基于姿态测量模块和闭环检测算法的仿生SLAM研究

Bio-Inspired SLAM Based on Gesture Measuring and Closed-Loop Detection

基于传感器模型的SLAM导航策略精度较高,但由于摩擦等因素误差长时间累计会造成里程计的漂移现象。依靠视觉里程计进行导航的Rat SLAM系统,通过在局部场景细胞中引入闭环检测策略实现累计误差的校正,在静态环境下具有一定的鲁棒性,但在复杂场景里,如移动障碍物的出现,视觉里程计会提取到错误的速度等姿态信息,导致航迹出现较大偏移,有时无法通过场景重定位进行闭环校正。结合两种模型的优势提出一种带姿态测量模块和闭环检测算法的仿生SLAM模型。实验表明,相较于仅带带闭环检测的Rat SLAM系统或仅由姿态测量模块构成的导航系统,本文提出的新系统更能适应长期复杂场景下的导航,且鲁棒性更强。

There has a relatively high accuracy on sensors in the Simultaneous Localization And Mapping（SLAM） problem, but odometer drift appears on account of accumulative errors for a long time, and this kind of accumulated error can be corrected by adding closed-loop detection to the local view cells, which is based on visual odometry of RatSLAM system, and this system has a good performance on robustness under the circumstance of static environ- ment. However, speed acquired from the visual odometry is incorrect in complex scenarios such as moving obstacles, which leads to large deviation during navigation, and sometimes this model is unable to relocate the surroundings and correct trajectory. Combined with the two algorithms above, bio-inspired SLAM based on gesture measuring and closed-loop detection is presented, and experiments show that in contrast to the RatSLAM system with only closed- loop detection or the system with only gesture measuring model, the new system proposed in this paper adapts to the navigation in a long time and has fairly better robustness.