SVR controller of a biped robot in the sagittal plane when pulling a mass

This paper describes the control of an autonomous biped robot capable of pulling a mass in the sagittal plane. The sagittal balance control uses an 8-link model of the robot, which is difficult to apply in real time due to the excessive computational effort. To solve this problem it is used a support vector regression (SVR) controller, which is able to perform the balance control in real time. As the ZMP balance control is non-linear, an SVR is appropriate. An SVR calculates the optimal hyper plane for the training data and computes faster than several other artificial intelligence techniques. An SVR that was trained based on experimental and simulation data, is used. This SVR uses the zero moment point (ZMP) position and its time variation as inputs, and the sagittal angle correction of the robot's body is obtained as the output. The ZMP is obtained making use of the force distribution on four force sensors placed under each robot's foot. This controller combines the use of the ankle and torso joints movements to correct the ZMP. The gait implemented in this biped is similar to a human gait that was acquired and scaled to the robot size. In this paper some experiments are also presented and the results show that the used gait combined with the SVR are appropriate to be used in the control of the biped robot, even when it drags a mass of about 65% of its own weight.