Research On Simultaneously Localization And Mapping Of Air-duct Cleaning Robot

Along with the high-speed development of national economy,the central air-conditioning system,which brings people a comfortable life and working environment,has been more and more widely used.At the same time,a variety of contaminants adhering to the wall of ventilation duct may produce serious threats to human health.The air-duct cleaning robot(ADCR)arises under this background.It can replace manual work to test and clean a large complex ventilation duct system,and has much advantages such as efficient,safe,no secondary pollution,etc.Simultaneous localization and mapping(SLAM)is a process by which a mobile robot can build a map of an environment and at the same time use this map to get its location.It is one of the hot research topic in the field of mobile robot,and is considered to be a pivotal technology that can truly make a mobile robot autonomously navigating in unknown environment.In order to further improve the automatic and intelligent level of ADCR,the research of this paper mainly focuses on the following two aspects:(1)design the control system and navigation system of ADCR to satisfy practical SLAM application requirement.(2)solve some basic and common problems of current SLAM research,and apply the research results to ADCR.Firstly,this paper designs a ADCR with the sensor system mainly composed of MEMS inertial measurement unit(IMU)and stereo visual sensor,and constructs an experiment platform simulating the characteristics of actual ventilation duct system.Then the kinematics model and the observation model of ADCR are developed on this basis.Besides,a data association algorithm is proposed to match the homogenous measurements and landmarks.Furthermore,a simulation system,whose sensor system is composed of encoder(for odometry)and two-dimensional laser range finder(for measurement),is built to verify the ef-fectiveness and universality of the SLAM methods proposed in this paper.Secondly,in order to solve the particle impoverishment and sample size dependency problem of FastSLAM,this paper propose a new FastSLAM framework based on improved shuffled frog leaping algorithm(SFLA).The main idea is to introduce SFLA to optimize the particle distribution before the resampling procedure,so that particle samples tend to be around neighbourhood regions with high likehood.For this reason,the required resampling procedure will be reduced distinctly while the particle diversity will be promoted greatly.In addition,to reduce the searching time and avoid local optimal solution,the leaping rule of SFLA is modified by properly extending step size coefficient,trend coefficient and leaping inertia component.Furthermore,unscented Kalman filter(UKF)is used to replace extended Kalman filter in the SLAM update process for achieving more suitable proposal distribution of robot pose and environment map.The simulation and experimental results indicate that the proposed method can achieve preferable precision of robot localization and environment map than the genic FastSLAM.Thirdly,the random-finite-set SLAM(RFS-SLAM)is a relatively new SLAM technique based on finite set statistics.It represents map and measurements as random finite set,while takes a Rao-Blackwellized(RB)implementation based on the Gaussian-mixture PHD filter for the map and a particle filter for the vehicle trajectory.Coompared to a vector-based SLAM approach,RFS-SLAM has two significant benefits:it does not rely on any heuristic data association process,and can account for detection statistics.However,similar to FastSLAM,RFS-SLAM also suffers from limitation of error accumulation introduced by incorrect model and inaccurate linearization of nonlinear functions.To overcome this problem,a fuzzy cerebellar model articulation controller(FCMAC)based RFS-SLAM algorithm,namely FCMAC-RFS-SLAM,is proposed in this paper.In this method,FCMAC is taken as the f-orward error compensator to integrate into RFS-SLAM framework,and is trained online to obtain the minimal error between the predictive robot pose with the motion model and the updated robot pose with RFS-SLAM filter.With the error compensation of robot trajectory,the particle distribution will be closer to its true distribution,moreover,the particle diversity will be improved to a certain extent.Hence,the estimation accuracy of robot trajectory will be further improved,which will play a positive role in the estimation of environment map.The simulation and experimental results show that the performance of the proposed algorithm is better than RFS-SLAM.Fourthly,to address the autonomous exploration problem for ADCR,this paper proposes a novel active SLAM framework,which effectively integrates FCMAC-RFS-SLAM with path planning method using harmonic functions,exploration strategy based on frontier and shortest path searching technology using Dijkstra algorithm.In the proposed framework,autonomous exploration can be regarded as a process of building and updating a topology network of global planning path(TNOGPP)iteratively.FCMAC-RFS-SLAM module produces accurate pose estimation of ADCR and occupancy grid map.Harmonic function path planning module obtains topology networks of local planning path(TNOLPP)between new frontiers and source nodes in current TNOGPP.Frontier based exploration strategy module selects the navigation target according to information gain,length of frontier edge and driving cost.Dijkstra module search the TNOGPP for the shortest path between current robot position and the navigation target.There are several benefits using the proposed active SLAM scheme:(1)TNOGPP is always in safe areas of the environment being explored.(2)Path planning procedure will be triggered only when the distance between ADCR and current navigation target is under a threshold.(3)There is no need to execute true global path planning,since TNOGPP is iteratively extended by integrating sequences of TNOLPP.The simulation and experimental results demonstrate that the proposed active SLAM framework is an effective,precise and safe autonomous exploration scheme.Finally,the paper summarizes the work and contribution,and the prospect in future work.