Research On Multi-Robot Active Environment Detection Based On Non-Gaussian State Estimation

With the continuous progress of science and technology,mobile robot has been more and more applied in every aspect of human life.Robot system often presents nonlinear,non-Gaussian and other complex characteristics,which poses great challenges to the application of gaussian noise filtering.The state estimation of robot is the basis of its environmental detection,which includes the mapping of the environment and the searching of objects in the environment.Most of the multi-robot SLAM algorithms have already planned the path and known the environmental information,but in practice it is difficult to obtain the environmental information in advance and plan the path in advance.Multi-robot active target search in unknown Marine environment is an important research content of active environmental detection.However,there is often non-Gaussian noise in target search in Marine environment,and there are related characteristics of multiple targets,so few scholars have been involved in it.Therefore,it is urgent to establish multi-target search algorithm in non-Gaussian environment.In summary,based on the state estimation,this paper makes the following researches on the relevant issues in active environmental detection:(1)In view of the single robot system is nonlinear and non-gaussian problems,an gaussian sum cubature Kalman filter algorithm is put forward.Firstly,an improved robust EM algorithm for parameter estimation of non-Gaussian noise is proposed,which overcomes the shortcomings of existing EM algorithms.Then,based on the Gaussian term merging method based on Mahalanobis distance and KL distance,a fusion scheme which can effectively combine the two Gaussian term merging methods is proposed.Finally,gaussian sum cubature Kalman filter frameworks are applied to estimate the motion state of robots in complex environments.(2)For multi-robot active SLAM under unknown environment is not completely ergodic problem such as environment,location accuracy is not ideal,put forward a kind of fusion based on observation and attractors multi-robot active SLAM algorithm.First of all,when the same landmarks observed by multiple robots,robot convex combination method fusion estimate for the target,reduce the uncertainty of landmark are estimated.Secondly,by introducing attractors,enhance the communication between the multi-robot system,improve robot positioning accuracy,and guide the multi-robot team to explore the environment at the same time.Finally,applied in unknown environments of multi-robot active SLAM build figure for unknown environment.(3)A Gaussian sum cubature Kalman filter algorithm based on perceptual adaptation is proposed to solve the nonlinear and non-Gaussian multi-target search problems in marine environment.First,the robot collects environmental information,perceives the environment,and makes active decisions.Secondly,in the search process,the associated knowledge is extracted by analyzing the trajectory of the underwater and surface targets,and the extracted associated knowledge is used in the search process to form an evolutionary search relationship and increase the search efficiency.Finally,applied to a wide range of marine environment of underwater target search,the water of non-gaussian environment more dynamic target search and location.