A hybrid elastic-hyperelastic approach for simulating soft tactile sensors

Efficient robotic grasping increasingly relies on artificial intelligence (AI) and tactile sensing technologies, which necessitate the acquisition of substantial data-a task that can often prove challenging. Consequently, the alternative of generating tactile data through precise and efficient simulations is becoming increasingly appealing. A significant challenge for simulating tactile sensors is balancing the trade-off between accuracy and processing time in simulation algorithms and models. To address this, we propose a hybrid approach that combines elastic and hyperelastic finite element simulations, complemented by convolutional neural networks (CNNs), to generate synthetic tactile maps of a soft capacitive tactile sensor. By leveraging a dataset of 53,400 real-world tactile maps, this methodology enables effective training, validation, and testing of each pipeline. This approach combines a fast elastic model for simple contact patches with a more detailed but slower hyperelastic model when greater precision is required. Our method automatically assesses contact patch complexity based on parameters associated with the object's mesh to determine the most appropriate modeling technique by still ensuring accurate deformation simulation. Tested on a dataset of 12 unseen objects, our approach achieves up to 97% Structural Similarity Index Measure (SSIM) for the hyperelastic model and 90% for the elastic model. This hybrid strategy enables an adaptive balance between simulation speed and accuracy, making it suitable for generating synthetic tactile data across tasks with varying precision demands and object geometrical complexities.