PaperlayerView exact revision on arXiv

Source-linked AI summary

VirSqueezer: Generating Realistic Deformations and Squeezing Dynamics in VR from Fine-Grained Squeezing Controls

Qian Zhang, Xiaoming Chen, Xiaorui Ma, Haisheng Li, Weidong Cai

arXiv:2609.01698v1cs.HC

TL;DR

Existing VR content-creation methods have limited ability to simulate fine-grained squeezing effects, including localized deformation and complex dynamics. VirSqueezer uses SenseGlove controls, physics simulation, and conditioned generative models to produce synchronized squeezing effects. The framework demonstrates realistic deformations, convincing dynamics, and consistency with fine-grained controls, while real-time generation and object-specific physical inference remain limited.

  • Problem

    Existing 3DGS-based and diffusion-based methods are limited in simulating fine-grained virtual squeezing effects such as localized deformation, rupture, and overflow.

  • Method

    VirSqueezer captures finger-level squeezing controls with a SenseGlove, maps them to physics-based deformation, and conditions generative models for synchronized secondary effects.

  • Results

    VirSqueezer generates realistic localized deformations and complex squeezing dynamics while maintaining consistency with fine-grained squeezing controls.

  • Takeaways & Limitations

    VirSqueezer extends VR content creation toward expressive, user-driven generation of dynamic virtual content beyond predefined animations and conventional deformation simulation.

  • Takeaways & Limitations

    Real-time generation remains beyond current practical capabilities because complex squeezing dynamics rely on computationally intensive generative models.

Abstract

from arXiv · show

Squeezing is one of the most natural forms of hand manipulation, inherently involving fine-grained, temporally evolving, per-finger flexion. In VR content creation, squeezing plays a unique role in enabling particular visual effects such as localized deformations and dynamic behaviors, e.g., bursting a Coke can or juicing a fruit, thereby expanding the expressive possibilities of VR content. However, existing techniques, such as 3D Gaussian splatting-based methods and diffusion-based video generation models, are limited in their ability to simulate fine-grained virtual squeezing effects. We introduce VirSqueezer, a framework designed to generate both localized deformations (primary effects) and complex squeezing dynamics, such as rupture and overflow (secondary effects). VirSqueezer captures squeezing control signals using a SenseGlove and provides the user with inferred resistance force feedback during the squeezing process. By estimating object contact areas, inferring physical properties, and simulating physical responses, VirSqueezer computes conditions that guide generation models for visual effect generation, ensuring both visual coherence and temporal synchronization with the simulation. Consequently, VirSqueezer enables the generation of physically realistic visual effects directly from continuous, fine-grained squeezing control signals. Our extensive evaluation demonstrates VirSqueezer's ability to reproduce realistic localized deformations, generate convincing visual dynamics, and maintain consistency in fine-grained squeezing controls.

1 INTRODUCTION

VirSqueezer addresses the difficulty of generating fine-grained squeezing effects by linking continuous finger-level controls to localized physical deformation and complex, temporally coherent visual dynamics.

  • Motivation: Squeezing combines temporally evolving per-finger flexion with localized deformation and effects such as rupture and overflow.These capabilities support interactions including bursting cans, crushing soft toys, and juicing fruit.
  • Limitations of Existing Methods: Existing 3DGS-based and generative techniques struggle to synchronize fine-grained controls with localized deformation and complex squeezing dynamics.MPM-based methods capture global deformation but have difficulty with localized effects, rupture, and overflow.
  • Framework: VirSqueezer captures per-finger flexion and motion trajectories with a SenseGlove and maps them to local object deformation through contact-area estimation and physics modeling.The framework also infers material properties and renders resistive force feedback through the SenseGlove.
  • Contributions: VirSqueezer is presented as the first framework specifically designed for fine-grained virtual squeezing, bridging continuous controls, localized deformation, and dynamic visual effects.The framework is intended to expand expressive, user-driven VR content creation.
  • Dynamic Effects: A physics-guided control diffusion method generates secondary effects such as rupture and overflow while maintaining visual coherence and temporal synchronization with simulation.Generative models are conditioned on physics-based deformation signals and user squeezing behavior.
  • Framework: VirSqueezer integrates Dirichlet boundary conditions with MPM to model realistic squeezing deformations and more accurate material responses.The approach jointly considers finger-level controls, localized contact, and object material properties.

2 RELATED WORK

Related work combines efficient 3DGS representations, physics-based MPM simulation, and diffusion generation, but existing approaches remain limited in fine-grained squeezing control and complex dynamics.

  • 3DGS-based Frameworks and MPM-based Methods: 3DGS enables efficient and realistic static scene reconstruction and has supported scalable VR content creation and editing.Subsequent work improves representation and motion modeling for dynamic content.
  • 3DGS-based Frameworks and MPM-based Methods: Combining MPM with 3DGS supports physical simulation of 3DGS objects and high-quality motion synthesis.These methods incorporate physics-based dynamics into Gaussian representations.
  • 3DGS-based Frameworks and MPM-based Methods: 3DGS- and MPM-based methods often require manual physical-parameter configuration and primarily focus on global object motion.They lack support for fine-grained squeezing deformations and complex visual dynamics.
  • Diffusion-based Models: Diffusion-based models provide scalable content creation and high-fidelity dynamic generation but offer limited physical realism.Many do not directly simulate precise physical interactions and instead use diffusion models to represent motion.
  • Diffusion-based Models: Existing video-generation approaches can edit 3DGS visual effects and motion, but commonly rely on prompts rather than fine-grained user-specific controls.Consequently, they cannot drive localized deformations through continuous squeezing signals.

3 METHOD

VirSqueezer maps continuous, fine-grained finger and hand squeezing controls into localized deformation constraints, simulates physical responses, and uses generated conditions to synthesize synchronized secondary dynamics.

  • Squeezing Control Mapping: SenseGlove signals synchronize five-finger flexion, fingertip positions, and palm poses before contact-area estimation and finger mapping produce deformation constraints.Signals are aligned to the SenseGlove sampling rate and smoothed with an exponential moving average.
  • Squeezing Control Mapping: Per-finger flexion is converted into time-varying force coefficients using activation thresholds, nonlinear sensitivity, and finger-specific baseline weights.The coefficients represent de-biased flexion responses and support different squeezing stages through bidirectional activation.
  • Squeezing Control Mapping: SC-CAE estimates finger-oriented contact patches by jointly enforcing geometric compatibility, normal alignment, curvature regularization, scale priors, and multi-finger coupling.The estimated contact area and force coefficients are combined into a continuous surface displacement field and converted into Dirichlet boundary constraints.
  • Squeezing-Driven Deformation: The MPM solver projects Dirichlet constraints onto a grid and advances squeezing deformation through particle-to-grid, grid-update, and grid-to-particle steps.The constitutive model updates stress and deformation, while the material-law correction helps determine whether squeezing produces primary or secondary effects.
  • Squeezing-Driven Deformation: An LLM infers material properties and object-specific squeezing thresholds, with threshold exceedance triggering effects such as rupture or overflow.Internal particles model object structure and are released opposite the squeezing deformation when the threshold is reached.
  • Conditional Multi-View Dynamic Effects Generation: VirSqueezer conditions secondary-effect generation with refined particle masks, normalized depth maps, RGB frames, and textual prompts for visual generation models.The resulting conditions guide multi-view generation while preserving spatial and temporal consistency with the simulated scene.

4 EVALUATION

The evaluation measures VirSqueezer against MPM-based and diffusion-based baselines using visual-quality and physical-realism comparisons, user studies, LLM assessments, and custom metrics.

  • Quantitative and Human Evaluation: Experiments compare VirSqueezer with MPM-based and diffusion-based baselines on visual quality and physical realism.The evaluation also includes a human user study, assessments by multiple LLM models, and custom metrics tailored to squeezing effects.

4.1 Quantitative Evaluation

VirSqueezer is evaluated across general video quality, physical commonsense, human and LLM judgments, and squeezing-specific control metrics using diverse objects and baselines. It achieves strong performance across these complementary evaluations.

  • Evaluation setup: The evaluation spans eight objects with diverse elastic, plastic, compressible, fluid-filled, and thin-shell characteristics.The objects include a can, plush toy, rubber duck, clay, tomato, orange, rubber ball, and paper cup.
  • Baselines: VirSqueezer is compared with MPM-based methods PhysGaussian and OmniPhysGS, and diffusion-based methods DreamGaussian4D and Gaussians-to-Life.
  • VBench evaluation: VirSqueezer outperforms all baselines in VBench imaging quality and motion smoothness, demonstrating suitability for fine-grained squeezing-controlled dynamic generation.The authors caution that this result does not establish general superiority in video generation.
  • Physical commonsense evaluation: VirSqueezer achieves the highest Physical Commonsense score among MPM methods and significantly outperforms diffusion-based baselines.
  • LLM evaluation: In LLM evaluations, VirSqueezer surpasses all baselines except PhysGaussian in visual quality and leads in motion realism and motion consistency.The reported GPT-5 results complement human assessments, while additional results use Gemini-3.1, Grok-4.1, and Manus-1.6.
  • Squeezing-controlled deformation: The study introduces BDC-T, BDC-S, SDC, and mIoU to measure temporal and spatial deformation consistency and squeezing-deformation alignment.BDC scores are normalized to [0,1], with higher values indicating better performance.

4.2 Ablation Study

The ablation studies examine Dirichlet boundary constraints and physics-guided conditions as components of VirSqueezer’s deformation and secondary-effect generation pipeline. Removing either component reduces supported deformation or visual-dynamics metrics.

  • Dirichlet boundary constraints: Removing Dirichlet boundary constraints lowers average BDC-T, SDC, and mIoU while leaving BDC-S comparable.
  • Dirichlet boundary constraints: The comparison uses PhysGaussian as a force-driven MPM baseline without explicit boundary constraints.
  • Physics-guided conditions: Removing G-buffer physics-guided conditions degrades Physical Commonsense, Motion Smoothness, Motion Realism, and Motion Consistency.The ablated model supplies only textual prompts to the generative models, omitting RGB frames, depth maps, and internal particle masks from simulation.
  • Physics-guided conditions: The G-buffer conditions constrain generated secondary effects with physical deformation, improving temporal coherence and physical plausibility.

4.3 Qualitative Results

VirSqueezer produces localized deformations and complex squeezing dynamics from fine-grained controls, outperforming qualitative baseline behavior in control fidelity, physical plausibility, and geometric consistency.

  • Demonstration of Deformation Effects and Complex Squeezing Dynamics: VirSqueezer demonstrates primary deformations across diverse objects, including rebound or non-rebound release behaviors after squeezing force is reduced.
  • Demonstration of Deformation Effects and Complex Squeezing Dynamics: Threshold-exceeding squeezes generate secondary effects such as cola ejection from soda cans and cotton stuffing spilling from plush toys.
  • Overall Comparison on Deformation Effects and Complex Squeezing Dynamics: MPM-based baselines generate physical deformations but do not accurately follow fine-grained controls or reproduce rupture and overflow.
  • Overall Comparison on Deformation Effects and Complex Squeezing Dynamics: Diffusion-based baselines synthesize dynamic effects, but their motions and geometries are less consistent with squeezing controls and physical deformation.
  • Extended Comparison on Complex Squeezing Dynamics: VirSqueezer preserves the original squeezing deformation while generating visually accurate and realistic complex dynamics.

5 LIMITATION

VirSqueezer remains constrained by computational cost, uncertain object-specific physical-property inference, limited interaction scope, and incomplete experimental coverage.

  • Real-time generation remains impractical because complex squeezing dynamics rely on computationally intensive generative models.
  • LLM-inferred physical properties and squeezing thresholds may not accurately characterize specific objects because they rely primarily on general material knowledge.
  • The current framework and evaluation primarily cover single-object squeezing and a predefined range of deformation and secondary effects.
  • Broader validation across object categories, material types, squeezing behaviors, and participants is needed to assess robustness and generalizability more comprehensively.

6 CONCLUSION

VirSqueezer maps continuous finger-level controls to physics-based deformations and generated secondary effects for 3DGS objects in VR. The framework supports more expressive, user-driven dynamic content beyond predefined animations and conventional deformation simulation.

  • SenseGlove captures finger flexion, fingertip positions, and palm poses, which are mapped onto 3DGS surfaces to construct Dirichlet boundary conditions for MPM deformation.
  • User squeezing controls condition generative models to synthesize secondary effects such as rupture and overflow alongside realistic primary deformations.
  • By bridging finger-level controls with physically grounded deformation and generative effects, VirSqueezer extends VR content creation toward more expressive user-driven dynamic content.
Loading 2609.01698v1…