Reduced-order Neural Modeling with Differentiable Simulation for High-Detail Tactile Perception

arXiv cs.RO / 5/7/2026

📰 NewsDeveloper Stack & InfrastructureIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper addresses the computational challenge of simulating high-resolution elastomer deformation for tactile perception, where FEM is accurate but expensive and MPM is memory-heavy.
It proposes a reduced-order neural simulation framework that combines coarse-grained MPM dynamics with an implicit neural decoder to reconstruct fine tactile details from compact latent states.
The model is trained on paired high- and low-resolution simulation data to learn a continuous deformation manifold, enabling physically consistent, differentiable inference.
Experiments report significant efficiency improvements versus TacIPC, including 65% faster simulation and 40% lower memory usage, while preserving (or improving) geometric fidelity.
The approach also boosts downstream tactile rendering and 3D surface reconstruction accuracy by 25% and generates realistic depth images and surface meshes with faster inference.

Abstract

Tactile perception is key to dexterous manipulation, yet simulating high-resolution elastomer deformation remains computationally prohibitive. Finite element methods (FEM) deliver high fidelity but demand costly remeshing, while Material Point Methods (MPM) suffer from heavy particle-memory tradeoffs. We propose a {reduced-order neural simulation framework} that couples coarse-grained MPM dynamics with an implicit neural decoder to reconstruct sub-particle tactile details from compact latent states. The framework learns a continuous deformation manifold from paired high- and low-resolution simulations, enabling physically consistent, differentiable inference. Compared to the TacIPC, our method achieves over 65\% faster simulation and {40\% lower memory usage}, while maintaining better geometric fidelity. In tactile rendering and 3D surface reconstruction, our methods further improve accuracy by 25\% and produce realistic depth images and surface mesh within a faster inference speed. These results demonstrate that the proposed reduced-order neural model enables high-detail, physically grounded tactile simulation with substantial efficiency gains for robotic interaction and optimization.