Research On Improved FastSLAM Algorithm And Its Application Based On ROS Platform

With the rapid development of the field of artificial intelligence,high degree of automation has brought a variety of convenient to life.Among them,the field of mobile robots due to its wide range of application scenarios is obtained the unprecedented development.Autonomous navigation is a prerequisite for mobile robots to perform various tasks in complex environments.Simultaneous Localization and Mapping(SLAM)is considered as a key technology to enable robots to have space exploration capabilities.In practical application scenarios,due to the interference of random noise and the nonlinearity of sensor observation model,the system needs to be able to track the state of the nonlinear stochastic dynamic system robustly.Therefore,how to design a core state estimation framework for SLAM to properly propagate the uncertainty of estimation has a great impact on the performance of SLAM.At present,the main mathematical framework can be divided into two categories: non-linear optimization and filter.Compared with the method of non-linear optimization,filter has certain advantages in system scalability and robustness.In some better designs,the filter algorithm based on local information fusion has real-time advantages under the same precision.Particle filter has been widely used in mobile robot navigation because of its excellent performance in reducing system modeling error and improving estimation accuracy.However,the particle filter SLAM method also has some shortcomings caused by particle filter.Among these defects,particle degradation and dilution seriously affect the performance of SLAM.In order to alleviate this problem,the gravitational field mechanism is used to improve the particle distribution during the resampling process.Gravitational field algorithm(GFA)is a heuristic search algorithm,which regards particle swarm optimization as a cosmic dust system.In the optimization process,the most adaptable particles in the state space are analogous to the central dust,which can exert gravitational and repulsive effects on the rest of the dust.When the distance between the surrounding dust and the central dust is larger than the preset threshold of the algorithm,the dust is subjected to the unidirectional gravitational action of the central dust,and each dust is subjected to the exclusion effect of the central dust due to its rotation.The location of each dust in the system is updated by fusing these two effects.The improved algorithm uses the moving factor to drive the particles to move to high likelihood region,and at the same time keeps the particles at a certain distance through the role of rotation factor,which enlarges the local search space of the particles and enhances the robustness of the tracking algorithm in case of state mutation on the basis of guaranteeing the estimation accuracy of the algorithm.Compared with the FastSLAM algorithm in the traditional framework,the improved algorithm has significant improvement in real-time performance and estimation consistency.The simulation results of MATLAB show that the estimation accuracy of GFA-FastSLAM2.0 algorithm is better than the result of EKF-SLAM and FastSLAM2.0 under the same noise level of interference.In order to further test the performance of the algorithm,a URDF model of lidar robot is built through Robot Operating System(ROS),and the improved algorithm is deployed to the ROS communication system.Finally,it is tested in the Gazebo simulation environment with physical properties to verify the practical feasibility of the improved algorithm.