| From industrial robots that are anchored to a specific position to the autonomous mobile robots that are able to travel through various environments,the mobility,environment adaptability and intelligence of robots have been improved remarkably with the list of applications for robots keeps growing in the past decades.The navigation technology in structured environments has increasingly turned mature and paced into industrial phase at present.However,with the expansion of applications,the working scenes of mobile robots becomes more complex,most of which are dynamic,unknown and unstructured.In the mainstream framework of robot navigation,high-level tasks given by human users are translated into simple and low-level tasks that the robot can understand,including localization,environment modeling,behavior planning and motion control.Due to the various interference factors in the environment,the mobile robot has to recover an accurate estimate of its own state and the environment model from the limited uncertain information,and then remember,reason,decide and act to achieve safe navigation to the target position.For the localization and motion control problems of mobile robots,this thesis draws the advantages of uncertain information processing methods including grey system theory,probabilistic statistics,and interval analysis,and proposes some new localization and control algorithms in a human way to improve the safety,robustness and environment adaptability of the mobile robots.The main work and contribution of this thesis include:(1)In view of the disadvantages of the classic points based scan matching approach for mobile robot pose estimation,this thesis proposes a robust point cluster based qualitative scan matching method to estimate the relative transformation of the robot pose over time,which consists of two separate stages,i.e.hierarchical clustering and point cluster matching.During the hierarchical clustering,the whole laser scan is first segmented into several continuous subsets on the basis of the continuity of surrounding surface,and then each subset is further divided into different clusters.Since the points in a cluster can be approximated by a Normal distribution,the discrete view of the environment is transformed into a continuous and compact representation with clustering analysis,which is more robust to robot's position and view-point.In the cluster matching stage,a pair of stable clusters in current observation is used as heuristic information to establish the corresponding cluster pairs in the reference observation by checking the compatibility between associated cluster pairings.Qualitative analysis is then employed to compute the relative transformation between associated cluster pairs and estimate the relative robot pose.Experimental results show that the proposed approach can provide accurate pose estimation results that are more robust to observation noises and initial pose errors.(2)A novel global localization approach based on particle swarm optimization algorithm and particle filter is proposed.The global localization problem proceeds in two stages called initial pose estimation and multiple-pose hypothesis tracking.In the first stage,particles are uniformly distributed in the free space of the environment in order to find the correct initial pose.Particles in a local neighborhood realize a joint effort to search for the best solution from iteration to iteration until the stopping criterion is met.As the density of particles near to the local extremum is higher than the other region in the solution space,a density based clustering technique is conducted to search for the pose hypotheses,each of which is associated with a subset of particles.In the second stage,each subset progresses as the robot moves,and particles in each subset iteratively searches for the best pose estimate for each hypothesis.As new observations are captured,the importance weights of particles in wrong pose hypotheses decrease,and they are discarded gradually through resampling until the whole population converges to the correct pose hypothesis.Experimental results on a public dataset illustrate that the proposed approach is robust to the ambiguity of the initial robot pose and the great motion uncertainty,and it is able to provide accurate pose estimation for both initial pose estimation and pose tracking.(3)In order to improve the safety of mobile robots during navigation,a velocity space based trajectory tracking approach is proposed for the tracking problem in dynamic environment.Affected by unmodeled dynamic obstacles in the environment,the control command at the next moment cannot be determined in advance.However,due to the input saturation constraints,the speed that the robot can reach in the next control cycle is within a limited area.Assuming that the robot speed changes linearly or remains constant during a control cycle,it is possible to predict the trajectory of the robot using the robot's motion model with a given control.Considering the tracking error and the distance of robot to the nearest obstacle,an evaluation function can be defined to find the optimal solution in the velocity space,which is used to control the robot in the next cycle.Simulations show that the proposed method has a fast tracking response speed,and it is able to keep smaller tracking errors while avoiding obstacles,since both the tracking errors and the moving obstacles are considered during trajectory tracking. |
PDF Full Text Request