 Hello everyone, this is Julie from QNU South Korea. I am going to present our work that are driven haptic texture modeling and rendering based on deep spatio-temporal networks. Haptic texture rendering is a process of reproducing contact responses based on the haptic texture model of the surface. During tool surface interaction, the feedback is in the form of vibration that propagates through the tool. One of the critical issues with the existing approaches like LPC and WST is that they fail to intake face information fully. Besides, many methods do not consider scalability in input parameters. If the dimensions of an input parameter increases, the model fails to reproduce accurate output anymore. In this work, we introduce a deep spatio-temporal network to synthesize the acceleration signal, which is used in the form of feedback. The network is trained using contact acceleration data collected through our manual scanning stylus and interaction parameter, such as scanning velocities, directions and forces. The proposed network is composed of attention-aware 1DCNN and attention-aware encoder-decoder networks to adequately capture both the local special features and the temporal dynamics of the acceleration signals. We shall further augmented with attention mechanism that assign weights to the features according to their contributions. Previous generated sequence becomes the input of the network, which predicts the next acceleration signal. Therefore, during rendering, the trained network generates synthesized signal in real-time by taking the previous acceleration signal XN, scanning velocity V, applied force F and direction D. At first, we performed numerical evaluation to assess the performance of the proposed approach. Recorded and synthesized contact acceleration signal is showed in the left figure. The relative spectral RMS error is used as an error metric to compare synthesized and collected acceleration sequence. We compared our approach with simpler models as well as other state-of-the-art algorithms, which include air model, frequency decomposed neural network, wavelet segment table-based approaches. The result shows the effectiveness of the proposed approach over the state-of-the-art. In this work, we performed experiment on six haptic textures. Later on, we asked participants to rate the subjective similarity between the virtually rendered feedback and real feedback from the physical surface. The result shows that compared to the existing approaches, participants preferred the proposed approach, specially for the anisotropic textures. Furthermore, during interview, most of the participants mentioned that they are clearly able to distinguish the virtual textures from each other.