Research On Mobile Robot Positioning Based On Neural Network Bayesian Filtering

In recent years,with the rapid development of information and intelligence,mobile robots are gradually entering people's work and life.With the rise of artificial intelligence technology,intelligent mobile robots are leading to a rapid development stage.More and more researchers are looking at using artificial intelligence technology to solve the bottleneck of traditional methods.To solve the mobile robot positioning problem is the premise to complete the robot movement,control,navigation and other functions,is a basic problem in mobile robot research,in some specific areas such as storage and logistics,intelligent patrol inspection,unmanned driving and so on has a rigid demand and important research value.However,the traditional positioning technology based on bayesian filtering still has some shortcomings in positioning accuracy and positioning error.The research work of this paper mainly focuses on the use of neural network to improve the positioning method of mobile robots based on bayesian filtering theory.The main research work includes the following aspects:First of all,this article in view of the robot localization problem,this paper expounds the framework,the bayesian filtering from the Angle of the expression of probability based on bayesian filtering method are introduced various forms for realizing,generally for kalman filtering and particle filtering,and analyze the performance of different filter,at the same time to the mathematical derivation method,then introduces the circulation network,neural network and the length of memory finally introduces in the robot localization need robot motion perception model and probability model.Then,inspired by the fact that cyclic neural network is very suitable for establishing longterm data dependence relationship between sequential data,a motion model based on long-short term time memory network is proposed to reduce markov limitations in traditional motion prediction model.In this paper,the differentiable particle filter network is based on the differentiable particle filter network.The differentiable particle filter network is a framework of the improved motion model.Firstly,it is verified that the motion prediction model based on LSTM has a more accurate regression trajectory than the ordinary motion model on a curve trajectory.Then,by collecting the robot trajectory roaming data based on the simulation environment developed by DEEPMIND,this paper establishes the parameter training set and completes the parameter training of the model.Finally,the performance of the algorithm is tested and verified in the simulation map environment.Finally,this paper analyzes the problem that single model is not enough to accurately describe the motion of the target in the extended kalman filter when the target has a complex and uncertain maneuver form.The starting point is the large range of model errors caused by uncertain maneuverability.In order to obtain the state of the mobile robot location precision estimate,with a model based on the length of the memory network was proposed to replace the traditional filtering method in the direct parameter model,study of model parameters directly from the training data were applied to predict and update phase error identification,and the data compensation,this paper use park dataset of piecewise training set,and use one paragraph to verify the feasibility and accuracy of the proposed algorithm,through the comparison with the traditional extended kalman filter algorithm error,gives the quantitative evaluation of trajectory;The validity of the extended kalman filter localization algorithm based on the long and short time memory network is verified.At last,this paper summarizes the work of this paper,and points out the shortcomings of the research content and the further direction of the future work.