Texture Dataset Collected by Tactile Sensors

Tactile perception of the material properties in real-time using tiny embedded systems is a challenging task and of grave importance for dexterous object manipulation such as robotics, prosthetics and augmented reality [1-4] . As the psychophysical dimensions of the material properties cover a wide range of percepts, embedded tactile perception systems require efficient signal feature extraction and classification techniques to process signals collected by tactile sensors in real-time. We have limited our study to the machine perception/discrimination of various textures that can be sensed by sensors attached to a probe/stick touching the material surfaces via a single touch point. For this purpose, we developed two embedded systems, one that served as a vibrotactile stimulator system and one that recorded and classified the vibrotactile signals collected by its sensors. As the probe rubs against the surface of the textured material on the stimulator, the sensors attached to the probe capture the vibrotactile signals for real-time classification. The probe is 3D printed with high printing density so that it transmits the vibrations at its tip without distortion. Our study has been submitted under the title: “An Embedded System for Collection and Real-time Classification of a Tactile Dataset”, in which the data has been further elaborated and analyzed using the proposed signal feature extraction method and the Fourier transform as input to machine learning classifiers. We performed experiments both offline and on the proposed embedded platform in real-time. Based on the limited memory and performance budget of the embedded system used in this study, we have chosen the 3-dimensional accelerometer sensor (MMA-7660 from NXP Company [5]) and an electret condenser microphone (CMA-4544PF-W from CUI Company [6]) as the sources of recordings in our tactile dataset. We have used commercial off-the-shelf embedded boards and electrical components (AVR-based embedded boards, stepper motors, etc.) as well as our own designed and 3D printed mechanical components (including the rotating drum glued with different texture strips). The collected tactile dataset has 12 texture classes, including sandpapers of various grits, Velcro strips with various thicknesses, aluminum foil, and rubber bands of various stickiness. For each texture, 20 seconds of recordings are collected (corresponding to nearly five rotations of the drum). We used the sampling rate of 200Hz for the accelerometer to collect the vibration data and to 8kHz for the microphone to collect the sound data. We used the dataset to show that low-cost, highly accurate, and real-time tactile texture classification can be achieved on embedded systems using an ensemble of sensors, efficient feature extraction methods [7], and simple machine learning classifiers [8]. 1- J. C. Gwilliam, Z. Pezzementi, E. Jantho, A. M. Okamura, and S. Hsiao, “Human vs. robotic tactile sensing: Detecting lumps in soft tissue,” in 2010 IEEE Haptics Symposium, March 2010, pp. 21–28.2- S. Okamoto, H. Nagano, and HN. Ho, “Psychophysical Dimensions of Material Perception and Methods to Specify Textural Space,” In: Kajimoto H., Saga S., Konyo M. (eds) Pervasive Haptics. Tokyo: Springer Japan, 2016.3- W. Duchaine, “Why tactile intelligence is the future of robotic grasping,” in IEEE Spectrum Automaton. IEEE, 2016.4- A. Schmitz, Y. Bansho, K. Noda, H. Iwata, T. Ogata, and S. Sugano, “Tactile object recognition using deep learning and dropout,” in 2014 IEEE-RAS International Conference on Humanoid Robots. IEEE, 2014, pp. 1044–1050.5- “3-axis orientation/motion detection sensor,” NXP Semiconductor, Document Number: MMA7660FC, 2012. [Online]. Available: https://www.nxp.com/docs/en/data-sheet/MMA7660FC.pdf6- “Electret condenser microphone sensor,” CUI Devices, Document Number: CMA-4544PF-W, 2013. [Online]. Available: https://www.mouser.com/datasheet/2/670/cma-4544pf-w-1309465.pdf7- E. Alpaydin, “Introduction to machine learning, third edition,” The MIT Press, Cambridge, 20148- M. Fernandez-Delgado, E. Cernadas, S. Barro, and D. Amorim, “Do we need hundreds of classifiers to solve real world classification problems?” Journal of Machine Learning Research, vol. 15, pp. 3133–3181, 2014. [Online]. Available: http://jmlr.org/papers/v15/delgado14a.html

借助微型嵌入式系统（embedded systems）实时感知材料属性的触觉信息，是一项极具挑战性的任务，同时对于机器人、假肢以及增强现实等灵巧物体操控场景而言至关重要[1-4]。由于材料属性的心理物理维度涵盖了广泛的感知范畴，嵌入式触觉感知系统需要高效的信号特征提取（signal feature extraction）与分类技术，以实时处理触觉传感器（tactile sensors）采集到的信号。本研究仅聚焦于通过单触点接触材料表面的探针/触杆上的传感器所感知的各类纹理的机器感知与判别。为此，我们开发了两套嵌入式系统：一套作为振动触觉刺激器系统，另一套用于采集并分类其传感器所获取的振动触觉信号。当探针在刺激器上的纹理材料表面摩擦时，附着于探针的传感器会采集振动触觉信号以进行实时分类。该探针采用高打印密度3D打印制成，可无失真地传递其尖端的振动。我们的研究以"An Embedded System for Collection and Real-time Classification of a Tactile Dataset"为题提交，其中针对所提出的信号特征提取方法与作为机器学习分类器（machine learning classifiers）输入的傅里叶变换（Fourier transform），对数据集进行了进一步的细化阐述与分析。我们分别在离线环境与所提出的嵌入式平台上开展了实时实验。基于本研究中使用的嵌入式系统有限的内存与性能预算，我们选用了恩智浦（NXP）公司的MMA-7660型三维加速度传感器（3-dimensional accelerometer）[5]以及CUI公司的CMA-4544PF-W型驻极体电容麦克风（electret condenser microphone）[6]作为触觉数据集中的采集源。我们使用了商用现货（commercial off-the-shelf）嵌入式开发板（embedded boards）与电子元件（基于AVR的嵌入式开发板（AVR-based embedded boards）、步进电机（stepper motors）等），以及自主设计并3D打印的机械组件（包括粘贴有不同纹理条带的旋转滚筒）。所采集的触觉数据集包含12个纹理类别，涵盖不同粒度的砂纸、不同厚度的魔术贴（Velcro）条带、铝箔以及不同粘性的橡皮筋。针对每一种纹理，我们采集了20秒的记录数据（对应滚筒旋转约5圈）。我们以200Hz的采样率采集加速度计的振动数据，以8kHz的采样率采集麦克风的音频数据。我们利用该数据集验证了：借助多传感器集成、高效特征提取方法[7]以及简单的机器学习分类器[8]，可在嵌入式系统上实现低成本、高精度且实时的触觉纹理分类。1. J. C. Gwilliam、Z. Pezzementi、E. Jantho、A. M. Okamura与S. Hsiao，"Human vs. robotic tactile sensing: Detecting lumps in soft tissue"，收录于2010 IEEE触觉研讨会（IEEE Haptics Symposium），2010年3月，第21–28页。2. S. Okamoto、H. Nagano与HN. Ho，"Psychophysical Dimensions of Material Perception and Methods to Specify Textural Space"，收录于：Kajimoto H.、Saga S.、Konyo M.主编《普适触觉技术（Pervasive Haptics）》，东京：日本Springer出版社，2016年。3. W. Duchaine，"Why tactile intelligence is the future of robotic grasping"，收录于IEEE Spectrum Automaton，IEEE，2016年。4. A. Schmitz、Y. Bansho、K. Noda、H. Iwata、T. Ogata与S. Sugano，"Tactile object recognition using deep learning and dropout"，收录于2014 IEEE-RAS人形机器人国际会议（IEEE-RAS International Conference on Humanoid Robots），IEEE，2014年，第1044–1050页。5. "3-axis orientation/motion detection sensor"，恩智浦半导体（NXP Semiconductor），文档编号：MMA7660FC，2012年。[在线]. 可获取：https://www.nxp.com/docs/en/data-sheet/MMA7660FC.pdf6. "Electret condenser microphone sensor"，CUI设备公司（CUI Devices），文档编号：CMA-4544PF-W，2013年。[在线]. 可获取：https://www.mouser.com/datasheet/2/670/cma-4544pf-w-1309465.pdf7. E. Alpaydin，"Introduction to machine learning, third edition"，麻省理工学院出版社（The MIT Press），剑桥，2014年。8. M. Fernandez-Delgado、E. Cernadas、S. Barro与D. Amorim，"Do we need hundreds of classifiers to solve real world classification problems?"，《机器学习研究期刊（Journal of Machine Learning Research）》，第15卷，第3133–3181页，2014年。[在线]. 可获取：http://jmlr.org/papers/v15/delgado14a.html