Abstract
This paper presents the development and implementation of a human-like robotic hand integrated with advanced triboelectric nanogenerator (TENG) based tactile sensors for shape and material recognition. Meanwhile, traditional piezo sensors’ effectiveness is limited, sensitive to the temperature, and the manufacturing cost is high. TENG sensors offer a self-powered alternative with simplified circuitry, cost-effective fabrication, and enhanced durability. To capitalize on these benefits, we propose a novel machine learning approach that represents time-series data as two-dimensional images processed using a two-dimensional convolutional neural network (2D CNN). This method is compared against the traditional one-dimensional convolutional neural network (1D CNN) method. The research methodology encompasses TENG sensor preparation, noise cancellation, robotic hand design, and control electronics. Experimental results demonstrate that the proposed 2D CNN method significantly improves shape and material recognition accuracy, achieving 98% and 99%, respectively, compared to 94% and 98% with the 1D CNN method. Real-time evaluation further validates the robustness and adaptability of the proposed model in unstructured environments. These findings underscore the potential of integrating TENG sensors with advanced neural network architectures for autonomous dexterous manipulation in various industrial applications, paving the way for future advancements in robotic tactile sensing.