Design And Application Of Developmental Network Based On The Principle Of Neuromodulation

In recent years,using the machine learning methods to study artificial intelligence has attracted the attention of researchers.Although machine learning methods such as deep learning play their powerful tools in the application field,they also have obvious drawbacks,such as poor generalization ability,low efficiency,and small deviations can produce large errors.How to increase the robustness and versatility of artificial intelligence,the development of general artificial intelligence has attracted more and more scholars' attention.Professor Juyang Weng from Michigan State University used development networks to solve the shortcomings in traditional machine learning.The human neuromodulation mechanism are combined with the developmental network in this thesis and a regulatory development network model based on the neuromodulation mechanism is proposed.Firstly,the basic principles and characteristics of the developmental networks are introduced in this thesis.Secondly,based on the neuromodulation principle of serotonin and dopamine,the thesis proposes a regulatory developmental network model containing serotonin and dopamine neurotransmitters to mimic the reward of dopamine and the punitive effect of serotonin.In order to verify the validity of the model,the thesis designs robot autonomous navigation experiments in three environments: static,dynamic and complex environments.In three environments,the thesis compares the distance between the agent and the target,the agent and the nearest obstacle,and the total number of running steps.The performance of the agents in different environments(great rewards,large penalties,and current regulatory development networks)is compared and analyzed.The experimental results show that compared with the original regulatory development network,the current regulatory development network shows better navigation effect and has higher practicability.Finally,based on the regulatory developmental network containing serotonin and dopamine neurotransmitters,the thesis simulates the response mechanism of acetylcholine and norepinephrine to uncertain events in the environment and proposes a regulatory developmental network model containing four neurotransmitters: serotonin,dopamine,acetylcholine and norepinephrine.Similarly,the paper first analyzes and compares the escape behavior of the obstacles in the static,dynamic and complex environments when the obstacles suddenly appear.Further,in order to fully demonstrate the superiority of the regulatory development network containing four neurotransmitters,the artificial potential field method is used as a comparison in a complex environment for experimental verification.The experimental results show that compared with the original regulatory development network,the regulatory development network containing serotonin and dopamine neurotransmitters,and the artificial potential field method,the agent can avoid obstacles in a timely manner and has better flexibility.