Research On A Self-growing Model For A Local CPG Network With Multiple Physical Factors

Control methods based on CPG networks have many advantages overtraditional ways such as dramatically reducing the dimension of the controlvariables and preserving the complexity of the output pattern, therefore are widelyused in locomotion control of robots. But current CPG study polarizes in twoorientations: Some emphasize function more than the rationality on biological sideand artificially design the structure of the network based on control objectives,which causes the network functions with low adaptability and can only be used inspecific occasions. If new control objects are proposed, previous research worksare normally not very instructive and the network has to be designed again at basiclevels. The others focus on the growing mechanisms of biological neuron netw orksand build models to achieve the similarity with them in morphologic and statisticalaspects. But the models are short of application significances.This paper tries to integrate the growing process and control function of theCPG network by introducing physical factors which can determine the parametersof the control model into the growing model, and achieve the control objectives onthe premise of corresponding to biological realities.Firstly, by investigating the biological mechanisms of axonal elongation, agrowing model of axon is established based on gravitation and stochastic motion;The Rall model is modified to acquire a signal transmission model of the axon thatsuits our application situation; A simplified self-growing model of the neuron withmultiple physical factors is established by integrating the growing model and thesignal transmission model.Secondly, by studying the Matsuoka model, a topology structure for a localCPG network is proposed as the growing object of the network; A gro wing modelis established for the network to achieve the desire structure stably; The output ofthe network is calculated by the signal transmission model with the parametersderived from the growing process, and is used as the control signals of the hipjoints on quadruped robots.Finally, a simulation platform is built by which the whole process of localCPG network growing, calculating the output of the network and controlling thevirtual robot to walk with certain gaits is realized. And the biological rationality ofthe self-growing local CPG network with multiple physical factors proposed bythis paper and its feasibility to be used in quadruped robots control are verified.