A Bionic Cerebellar Control Model And Its Application In The Arm Motion

The cerebellum is an important nerve regulation center of the human body and plays an important role in motor control and motor learning.When the human body causes cerebellar damage due to genetics or external factors,ataxia symptoms such as"staggering and unstable standing"will appear.Ataxia seriously reduces the quality of life of patients,but there is currently no effective treatment.Therefore,in-depth study of the cerebellar motor control mechanism and the establishment of a cerebellar model with bionic significance can not only improve our cognition of the cerebellar motor control mechanism,but also have important reference value for the treatment of ataxia.The existing cerebellum model only aims at controlling the effect.Although it draws on the functional model of the cerebellum,it ignores the structural characteristics and learning mechanism of the cerebellum.Therefore,according to the cell type of the cerebellum and its connection method,this paper establishes a bionic cerebellar functional module,and combines supervised learning and reinforcement learning to build two bionic cerebellar motion control models with different learning mechanisms.The main contents include:(1)On the basis of in-depth study of the physiological characteristics and structural functions of the cerebellum,a functional module of the cerebellum that can clarify the main pathways and mechanisms of the internal neurons is established by using neurocomputing methods and neuron simulators.(2)A supervised learning cerebellum model was built based on the cerebellum function module and supervised learning algorithm,and the correctness of the established model was verified on the neural network simulator Emergent.On this basis,further design the robot arm tracking control simulation experiment.The results show that the errors of the 6 and 10 basic units are 5.12?1.09 mm and 4.28?0.63 mm,respectively.Even if more basic units are used,the control performance can be improved.Aiming at the problem of the model's insufficient ability of multi-dimensional motor coordination,a cerebellum model based on an array of adjustable pattern generators is designed.This model uses multiple adjustable pattern generators to jointly operate the multi-degree-of-freedom arm,realizing multi-dimensional motion control.The results show that the distance error of the elbow and wrist joints of this model is 4.12±1.82 mm and 5.66±2.00 mm,and it has good tracking control performance.(3)According to the results of recent experiments on the anatomy and physiology of the cerebellum,there are reward signals in the cerebellum during motor learning.Based on this,this paper uses the actor-critic algorithm to construct a bionic cerebellar motion control model based on the reinforcement learning mechanism.The simulation results show that the distance errors between the elbow joint and the wrist joint of this model are 5.16±2.51 mm and6.82±1.89 mm,and good motion control can also be achieved.However,the execution speed of the biomimetic cerebellar model based on supervised learning is better than that of the biomimetic cerebellar motion control model based on the reinforcement learning mechanism(t(28)6.01(29)t(39)_{0.0 5}(28)1.68,p(27)0.05).(4)In order to further verify the feasibility of the proposed model,this paper developed a bionic arm control system based on force feedback,including the software and hardware design of the system.The force feedback control experiments show that the control errors of the two bionic cerebellar motion control models constructed in this paper are less than 10 mm.Both models have good trajectory tracking performance.In this paper,two bionic cerebellar motion control models with supervised learning and reinforced learning capabilities are established by using neural computing methods,and both have good control capabilities and biological rationality.