TacWgan-gp: Fabric Texture-Based Tactile Friction Coefficient Generation

Generating tactile data from visual modality is an important technology in various fields, including haptic rendering, virtual reality, and robotics. This technique effectively bypasses the labor-intensive process of tactile data recording and addresses the limitations inherent in physical contact. However, existing generation methods exhibit limited consistency in generating multiple types of tactile data. To address this issue, we propose the TacWgan-gp cross-modal generation framework, which uses the WGAN-GP as the base, incorporates a center image condition constraint and L1 loss to regulate the generation quality. This framework is capable of generating spectrograms of friction coefficient signals from fabric texture images and ultimately converting them into one-dimensional friction coefficient signals using the Griffin-Lim algorithm. We utilize RMSE and MAE to quantify the differences between the generated spectrograms and reconstructed signals and their corresponding ground truth, and compare them with existing methods. The results show that our proposed method significantly outperforms existing methods in terms of generation accuracy and stability.