PVT3: a pruned video-vision transformer for tactile texture classification

• Main Author: Ouyang, Yanjia
• Other Authors: Lin Zhiping
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/158296
• Institution: Nanyang Technological University

With the newly involved technologies in tactile sensory, variants tactile sensors have been deployed on robots which provides them touching ability to perceive complex environments. One typical example of robot touching task is to recognize different materials based on the tactile data generated from different textures. In this report, we propose PVT 3 , a light-weight Transformer-based architecture with pruning layers to model the texture representation. By using a Video-Vision Transformer backbone, the spatial and temporal features will be well preserved and utilized. The multi-dimensional pruning layers will reduce model complexity and size without sacrificing the performance. Three tactile datasets are used for 3 testing the PVT model. Overall, our proposed model achieves higher accuracy on material classification results with a smaller model size compared to the state-of-the-art tactile texture models. This work was written as a paper and submitted to the International Conference on Intelligent Robots and Systems (IROS) 2022.