Acquisition of adaptive behavior of robot through bow-tie structure

摘要

In this paper, we discussed the role of cognitions in biological control systems introducing the bow-tie structure. In bow-tie structure, sensory inputs with great diversity are translated into smaller diversity but semantic information, and the semantic information is re-translated to great diversity signals to create variety of behaviors. The process of reducing the signal diversity can be considered as the cognition in biological system. Our simulation results with the 2DOF inverse pendulum and the 27DOF humanoid robot showed that the lower diverse signal tuning is able to adapt the robot behaviors to the environment through body/environment interactions, suggesting that how recognize the environment can be useful to choose appropriate behaviors. Obviously, the behavior modifications in our method are strongly constrained by the diversity translating algorithm, mechanical resonance mode in this paper. We need further discussions for the relationship between signal diversity controls and behavior adaptations.