Research On Brain-inspired Intelligence Decision And Planning For Multiple Unmanned Aerial Vehicles In Complex Environment

In the 20 st and 21 st century,the development of unmanned aerial vehicle(UAV)extremely rapid and the multi-UAV system gradually plays an important role in many fields.However,with the increasing complexity of environment,the current intelligence level of UAV is difficult to meet the requirements of dynamic game environment complex tasks.This dissertation investigates the intelligent task decisionmaking and cooperative motion-planning problem for multi-UAV,in order to improve the intelligence and cooperation capabilities of multi-UAV system.The main ideas of this thesis are as follows:First of all,the emergence task decision-making problem for multi-UAV is investigated.Consider the effect of task attribute stochastic and the uncertainty of UAV execution efficiency,the request-response dynamic task decision-making network is proposed by referring the request-response mechanism adopted by human when facing the same problem.Then a brain-inspired intelligence solution strategy is investigated based on dynamic game,iterative training,online decision by imitating the process of human decision-making(experience,learning and decision)to solve the multi-UAV decision-making mission,which can improve the ability to respond to dynamic environment and the solving speed,then the appropriate attributes UAV for emergence task are assigned appropriately.Secondly,the target threat behavior prediction problem in high dynamic confrontational environment is investigated.Considering the impact of environmental uncertainty,heavy clutter and the correlation between the behavior between the target and us,the star topology prediction network based on long and short-term memory(LSTM)neural network is investigated,including the hub network for extracting coupling interaction features and a host network for predicting target threat behavior.Compared with non-intelligent prediction methods,our method improves the prediction accuracy and speed of the real-time target threat behavior prediction problem.Thirdly,the distributed motion planning problem for pursuit-evasion game task is studied.Consider the multi-obstacles in dynamic environment and the effect of constraints of the environment,the problem is also solved by the brain-inspired intelligence solution strategy.First,the reward function is established and the experience buffer is created according to the experience.Then,the future trajectory of the target is predicted based on the deep learning algorithm.Furthermore,the actor network and critic network are established based on the scalability of LSTM,and the policy of multi-UAV is improved based on the centralized-training & decentralizeddecision structure.Finally,the online motion planning problem for pursuit-evasion game task is simulated in the virtual simulation interactive environment which is built by Unity3 D technology.Finally,the multi-UAV formation reconfiguration problem based on braincomputer interface(BCI)is investigated.Consider the situation in which the UAV is unable to make decisions autonomously or the decision result is not satisfactory,the multi-UAV formation reconfiguration based on BCI is developed.The spatial and temporal features of the motor imagery(MI)signal simultaneously is extracted and learnt by a hybrid network consisting of the convolutional neural network and long short-term memory network.The virtual simulation environment of multi-UAV formation reconfiguration is established by Unity3 D and the commander can control the multi-UAV formation by BCI in the virtual simulation.