Evolving a CPG controller for a biped robot with neuromodulation

In order to create agile locomotion for legged robots, so far various methods have been proposed with the concept of Central Pattern Generator (CPG). In this paper, the concept of neuromodulation widely observed in biological nervous system has been introduced to CPG circuits for a 3-D biped robot. The proposed method is able to dynamically change the properties of CPG according to the current situation. Simulation results show that the biped robot walks on uneven terrain by changing the torque outputs applied to each joint. Furthermore, the proposed approach outperforms the monolithic CPG approach with fixed sensory feedback not only in terms of adaptability but also evolvability.