A large amount of information on tactile sequences can be collected by using a dexterous hand with a tactile sensor to grasp different objects. The abilities of a robot's environmental perception and dexterous manipula?tion are significantly improved after these tactile sequences are classified. Therefore, tactile sequences are separated into a series of subgroups and features are extracted by using a method based on the linear dynamical system ( LDS) . Since the features extracted by LDS are located in the non?Euclidean space, when dealing with these fea?tures, the Martin distance which is a measurement different from Euclidean distance is applied to represent the dis?tance between two LDS features, and the K?Medoid algorithm is used for clustering. Then, the codebook which is formed after clustering is used to represent the tactile sequence, the model of bag?of?system is formed, and the sup?port vector machine ( SVM) is used to classify these objects efficiently. Finally, a dataset based on 16 objects is used to evaluate the algorithm and the result of recognition is good, which proves this algorithm can be used in tact?ile sequences for object classification.