresses or the upper body begins to rotate.
Why is stopping harder than sprinting?
Acceleration can be optimized for a straight, prepared lane. Stopping requires the robot to absorb energy without exceeding motor limits, losing traction or taking several unsafe steps. A machine that relies on a padded barrier has demonstrated speed, not complete locomotion control.
Real workplaces add people, loose objects, ramps and surfaces with changing friction. A useful humanoid must detect those changes, reduce speed early and remain stable while carrying a load.
What should a good demonstration report?
A lap time or top speed is only one measurement. Stronger evidence includes the number of successful trials, fall rate, stopping distance, energy use, payload, surface conditions and whether a human operator supplied remote corrections.
The most informative test is repeatable safe motion under varied conditions. Running fast is visible progress, but reliable balance appears when the robot can recover, stop and start again without a crew resetting it.
First appeared in
A Humanoid Robot Ran 100 Metres in 9.39 Seconds
