Gait adaptation method of biped robot for various terrains using central pattern generator (CPG) and learning mechanism

There is evidence showing that animals have inherent rhythmic pattern generators called central pattern generator (CPG) that produce locomotion, respiration, heartbeat, and etc. Until now, the CPG has been widely used in robotic systems especially for locomotion. In this paper, we propose a gait adaptation method of biped robot for various terrains. The CPGs are used for creating desired joint angle of a biped robot. And a learning mechanism is realized with genetic algorithms (GAs) and Neural Network (NN). From an each terrain, the most suitable parameters of CPGs that can produce stable biped robot gait are founded by GAs. The parameter set founded by GAs and sensor data at that time are used for training the NN. After finishing training of the NN, the NN can be used for producing suitable parameters of the CPGs according to a sensor input. Finally the gait of the biped robot is changed according to environment.