부산대학교 도서관

In order to create agile locomotion for legged robots, so far various methods have been proposed with the concept of neural circuits, or so-called Central Pattern Generator (CPG). In contrast to these approaches in which monolithic CPG neural circuits are employed to control locomotion, in this paper a polymorphic neural circuit is employed instead. This allows the dynamic change of its properties according to the current situation in real time. This is because what we are intrigued with is not to simply generate stable locomotion patterns but to investigate how sensory information modulates locomotion patterns according to the current situation.To this aim, the concept of neuromodulation with a diffusion-reaction mechanism of chemical substances called neuromodulators is introduced to modulate a CPG circuit. To verify the feasibility of this approach, this concept is applied to the control of a 3-D biped robot which is intrinsically unstable. In order to explore the parameters of the CPG circuit with a neuromodulation mechanism, an evolutionary, algorithm is employed in this study.Simulation results show that the neuromodulation mechanism dynamically changes synaptic weights of the CPG circuit. This synaptic change leads to the modulation of oscillator outputs and creates the appropriate walking pattern according to the situation in real time. In addition, simulation results indicate that the proposed approach outperforms the monolithic CPG approach not only in terms of adaptability but also evolvability.