基于粒子滤波器的SLAM的仿真研究

机器人同时定位与精确地图创建能力是自主移动机器人的先决条件。SLAM的很多实现方法无法解决有大量环境特征的环境。应用粒子滤波器和卡尔曼滤波器分别估计机器人位姿和环境特征的后验概率分布。这个算法的基础是把后验概率分解成路径的后验概率和环境特征的后验概率分布。为避免衰竭问题,在粒子滤波器的重采样阶段,除了用权值选取粒子,还在更新阶段直接注入从传感器数据生成的少量粒子。仿真结果显示这个算法可以用100个粒子处理5000个环境特征的优越性。

Simulation Research on Simultaneous Robot Localization and Mapping Based on Particle Filter

The ability to simultaneous localization and mapping is a predetermination of antomomous robots. Now, few approaches can manage the environment with masses of landmarks. The posterior distribution over robot pose and landmark locations was estimated with paticle filter and Kalman Filter respectively. The key idea of the algorithm is factorafing of the Bayes filter into an estimation of robot path estimation and an estimation of landmarks. To avoid the depletion problem, the particle population was injected during the update phase with a small number of particles created directly from the sensor data as well as using the weights of the particles to decide which ones were going to be progated forward. Simulation results demonstrate 5000 landmarks with 100 particles can be handled successfully.