Research On Quadruped Robot Control Method Based On A Multilayer CPG Neural Network

Dynamics,adaptability,and stability are the major difficulties in quadruped robotics research.The model-based control method needs to accurately establish the model of itself and the environment,has the disadvantages of large workload and complex motion planning,which reduces the real-time control.Behavior-based methods cannot make global plans due to the lack of high-level adjustments,result in less flexible and more limitations.Bio-based CPG control can spontaneously generate stable rhythmic signals in the absence of high-level regulatory and sensory information,and control organisms to achieve basic rhythmic m ovements.Due to its simple structure and strong adaptability,it does not need to model the environment and itself,reduces the amount of calculation,and is widely used in the motion control of foot robots.However,due to the fact that the robot's CPG c ontrol method cannot simulate the characteristics of the creature well,it needs further improvement in coordination and adaptability.In this paper,from the perspective of biology,the mapping of CPG output and joint drive is studied.The mapping functions of the two are established,and a multi-layer CPG network quadruped robot motion control system based on reflection coordination is constructed.Firstly,the movement characteristics of the quadrupedal walking gait and the trajectory of each joint were analyzed,and the motion law between the joints was summed up.According to the animal's stratified reflection mechanism,basic reflection behaviors such as stretch reflex and postural reflex were introduced into the multi-layer CPG network to adjust the output of the CPG neural network.The quadruped robot CPG output and joint-driven mapping functions of the hip joint,knee joint and ankle joint were constructed to make the multi-layer CPG network model more in line with biological characteristics.Secondly,the multi-layer CPG neural network model structure is studied,and a multi-layer CPG neural network based on reflection coordination is established and constructed,including: rhythm generation layer,pattern generation layer,motor neuron layer,reflection coordination layer.The improved Van der Pol oscillator model is used to model each layer network,the parameters in the model are analyzed,and the feasibility of the multi-layer CPG quadruped robot motion control based on reflection coordination is verified through simulation experiments.Finally,on the simulation platform and the physical experiment platform,the quadruped robot coordination between the limbs and joints,as well as the quadruped robot's in situ walking and walking gait were tested.The experimental results verify the feasibility and coordination of multi-layer CPG neural networks with reflective coordination layer applied to the stable walking of quadruped robots.