Skip to main content
. 2021 Jul 19;8:702845. doi: 10.3389/frobt.2021.702845

FIGURE 2.

FIGURE 2

The overall learning process of the integrated robust motion controller. We construct the learning controller as a Multi-Layer Perception (MLP) neural network that consists of three hidden layers. The neural network parameters are updated using the policy gradient method. The network produces joint target positions, which are translated into torque-level commands by PD (Proportional Derivative) control.