Humanoid Robot Posture-Balance Control

Controlling humanoid robot posture with the consideration of balance is a difficult problem. In this paper, we present a posture-balance control approach for users to program humanoid robot postures. To regulate posture and balance control, the proposed idea is to develop a system which helps to program robot motion by human demonstration and maintain robot balance as well. Two components are investigated in the system. First, human motion for controlling robot posture is captured by a camera. We use a Kinect RGB-D camera to acquire human joint position information and derive the joint angles to control the robot by the simple geometry model of a human body. Since the image information from Kinect is not always stable, the mechanism to prevent robot self-collision is designed by constraining robot joint ranges. Second, this paper adopts the inverse COG jacobian method to adapt the robot posture to maintain its balance. The adapted posture is quite similar to the taught one but still maintains the balance. In the experiment, we provide extensive validation of the proposed system which shows good results with forward-lean and side-lean postures. Regarding to the computational time, the proposed method is superior to our previous work by Genetic algorithm. In addition, there does not exist any degradation of posture similarity compared to our previous work.