← ArchivePaper2024
A Neural Network Model for Efficient Musculoskeletal-Driven Skin Deformation
Yushan Han, Yizhou Chen, Carmichael Ong, Jingyu Chen, Jennifer Hicks, Joseph Teran
Neural network approximates full musculoskeletal FEM skin deformation at interactive rates, trained on biomechanically accurate simulation data.
Abstract
We present a comprehensive neural network to model the deformation of human soft tissues including muscle, tendon, fat and skin. Our approach provides kinematic and active correctives to linear blend skinning [Magnenat-Thalmann et al. 1989] that enhance the realism of soft tissue deformation at modest computational cost. Our network accounts for deformations induced by changes in the underlying skeletal joint state as well as the active contractile state of relevant muscles. Training is done to approximate quasistatic equilibria produced from physics-based simulation of hyperelastic soft tissues in close contact. We use a layered approach to equilibrium data generation where deformation of muscle is computed first, followed by an inner skin/fascia layer, and lastly a fat layer between the fascia and outer skin. We show that a simple network model which decouples the dependence on skeletal kinematics and muscle activation state can produce compelling behaviors with modest training data burden. Active contraction of muscles is estimated using inverse dynamics where muscle moment arms are accurately predicted using the neural network to model kinematic musculotendon geometry.
How to read this
- Category
- Method: neural musculoskeletal skin deformation
- Contributions
- A neural network that models deformation of muscle, tendon, fat and skin as kinematic and active correctives to linear blend skinning.
- A layered data-generation scheme (muscle, then inner skin/fascia, then fat) producing quasistatic equilibria from physics-based hyperelastic simulation for training.
- A network that decouples skeletal-kinematics from muscle-activation dependence, plus inverse-dynamics muscle-activation estimation using NN-predicted musculotendon moment arms, at interactive cost.
- Context
- Builds on muscle-actuated human simulation (cf. Lee et al., 2019, Scalable Muscle-Actuated Human Simulation and Control) and on linear blend skinning (Magnenat-Thalmann et al., 1989) as the base it corrects.Builds on: Scalable Muscle-Actuated Human Simulation and Control
- Correctness
- Trained to approximate quasistatic equilibria from physics-based simulation, so accuracy is bounded by the training data and the quasistatic assumption (no dynamics); the decoupled-network design is reported to work with modest data but is an approximation of the full FEM solve.
- Clarity
- Technical; a first pass conveys the corrective-to-LBS idea and the layered pipeline, a second/third pass is needed for the inverse-dynamics and moment-arm modeling.
- How to read it
- Read for the layered training-data strategy and the LBS-corrective framing first; do a second and likely third pass on the inverse-dynamics muscle-activation estimation if you want to reproduce the biomechanics.
Builds on
Built upon by
Nothing yet.
Related work
- NeuroSkinning: Automatic Skin Binding for Production Characters with Deep Graph Networks 2019 / SIGGRAPH
- How to Build a Human: Practical Physics-Based Character Animation 2016 / DigiPro
- Data-Driven Physics for Human Soft Tissue Animation 2017 / SIGGRAPH
- NiLBS: Neural Inverse Linear Blend Skinning 2020 / arXiv
Keywords
This page summarises the entry and links to its original source. The archive never hosts or redistributes the publication itself.Show it in the full archive list →