Research On Construction Method Of Neural Network Model With Spatiotemporal Coordination Based On Postural Synergies During Arm Prehension

The rapid development of behavioral and cognitive neuroscience not only deepens people's understanding of brain behavior and function mechanism,but also promotes its combination with computational neuroscience,and greatly stimulates progress in machine intelligence.In real-time tracking of gestures,the spatiotemporal coordination between arm transport and hand grasping and the high dimension of gestures are the main bottlenecks in the research of grasping motion.Therefore,this paper,from the perspective of physiology and cognitive neuroscience,explores the construction method of neural network models related to these two research hotspots.Based on the VITE(Vector Integration to Endpoint)point-to-point trajectory formation model,a new neural network model which has biological meaning is proposed in this paper to explain the spatiotemporal coordination among arm transport component,hand preshape component and palm orientation component during reach-to-grasp tasks.Model changes the gating input signal,the basal ganglia thalamus cortex loop is exploited to adjust the overall speed of movement components,and optimizes the updating method of maximum grip aperture,coupled neurons are set up to ensure the exchange of status information with one another,and also considers experimental perturbation conditions.By comparing with the original method,the new model effectively extracts the important kinematic characteristics of arm prehension.In view of the high dimension and complex characteristics of human prehension,a task-oriented neural network model for arm movement and hand prehension is proposed.Model includes a collaborative control strategy for gesture description,using three synergy coefficients to define the evolution process of grasping,and builds a feedforward neural network which combines task demands and object features to generate synergy coefficients and achieve gesture grasping.The model also divides the prehension motion into three main channels: arm movement,hand preshaping and palm orientation.VITE trajectory generation model is used for movement command updating,and the inverse dynamic arm model is processed by the cerebellar inverse internal model.The validity of neural network and the coordination of grasping system are verified by experiments.Simulation results show that the gesture grasping model constructed by the above method has good applicability and grasping characteristics.