Optimized central pattern generator network for NAO humanoid walking control

Abstract

In this paper, an optimized central pattern generator (CPG) network is proposed for humanoid walking control. The CPG controller targets three joints (hip, knee and ankle) of each leg including 4 degrees of freedom (DOFs). The connections for CPG units of related joints are simplified and optimized hierarchically. The total number of CPG parameters is greatly decreased in this way. Moreover, the genetic algorithm (GA) is adopted to acquire the optimal parameters in batch and the complexity of the algorithm is decreased greatly. Finally, the proposed CPG controller is applied to the straight and circular walking on a commercial humanoid robot NAO, both in simulations and practices.