未知观测噪声时机器人同步定位与地图构建

Robot simultaneous localization and mapping with unknown observation noise

对未知观测噪声的机器人同步定位与地图构建问题,提出基于神经网络PID自适应学习观测噪声的机器人同步定位与地图构建算法.已知系统噪声为高斯分布,噪声的方差未知,但其真值是在某个有限集合内.设计一个由神经网络PID控制器、观测噪声调整以及中值滤波构成的噪声在线辨识单元.通过自适应在线辨识观测噪声,并进行新息协方差平均值滤波,迭代修正观测噪声协方差,实现机器人同步定位精度的在线提高.实验表明,该算法可降低观测噪声先验信息不足的影响,减小定位误差.

To deal with the unknown observation noise for a robot in simultaneous localization and mapping （SLAM）, we propose an adaptive learning observation noise algorithm based on neural network PID for robot SLAM （ALON-SLAM）. In this algorithm, the system noise is with Gaussian distribution and the noise variance is unknown, but the true values of the noise are within a finite set. An online noise identification structure consisting of neural network PID controller, observation noise adjustor and median filter is developed. The observation noise covariance is adaptively learned online. Meanwhile, the innovation covariance is used to match the observation noise covariance and to iteratively revise it. Then, the estimation accuracy of the robot position is improved online. Experimental results show that this algorithm reduces the impact of the observation noise without prior knowledge, and lowers the positioning error.