Research On Indoor Localization And Navigation Technology Of Mobile Robot

With the rapid development of technologies such as artificial intelligence and advanced manufacturing,mobile robots are playing an important role in more and more fields.Autonomous positioning and autonomous navigation are two major tasks of mobile robots,and the problem of robot kidnapping is a difficult point in localization technology.The Adaptive Monte Carlo Localization(AMCL)algorithm has low localization recovery efficiency in the kidnapping problem.The traditional navigation method relies on the environment map established by the high-precision mapping sensor throughout the process.In this thesis,the research on indoor localization and navigation technology of mobile robots is carried out in response to the abovementioned shortcomings.Based on the Adaptive Monte Carlo Localization algorithm and combined with the idea of template matching in image science,this thesis proposes an Adaptive Monte Carlo Localization algorithm based on fast affine template matching(Adaptive Monte Carlo Localization-Fast Matching,AMCL-FM).Although AMCL based on particle filtering has been able to solve the robot kidnapping problem,the method of generating new particles at random positions in the localization recovery process causes the effectiveness of the new particles to be random.The ACML-FM algorithm proposed in this thesis uses the global cost map and the local cost map to estimate the true position of the robot,and then places new particles at the estimated position,which improves the effectiveness of the new particles.Compared with AMCL,experiments show that the algorithm proposed in this thesis can speed up the convergence rate of the particle swarm and improve the localization recovery efficiency when the mobile robot encounters a kidnapping.The traditional navigation method based on Simultaneous Localization and Mapping(SLAM),global path planner and local planner relies on the environment map established by the use of high-precision mapping sensors throughout the whole process.Aiming at the above shortcomings,this thesis proposes a navigation model based on deep reinforcement learning algorithm named Soft Actor-Critic(SAC),which combines deep learning and reinforcement learning technology.During the navigation process,the speed of the mobile robot,the relative position of the navigation destination and the radar data are used as the input and output the action that the mobile robot should perform in the next step is to repeat the process until it reaches the target point,which realizes the map-free autonomous navigation of the mobile robot.Experiments show that the proposed navigation model has a certain versatility and can successfully complete navigation tasks in an unexplored environment.It can reduce the dependence on high-precision mapping sensors and the cost of mapping work in the practical application of mobile robots.