What are the commonly used balance algorithms for biped robots?

At first, the balance control strategy used by biped robot was "static walking".

The characteristic of this strategy is that the projection of the center of gravity is always in the polygon support area during the robot walking.

The advantage of this control strategy is that the robot can stop during the whole walking process without falling down, but at the cost of very slow movement speed (each step needs 10 seconds or even longer) (because the projection of the center of gravity needs to be in the supporting area all the time, otherwise it will be unstable), because the static walking is far from the expectation of human beings.

So human beings developed another walking balance strategy: "dynamic walking".

In dynamic walking, the speed of the robot is increased to no more than 1 second per step. However, the shortcomings are also obvious. It is difficult for the robot to stop immediately in the state of motion (inertia), which makes the robot unstable in the process of state transition. In order to solve the influence of inertia, zero moment point (ZMP) is introduced into this control strategy. In the one-leg support stage, ZMP = center of gravity. The advantage of introducing ZMP is that if ZMP exists strictly in the support area of the robot, the robot will never fall down.

XZMP stands for positive ZMP, Xmc stands for center of gravity moving forward, l is the length of inverted pendulum, and g is the acceleration of gravity.

Because a single controller can't achieve biped balance in complex terrain, the switching strategy of multiple controllers is adopted to solve the balance problem. In this strategy, the robot's walking is set as a cycle, and each cycle is divided into different walking stages.