基于极限学习机与模糊积分融合的机器人地面分类

Terrain classification for mobile robots based on extreme learning machine and fuzzy integral fusion

为了提高移动机器人地面分类的准确率,采用奇异值分解和功率谱密度估计两种方法对振动信号进行特征提取。针对极限学习机的隐层节点冗余问题,给出改进的算法,并采用改进的极限学习机对地面分类。针对模糊积分参数耗时和积分函数不确定的问题,给出改进的方法,并基于2种特征采用改进的模糊积分对2个改进的极限学习机进行融合。在四轮移动机器人左前轮轮臂上安装三向加速度计和z向传声器,使之在沙、碎石、草、土、沥青地面上分别以5种速度行驶,采集车轮与地面相互作用的加速度和声压信号。根据改进的极限学习机和模糊积分融合算法,分别对每种速度下的5种地面进行分类,分类平均准确率为95.22%。实验验证了算法的有效性。

To enhance the accuracy of terrain classification in mobile robots,we applied two feature extraction methods-singular value decomposition and power spectrum density-to the vibration signals.With respect to the redundancy in the hidden-layer nodes of extreme learning machines(ELMs),we proposed an improved ELM for classifying the terrains.With respect to the time-consuming parameter and uncertain integral function in fuzzy integral solutions,we also proposed an improved fuzzy integral fusion algorithm for the two improved ELMs based on two features.To acquire the acceleration and sound pressure signals of the wheel-terrain interaction,we used a fourwheeled mobile robot equipped with a three-direction accelerometer and a microphone in the z direction on the leftfront wheel arm.This robot then traversed sand,gravel,grass,soil,and asphalt terrains at five different velocities,respectively.We classified each of the five terrains at each velocity with the improved fuzzy integral fusion algorithm of the two improved ELMs and achieved an average classification accuracy of 95.22%.The proposed algorithms have been validated by corresponding experiments.