To solve the problem of AGV obstacle avoidance in intelligent warehousing, the method with lidar to identify the obstacle category in front of AGV was proposed to make reasonable decision by combining obstacle position information to assist AGV. The lidar data was filtered and clustered to obtain clusters with high purity. The feature vector was extracted by the proposed feature extraction method. The particle swarm optimization (PSO) algorithm was used to find the optimal parameters of the radial basis kernel (RBF) support vector machine (SVM) on the training set, and the model was trained. The method was tested on the data set of the intelligent warehousing simulation environment. The results show that the accuracy reaches 94.58%, which can accurately and effectively identify the categories of obstacles in front of the AGV.