 Hi everyone, I am Prey Darshani and today I am going to present my work enhancing haptics distinguishability of surface materials with boosting technique. The haptics sensation is crucial for many robotics applications. In recent year, many works in the haptics domain have focused on surface material classification. However, most of them rely on multimodal data for higher accuracy or high end haptic devices for recording the data. In this work, we propose a set of novel discriminative features CQFB for surface material classification. Additionally, we demonstrate the effectiveness of the metric-based feature transformation technique in enhancing the distinguishability of haptic signals. Our framework needs single modality acceleration data as an input and generalizes well for different predictors. The figure shows an overview of our algorithm. Our data consists of traces of acceleration signals of different surface materials such as a stone, wood, etc. In order to reduce the dimension of signal, we combine three components of acceleration signal using DFT321 technique. We extract the feature from signal spectrum by dividing them into n-beans using Gaussian filter of varying bandwidth. We further improve the distinguishability of haptic signals by transforming the extracted features by a metric-based linear transformation technique called boost metric. Our goal is to learn a linear feature transformation matrix M such that in the projected space signal from the same class form a compact cluster. The transformation matrix is obtained by solving this optimization function which scales the data in the embedding space such that signal from a very dissimilar group are placed far apart and dissimilar class signals are placed close together. We conduct various experiments to evaluate the accuracy and effectiveness of our algorithm. First we evaluate how effective our method is in improving the classification performance of haptic textures. The figure shows the performance improvement by the learned distance metric over the baseline Euclidean distance metric for different classifiers. As we see, there is a significant improvement in the classification performance with boosting embedding technique. We visualize the learned space using test-net plot. The left figure shows the input haptic signals in the original space and the middle figure shows the proposed CQ FB features projected in 2D space. As seen clearly, CQ FB features form a clear class-based clusters as compared to original haptic signals. On the right, the learned boosting-based features induced more compact cluster with low-interclass variance and high-interclass separation. To conclude, we proposed handcrafted spectral features to enable better discrimination of real-world surface textures. The learned boost metric is effective in improving distinguishability between different classes. Although our framework is effective in improving class-based separation, it is limited in mimicking the human-perceived dissimilarity between haptic signals. The limitation has been addressed in our follow-up work. Thank you.