Data-driven texture modeling and rendering has pushed the limit of realism in haptics. However, the lack of haptic texture databases, difficulties of model interpolation and expansion, and the complexity of real textures prevent data-driven methods from capturing a large variety of textures and from customizing models to suit specific output hardware or user needs. This work proposes an interactive texture generation and search framework driven by user input. We design a GAN-based texture model generator, which can create a wide range of texture models using Auto-Regressive processes. Our interactive texture search method, which we call preference-driven, follows an evolutionary strategy given guidance from user's preferred feedback within a set of generated texture models. We implemented this framework on a 3D haptic device and conducted a two-phase user study to evaluate the efficiency and accuracy of our method for previously unmodeled textures. The results showed that by comparing the feel of real and generated virtual textures, users can follow an evolutionary process to efficiently find a virtual texture model that matched or exceeded the realism of a data-driven model. Furthermore, for 4 out of 5 real textures, 80% of the preference-driven models from participants were rated comparable to the data-driven models.

Haptics plays an important role in training users to assemble mechanical components such as airplane or car parts. Because mechanical components are often geometrically complex, efficient collision detection and 6-DoF haptic rendering of contact are required for virtual assembly, and this has been extensively explored in prior work. However, as our work shows, this alone is not sufficient for efficient virtual assembly training. Our work asks how to augment 6-DoF haptic rendering of contact to maximize virtual assembly training efficiency, and proposes and measures several visual and haptic guidance strategies. Our visual strategies consist of displaying animations of the correct assembly path, motion indicator cues, and close-ups on difficult assembly path sections. Our haptic guidance consists of forces and torques that correct the trainee's deviation from the path. We investigate several haptic guidance strategies, including continuous forces and torques, force/torque nudging and anti-forces/torques. We designed a user study to evaluate the training efficiency of our proposed strategies quantitatively, using ANOVA and Tukey statistics. Our main finding is that the most efficient training approach is to use haptic rendering of contact in combination with visual animation-based guidance. Continuous forces, nudging, anti-forces and motion indicator cues were measured to be less effective.

We present a hard-real-time multi-resolution mesh shape deformation technique for skeleton-driven soft-body characters. Producing mesh deformations at multiple levels of detail is very important in many applications in computer graphics. Our work targets applications where the multi-resolution shapes must be generated at fast speeds ("hard-real-time", e.g., a few milliseconds at most and preferably under 1 millisecond), as commonly needed in computer games, virtual reality and Metaverse applications. We assume that the character mesh is driven by a skeleton, and that high-quality character shapes are available in a set of training poses originating from a high-quality (slow) rig such as volumetric FEM simulation. Our method combines multi-resolution analysis, mesh partition of unity, and neural networks, to learn the pre-skinning shape deformations in an arbitrary character pose. Combined with linear blend skinning, this makes it possible to reconstruct the training shapes, as well as interpolate and extrapolate them. Crucially, we simultaneously achieve this at hard real-time rates and at multiple mesh resolution levels. Our technique makes it possible to trade deformation quality for memory and computation speed, to accommodate the strict requirements of modern real-time systems. Furthermore, we propose memory layout and code improvements to boost computation speeds. Previous methods for realtime approximations of quality shape deformations did not focus on hard real-time, or did not investigate the multi-resolution aspect of the problem. Compared to a "naive" approach of separately processing each hierarchical level of detail, our method offers a substantial memory reduction as well as computational speedups. It also makes it possible to construct the shape progressively level by level and interrupt the computation at any time, enabling graceful degradation of the deformation detail. We demonstrate our technique on several examples, including a stylized human character, human hands, and an inverse-kinematics-driven quadruped animal.

The elastic parameters of soft tissues are important for medical diagnosis and virtual surgery simulation. In this study, we propose a novel nonlinear parameter estimation method for soft tissues. Firstly, an in-house data acquisition platform was used to obtain external forces and their corresponding deformation values. To provide highly precise data for estimating nonlinear parameters, the measured forces were corrected using the constructed weighted combination forecasting model based on a support vector machine (WCFM_SVM). Secondly, a tetrahedral finite element parameter estimation model was established to describe the physical characteristics of soft tissues, using the substitution parameters of Young’s modulus and Poisson’s ratio to avoid solving complicated nonlinear problems. To improve the robustness of our model and avoid poor local minima, the initial parameters solved by a linear finite element model were introduced into the parameter estimation model. Finally, a self-adapting Levenberg–Marquardt (LM) algorithm was presented, which is capable of adaptively adjusting iterative parameters to solve the established parameter estimation model. The maximum absolute error of our WCFM_SVM model was less than 0.03 Newton, resulting in more accurate forces in comparison with other correction models tested. The maximum absolute error between the calculated and measured nodal displacements was less than 1.5 mm, demonstrating that our nonlinear parameters are precise.