Skip to main content
. 2019 Oct 4;6(5):579–594. doi: 10.1089/soro.2018.0126

FIG. 1.

FIG. 1.

Overview of SenseCPG-PGPE. Displacement distance or angle of each actuator in a control objective (depicted here as a caterpillar-like robot) obeys the phase of an active rotator in the CPG-based controller. The feedback controller receives the mechanosensor values obtained across the body and calculates feedback to the CPG-based controller according to the equation. After running this cycle for several steps, the PGPE algorithm updates the parameters in the feedback controller based on performance. CPG, central pattern generator; PGPE, policy gradients with parameter-based exploration.