Real-Time Physics-Based Mesh Deformation with Haptic Feedback and

Material Anisotropy

Avirup Mandal

1 a

, Parag Chaudhuri

2 b

and Subhasis Chaudhuri

1 c

Department of Electrical Engineering, IIT Bombay, Powai, Mumbai, India

Department of Computer Science and Engineering, IIT Bombay, Powai, Mumbai, India

Keywords:

Virtual Sculpting, Haptic Feedback, Wetting, Deformation.

Abstract:

We present a real-time, physics-based framework to simulate porous, deformable materials and interac-

tive tools with haptic feedback that can reshape them. In order to allow the material to be moulded non-

homogeneously, we propose an algorithm to change the material properties of the object depending on

its water content. To enable stable visual and haptic feedback at interactive rates, we implement a multi-

resolution, multi-timescale solution. We test our model for physical consistency, accuracy, interactivity and

appeal through a user study and quantitative performance evaluation.

1 INTRODUCTION

Traditionally simulated virtual shape editing tools of-

fer a visual rendering of the object but entirely miss

the haptic aspect of it. Moreover, many of these

tools edit shapes in a purely geometric approach and

thus, are not physically accurate (De Goes and James,

2017). In this paper, we present a real-time, stable, in-

teractive, physics-based deformation framework with

faithful haptic feedback. We implement it as a multi-

resolution solution to be able to handle deformation of

high-resolution meshes. Our method works at mul-

tiple timescales to synchronise the haptic and visual

modes of interaction.

In the real world, an artist deforms a lump of clay

to sculpt it into a model. Adding water to the clay

makes it malleable, which in turn helps to reshape

parts of the same model differently. Our method

allows a user to perform these operations virtually

while receiving the appropriate haptic feedback for

the same. The major contributions of the work pre-

sented in this paper are as follows.

• Develop a stable, interactive, physics-based hap-

tic and visual simulation framework, with a multi-

resolution implementation.

• Develop a physically valid model for deformable,

porous soft volumetric objects.

https://orcid.org/0000-0002-2322-4440

https://orcid.org/0000-0002-1706-5703

https://orcid.org/0000-0002-1680-0016

• Modeling of objects with anisotropic elasticity

when parts of the material are made wet with

an interactive wetting tool. This allows the user

to deform the object differently in different parts

while applying the same force.

The rest of the paper is organized as follows. After

presenting a discussion on the related works, we de-

tail our simulation of variable elasticity produced by

material wetting. Next, we present full technical de-

tails of our haptic rendering solution for faithful force

feedback at interactive rates. Finally, we present qual-

itative, quantitative and user study results generated

using our framework.

2 RELATED WORK

In this section, we review the methods present in

the literature that are closely related to our work.

One of the most popular approaches to modelling

deformable objects is the Finite Element Method

(FEM). O’Brien (O’Brien et al., 2002), M

uller et

al. (M

uller and Gross, 2004) used FEM on tetrahe-

dral meshes with linear elasticity to model deformable

objects including plasticity and fracture. Non-linear

elasticity with the large plastic ﬂow is rendered in

more recent works by Bargteil et al. (Bargteil et al.,

2007), Irving et al. (Irving et al., 2004).

Fluid ﬂow and material wetting is a well-studied

subject in material physics (Bear, 1972) (Scriven,

Mandal, A., Chaudhuri, P. and Chaudhuri, S.

Real-Time Physics-Based Mesh Deformation with Haptic Feedback and Material Anisotropy.

DOI: 10.5220/0011611800003417

In Proceedings of the 18th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2023) - Volume 1: GRAPP, pages

153-161

ISBN: 978-989-758-634-7; ISSN: 2184-4321

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

153

1994). In computer graphics, work by Patkar et

al. (Patkar and Chaudhuri, 2013) offers a geometri-

cally modelled solution to the absorption, diffusion

and dripping of water in porous materials. In more

recent work (Fei et al., 2018), the wetting of different

kinds of clothes is explored. The change of material

properties due to ﬂuid absorption is well investigated

in material science (Yoon and Cowin, 2009) (Schraad,

2014). Here the authors hypothesize a mathematical

relationship between object elasticity with ﬂuid con-

tent and verify their hypothesis with empirical results.

Zilles and Salisbury (Zilles and Salisbury, 1995)

present God Object-based rendering where a god ob-

ject is constrained to stay on the surface of the mesh

object while a haptic proxy penetrates into the ob-

ject and the difference of their acceleration generates

haptic feedback. Ortega et al. (Ortega et al., 2007)

extended it to all six degrees of freedom. Another

broad category of haptic force rendering is penalty-

based rendering (Barbi

c and James, 2009) (Otaduy

and Lin, 2005) (McNeely et al., 1999). Here (Xu

and Barbi

c, 2017), the colliding objects penetrate

each other and force feedback is rendered depending

on the depth of penetration. Discrete penalty-based

rendering suffers from discontinuous and jerky force

feedback when the contact stiffness is high. These

problems are circumvented using continuous colli-

sion detection (Tang et al., 2012) (Xu and Barbi

2017). Even though constrained-based methods are

slightly more robust against the pop-through effect

of the proxy, we opted for the continuous penalty-

based method for our haptic feedback as it produces

smoother force feedback (Xu and Barbi

c, 2017).

Notable works in virtual shape editing include

methods presented in (Blanch et al., 2004) (Chen and

Sun, 2002) which build a rigid model using a small

cubic grid-based ﬁeld. These methods do not emulate

physically accurate material behaviour and are time-

consuming. In the works (Gunn, 2006) (Dachille

et al., 1999), the authors present frameworks that

deform polygonal mesh based on mass-spring-based

models in a strictly geometric way. The key draw-

back of all these existing works is that none of them

preserves physical plausibility. Moreover, the haptic

feedback provided in all these works is based on dis-

crete collision handling which suffers from jitters. We

use continuous collision-based smooth haptic feed-

back to tackle this problem. The need for physically

realistic virtual sculpting has been explored recently

by De Goes et al. (De Goes and James, 2017). Using

Kelvinlets (fundamental solutions of linear elasticity

for singular loads), they render the accurate mesh de-

formation in real time, but their work lacks the aspects

of haptic feedback.

Θ, t=0

Θ, t=t'

Figure 1: Material (left) coordinate to world (right) coordi-

nate.

In more recent work (Mandal et al., 2022) we ﬁnd

a method for cutting meshes with accurate physics-

based simulation and haptic feedback. Our work com-

plements this work, as we present a method to deform

the mesh while simulating the physics of wetting and

consequent material anisotropy.

We present a framework to efﬁciently reshape

meshes in a physically realistic manner with smooth

haptic feedback. Additionally, real-life objects have

anisotropic elasticity that can be modelled implicitly

using wetting in our framework.

3 DEFORMABLE POROUS

OBJECTS

In this section, we will brieﬂy describe the modelling

of a deformable object using Cauchy’s linear strain

model and then discuss how we model the change of

elasticity due to the wetting of the object.

We use a standard ﬁnite element discretization to

solve the governing differential equations of a de-

forming object (M

uller and Gross, 2004) (Erleben

et al., 2005). Let Θ ∈ R

be a three-dimensional do-

main which is discretized into a mesh of n

tet

tetrahe-

dra. The number of nodes shared by these tetrahedral

elements is n

. As shown in Figure 1, a displacement

function, u : Θ× [0,∞) −→ R

is a mapping of a ma-

terial point ζ ∈ Θ at time t ∈ [0,∞) to its deformed

location x in the world space.

u(ζ,t) =

∑

i=1

(ζ)u

(t), ∀ζ ∈ Θ (1)

where N(ζ) and u

(t) represents the shape function

and displacement vector at the node i respectively.

The system dynamics of a deformed object can then

be written in Lagrange’s form as

u + f

int

= f

ext

, u =



...u



(2)

where M, f

ext

, f

int

are respectively mass matrix, exter-

nal and internal element force vector of the full sys-

tem. Plastic ﬂow is also enforced in this model when

the strain exceeds a yield threshold as presented in the

work by M

uller et al. (M

uller and Gross, 2004).

For wetting the material, we follow the method

proposed in (Patkar and Chaudhuri, 2013) barring the

GRAPP 2023 - 18th International Conference on Computer Graphics Theory and Applications

154

dripping part of the algorithm that reduces the ﬂuid

content in the object. Once collision occurs between

the wetting tool in our framework and the boundary of

the tetrahedral mesh, the ﬂuid content in the tetrahe-

drons in contact with the tool increases in incremental

steps till the saturation value becomes one. The sat-

uration of a tetrahedron is deﬁned as S

= m

Here m

is the mass of water absorbed and V

is the

volume of the tetrahedron. After the absorption of

ﬂuid, diffusion happens between any two neighbour-

ing tetrahedra depending on the saturation gradient

between them (Patkar and Chaudhuri, 2013). A dry-

ing tool is provided whose action is complementary

to the wetting one.

3.1 Variable Elasticity

In order to formulate a relationship between ﬂuid con-

tent and elasticity of the material we followed the line

of thought presented in (Yoon and Cowin, 2009). To

the best of our knowledge, our framework is the ﬁrst

use of this method in an interactive, real-time setting.

The Voigt upper bound on the elasticity tensor of a

solid-ﬂuid mixture is given by

= (1 −φ)C

+ φC

(3)

where C

, C

and C

are the elasticity tensors of

the mixture, solid and ﬂuid respectively. The quan-

tity φ denotes the ﬂuid volume fraction in the solid.

The Reuss lower bound on the compliance tensor of a

solid-ﬂuid mixture with any kind of solid and ﬂuid is

given by

= (1 −φ)S

+ φS

(4)

where S

, S

and S

are the compliance tensors of the

mixture, solid and ﬂuid respectively. The Voigt and

Reuss bounds together give the bound on the effective

elasticity tensor as





−1

≤ C

e f f

≤





(5)

Putting everything together, the effective compliance

tensor for the solid-ﬂuid mixture with any kind of

solid and ﬂuid is given by

e f f

1 + φ



− P



−1

(6)

where Q

≡



− C



−1

, P

is the Eshelby ten-

sor (Eshelby, 1957), S

is the matrix compliance ten-

sor, C

is the matrix elasticity tensor and C

is the

inclusion elasticity tensor. For our framework, we as-

sume C

= C

, S

= S

and C

= C

. Using this

formulation we determine the effective elastic tensor

of a solid-ﬂuid mixture system. We always use water

for ﬂuid in our framework. As the change of elastic-

ity is dependent on the fraction of water content in

the tetrahedral element and water content is depen-

dent on the gradient of saturation, there is never any

abrupt change of elasticity in the model, thus main-

taining the stability of the system.

4 HAPTIC RENDERING

The model developed in the previous section for the

deformation of an object made of a porous material is

used in conjunction with haptic rendering. The hap-

tic interaction process with an object consists of the

following components:

• Continuous Collision Detection (CCD) between

the haptic proxy and tetrahedral simulation mesh.

• Continuous penalty-based haptic rendering while

deforming the mesh.

Haptic force feedback is performed on a volumetric

simulation mesh with tetrahedral elements. But to im-

prove the quality of visual rendering, we transfer the

deformation to a higher-resolution visualization sur-

face mesh. This is explained in Section 5.

t=1

t=0

t=t''

t=t'''

t=t'

p=q

Figure 2: Three contact times between colliding vertex ‘p’

and face ‘q’ are t

′

′′

′′′

. Penetration intervals are [t

′

′′

] and

′′

,1].

4.1 Continuous Collision Detection

We calculate continuous collision between the outer

boundary of the tetrahedral mesh and haptic proxy,

consisting of triangular face elements. We resolve

vertex-face and edge-edge collisions that arise when

two triangular face elements collide. In order to detect

a continuous collision, we begin by interpolating the

positions of each primitive i.e., vertex, edge and face

in the simulation time step, ∆t, normalized to [0,1].

Then a 3

order equation in t is solved to ﬁnd out

the number of collisions that occur between vertex-

face or edge-edge interactions during that simulation

time step. To detect the collisions fast, we used an

Axis Aligned Bounding Box for each of these primi-

tives and also followed a non-penetration ﬁlter-based

technique (Tang et al., 2010).

Real-Time Physics-Based Mesh Deformation with Haptic Feedback and Material Anisotropy

155

4.2 Haptic Rendering of Deformation

We classify the deformation of a mesh in two cate-

gories: (1) push deformation of mesh and (2) pull

deformation of mesh, both of which follow the same

principle except the applied force direction which is

inward for push and outward for pull. We detect col-

lision using CCD in each time step, ∆t, and then in-

tegrate over those particular time intervals when pen-

etration depth between collided primitives is positive,

implying that they are in a colliding state (Figure 2).

4.2.1 Vertex-Face Penalty Force

If a collision occurs between any vertex of the haptic

proxy and the triangular mesh boundary of the object

or vice-versa, then we calculate a penalty force (Tang

et al., 2012) as

V F

= k

v f

i<N

∑

i=0

− q

dt (7)

Here k

v f

is a scalar stiffness constant. Time inter-

vals [t

] ∈ [0,1] are called penetration time inter-

vals. These are deﬁned as time while the vertex is in-

side the object mesh and i is the number of penetration

time intervals (See Figure 2). Moreover n

, p

and q

denote contact normal, the position of vertex and con-

tact point on boundary mesh respectively during ∆t.

The collision point on the triangular mesh can be ex-

pressed using barycentric coordinates of three vertices

of the mesh as q

= w

+ w

. Once we get

the penalty force I

V F

we apply it to the object mesh



−I

V F

for pulling



and apply a reaction force of the

same magnitude but opposite direction to the proxy.

4.2.2 Edge-Edge Penalty Force

Similar to vertex-face penalty force, we also calcu-

late penalty force I

due to edge-edge collision. If a

collision occurs between the edge of the haptic proxy

mesh and the edge of the simulation mesh boundary

of the object, then the penalty force is calculated as

= k

i<N

∑

i=0

− q

dt (8)

Here k

is a scalar stiffness constant.

4.2.3 Clay-like Behaviour

While deforming the object we want to replicate a

clay-like behaviour in our model, i.e., the object

should be malleable near the point where an exter-

nal force is applied but the movement of the whole

structure of the object should be negligible due to this

Volumetric Simulation

Mesh (low res.)

Surface Visualization

Mesh (high res.)

1. Deform the Simulation Mesh

2. Update the

Visualization Mesh

Figure 3: Multi-resolution, multi-timescale deformation.

external force. To this end, we deﬁne a kernel func-

tion G

in Equation 9 around the position of the hap-

tic proxy. The velocities of the object mesh are scaled

with the weights of the kernel. If r = ||x − x

, then,

(x) =











1 + k

if r < R

1 + k

r + exp(k

if r ≥ R

(9)

where k

, k

are stiffness constants, x

denotes the po-

sition of haptic proxy and || · ||

denotes the l

norm.

is the inﬂuence radius of the damping kernel. As a

result, velocities of points further away from the hap-

tic proxy are more damped than closer points.

5 MULTI-RESOLUTION

FRAMEWORK

Physics-based simulation is computationally expen-

sive and cannot be performed on extremely high-

resolution meshes at interactive rates. However, vi-

sual ﬁdelity suffers a lot when low-resolution meshes

are used. On the other hand, haptic ﬁdelity requires

simulations to run at very high frame rates. Our

framework allows us to ﬁnd common ground between

all these disparate goals.

Our simulation runs on a coarse, low-resolution

volumetric mesh with tetrahedral elements that en-

closes a high-resolution surface mesh with triangle el-

ements like a cage. As shown in Figure 3, the vertices

of the surface mesh are expressed in the local space

of the simulation mesh using barycentric coordinates.

When the simulation mesh is deformed, the barycen-

tric coordinates of the surface mesh vertices in the

local space of the simulation mesh do not change.

This lets us calculate new coordinates of the surface

mesh vertices in a global coordinate system. Similar

ideas can be found in (Ju et al., 2005) (Chuhua Xian

et al., 2009). In Figure 4 two high-resolution sur-

face meshes, T-Rex (left) & Panther (right), and their

corresponding low-resolution volumetric simulation

meshes are depicted. Any manipulation performed on

the simulation mesh gets transferred to the visualiza-

tion surface mesh using a weight kernel. This sets up

the multi-resolution component of our framework.

GRAPP 2023 - 18th International Conference on Computer Graphics Theory and Applications

156

Figure 4: Surface visualization mesh embedded inside vol-

umetric simulation mesh: T-Rex (left) and Panther (right).

Figure 5: A deformation tool is colliding with the simu-

lation mesh (left). The circled colour gradient indicates the

region of deformation on the visualization mesh. The defor-

mation is projected onto the surface mesh for visualization

(right).

When a deformation (push/pull) tool collides with

the outer surface of the tetrahedral simulation mesh,

we visualize it by projecting a region with a colour

gradient on the surface mesh to denote the deforma-

tion region (Figure 5 left). The deformations are per-

formed on the simulation mesh. Using barycentric

coordinates this deformation is then transferred to the

surface mesh for visualization (Figure 5 right).

During wetting, any node of secondary mesh gets

the same saturation value as the tetrahedron which

contains it.

6 MULTI-TIMESCALE

FEEDBACK

For smooth haptic force feedback, a minimum refresh

rate of 1000 frames/sec is required. On the other

hand for smooth visual feedback, a refresh rate of

60 frames/sec is sufﬁcient. To achieve both these re-

quirements, the whole simulation is run in two distinct

threads. On one thread, physical simulations along

with graphics rendering are performed while another

thread is used for rendering haptic feedback. We have

kept the haptic thread running at 1000 frames/sec all

time using HAPI API which samples the haptic force

feedback at the required rate. The refresh rate of the

visual thread varies from 70 − 900 frames/sec, de-

pending on the underlying object mesh. So, the hap-

tic thread keeps rendering the same old force feed-

back at the higher frame rate, until it gets a force

update from the visual thread which runs at a much

slower rate. The synchronization between the two

threads is obtained implicitly due to the rapid update

rate of the haptic thread, instead of using a blocking,

explicit synchronization construct. Because of this

construct, our framework can work on high-resolution

mesh models with intricate details without degrading

the quality of haptic experience.

7 RESULTS

In this section, we present results that help evalu-

ate the performance of our mesh reshaping solution.

First, we show the results of deforming the object

mesh. We then demonstrate the effect of wetting the

material. Further, we present the results of a user

study, conducted to evaluate the qualitative perfor-

mance of our solution. A quantitative evaluation of

our framework is also conducted to afﬁrm that we sat-

isfy real-time interaction constraints.

All the results presented here are obtained on a

system with an Intel i7-4770K octa-core processor at

3.5GHz, 32GB RAM, a single Nvidia Geforce GTX

Titan GPU with 5860 MB of graphics memory and a

6-DOF haptic device from Geomagic Touch.

Figure 6: Illustration of the original model (left), the in-

teraction of the push tool with the model (middle) and the

interaction of the pull tool with the model (right).

As shown in Figures 6, whenever a haptic proxy

touches the simulation mesh, a colour gradient gets

projected on the surface of the mesh near proxy within

radius R

as mentioned in Equation 9. This helps the

user to get a better perception of the deformation re-

gion. In Figure 6, a zombie object mesh with a push

deformation (middle) and a pull deformation (right) is

shown. For push deformation, the mesh collapses and

bulging takes place for pull deformation.

Figure 7: Deforming a dry (left) and partially wet (right)

T-Rex model with a push haptic tool.

Real-Time Physics-Based Mesh Deformation with Haptic Feedback and Material Anisotropy

157

Figure 8: Illustration of haptic feedback force while inter-

acting with dry and wet object. As expected the force is

much less for the wet case.

Using a wetting tool we can wet material by trans-

fer of ﬂuid. Any vertex of the surface mesh gets the

same saturation value as the tetrahedron from the sim-

ulation mesh that contains the vertex. In Figure 7 a

user is shown interacting with a dry and a partially

wet T-Rex model. Except for the effect of wetting on

elasticity, the other material properties of the object

mesh and the area where the user interacts remain the

same in both cases. The haptic feedback force during

this interaction is shown in Figure 8. The perceptive

change of haptic feedback force after ﬂuid absorption

is evident from the plot which indicates that the wet

model offers less resistance compared to the dry one.

Moreover, as shown in Figure 7, the wet portion of the

mesh exhibits more deformation due to the change in

the material property after water absorption.

In Figure 9, we present a T-Rex model reshaped

using our framework.

Figure 9: Original (left) and deformed (right) T-Rex model.

7.1 User Study

Two different user studies were conducted to evaluate

the subjective quality of our virtual mesh reshaping

solution compared to real-world experience.

• Haptics-Visual Feedback Study to analyse the

effect of haptics and visual feedback in virtual

mesh reshaping. We perform an ANOVA analy-

sis for this study.

• Double Stimulus Comparison Study to evaluate

how close the virtual sculpting experience is com-

pared to real-world sculpting.

20 subjects in the age group 20 − 35 years partici-

pated in the user study. All the participants conﬁrmed

that they are not differently abled either physically or

Table 1: p-value for ANOVA study.

Compared strategies p-value

1 vs 2 0.00070

1 vs 3 0.00001

mentally. None of the participants had any prior ex-

perience with a haptic setup.

As the subjects participating in our experiment

were not familiar with any kind of haptic setup, we

ﬁrst trained them to use a haptic device. For that

purpose, we used a model scene provided with Ge-

omagic Touch haptic device. The scene contains two

boxes and using a haptic proxy, a user can move or

lift those boxes while getting appropriate force feed-

back. We ask each of the subjects to move and lift the

boxes with the haptic proxy repeatedly until he/she

feels comfortable handling a haptic device.

7.1.1 Haptics-Visual Feedback Study

The Analysis of Variance (ANOVA) (Fisher, 1954) is

a commonly used tool to evaluate whether the differ-

ences between groups of data are statistically signiﬁ-

cant. ANOVA is used in our work here to determine

the importance and effectiveness of both visual and

haptic feedback in a virtual mesh editing framework.

The participants are asked to perform virtual defor-

mation and wetting of a mesh in the following manner

and rate their experience on a scale of 1 (very poor) to

5 (very good) for each case.

• Strategy 1 - Visual On & Haptics On: Perform

both the virtual deformation and wetting opera-

tions with both visual and haptic feedback.

• Strategy 2 - Visual On & Haptics Off: Per-

form both the virtual deformation and wetting op-

erations with visual feedback but without haptic

feedback.

• Strategy 3 - Visual Off & Haptics On: Per-

form both the virtual deformation and wetting op-

erations without visual feedback but with haptic

feedback.

By ”without visual feedback” we do not denote a

complete absence of the virtual scene. It means that

the effect of the deformation and wetting operations

are not rendered on the high-resolution visualization

mesh. All the operations are performed only on the

outer cage mesh.

The null hypothesis in ANOVA suggests that all

groups are random samples from the same popula-

tion, which in our work means that all the there strate-

gies are equally effective. Thus, any observed dif-

ference between them is due to random noise. The

p-value deﬁnes the probability of obtaining results at

GRAPP 2023 - 18th International Conference on Computer Graphics Theory and Applications

158

Table 2: Mean and standard deviation of user feedback (1 -

very poor to 5 - very good).

Parameter Mean Median Std

Visual On, Haptics On 4.63 4.71 0.26

Visual On, Haptics Off 4.12 4.27 0.28

Visual Off, Haptics On 2.62 2.75 0.51

Realistic 4.61 4.65 0.35

Visual-haptic sync 4.72 4.90 0.25

Physical consistency 4.48 4.50 0.29

least as extreme as the observed results of a statisti-

cal hypothesis test, assuming that the null hypothesis

is correct. Thus, if the p-value falls below a certain

threshold, the null hypothesis is considered invalid. In

our study, we use p-value of 0.05, which is a widely

accepted choice. The p-value (see Table 1) for Strat-

egy 1 vs Strategy 2 is 0.00070 < 0.05, which rejects

the null hypothesis. From the top half of Table 2, the

higher mean and median score for Strategy 1 com-

pared to Strategy 2 denotes that user experience for

virtual sculpting improves when haptic feedback is

on. Similarly, the p-value for Strategy 1 vs Strategy

3 is 0.00001 < 0.05, which again rejects the null hy-

pothesis. The much higher mean and median score for

Strategy 1 compared to Strategy 3 proves that visual

feedback is very useful for a faithful user experience.

Finally, based on the previous results, we can

make the following observations.

Strategy 1 vs Strategy 2: According to the users’

rating, Strategy 1 is better than Strategy 2. It reveals

that if the visual feedback remains the same, turning

on the haptic feedback improves user experience.

Strategy 1 vs Strategy 3: We can observe from

the user ratings that Strategy 3 performs very poorly

compared to Strategy 1. This implies the importance

of appropriate visual feedback for our virtual mesh

editing framework. This closely resembles our real-

life experience too, where visual cues and motions are

most dominant among all the six sensory senses used

for perception.

7.1.2 Double Stimulus Comparison Study

We perform our subjective evaluation based on a dou-

ble stimulus comparison (Union, 2013) method. The

steps of the evaluation are as pointed out below.

• First Stimulus: We ask the subjects to mould a

ball of clay to any shape of their choice using their

hands and a pencil to get a feel of real-world shape

shaping.

• Second Stimulus: The subjects are then asked to

reshape object models virtually using our frame-

work with and without haptic feedback.

Our experimental setup is shown in Figure 10. A user

deforming a real clay sphere (left) and a virtual clay

sphere (right) is shown in the ﬁgure. After the ex-

periment is ﬁnished, the participants are asked to rate

their experience on a scale of 1 (very poor) to 5 (very

good) for the following parameters.

• Realistic: The users are asked to rate how close

is their experience compared to the real world.

• Visual-haptic Synchronization: The partici-

pants are asked if they experienced any delay be-

tween visual change and haptic force feedback.

• Physical Consistency: Consistency of the vi-

sual simulation of our framework with real-world

physical objects.

Figure 10: Comparing the experience of deforming a real

clay sphere on the left to the haptic feedback of deforming

a virtual clay sphere on the right.

The mean and standard deviation of the scores of

the user feedback opinions are listed in Table 2. The

ratings reﬂect highly realistic experiences with little

difference of opinion (low standard deviation).

7.2 Quantitative Evaluation

As mentioned earlier, in our multi-timescale frame-

work, the haptic thread updates at 1000 frames per

second for smooth interaction. Depending on the

model structure used, the frame rate of the visual ren-

dering thread varies between 70 and 900 frames per

second which is sufﬁcient for smooth visual feedback.

Further, to speed up the interaction frame rate in the

graphics thread we parallelized the computations on

the GPU wherever possible, using Nvidia CUDA. The

interactive frame rates of different tools of our frame-

work are presented in Table 3.

Table 3: Average frame-rate for different tools.

Model #tet Push/Pull Wet/Dry

T-Rex 2.5k 71.3 792.6

Zombie 1.2k 87.9 863.7

Sphere/Cylinder 1.5k 85.9 841.1

Real-Time Physics-Based Mesh Deformation with Haptic Feedback and Material Anisotropy

159

8 CONCLUSION AND FUTURE

WORK

We present a novel approach for a stable, real-time

simulation framework for mesh reshaping, enhanced

with haptic feedback and physically accurate material

simulation. We devise solutions to numerous chal-

lenges like the wetting of materials and the conse-

quent simulation of variable elasticity and deformable

porous solid simulation. Finally, we evaluate the ap-

peal and interactivity of our solution via a user study

and a variety of simulation results. One of the ma-

jor limitations of our work is that it works only with

one initial mesh. There is no provision for adding

more meshes on top of the initial mesh and cannot

thus model the functionality of material deposition.

In future, we want to work in this direction.

REFERENCES

Barbi

c, J. and James, D. L. (2009). Six-dof haptic ren-

dering of contact between geometrically complex re-

duced deformable models. In World Haptics 2009,

pages 393–394.

Bargteil, A. W., Wojtan, C., Hodgins, J. K., and Turk, G.

(2007). A ﬁnite element method for animating large

viscoplastic ﬂow. ACM Trans. Graph., 26(3):16:1–

16:10.

Bear, J. (1972). Dynamics of ﬂuids in porous media / Jacob

Bear. American Elsevier New York.

Blanch, R., Ferley, E., Cani, M.-P., and Gascuel, J.-D.

(2004). Non-realistic haptic feedback for virtual

sculpture. Technical Report RR-5090, INRIA.

Chen, H. and Sun, H. (2002). Real-time haptic sculpting in

virtual volume space. In Proceedings of VRST, pages

81–88. ACM.

Chuhua Xian, Hongwei Lin, and Shuming Gao (2009).

Automatic generation of coarse bounding cages from

dense meshes. In 2009 IEEE International Confer-

ence on Shape Modeling and Applications, pages 21–

27.

Dachille, IX, F., Qin, H., Kaufman, A., and El-Sana, J.

(1999). Haptic sculpting of dynamic surfaces. In Pro-

ceedings of I3D, pages 103–110.

De Goes, F. and James, D. L. (2017). Regularized kelvin-

lets: Sculpting brushes based on fundamental solu-

tions of elasticity. ACM Trans. Graph., 36(4):40:1–

40:11.

Erleben, K., Sporring, J., Henriksen, K., and Dohlman, K.

(2005). Physics-based Animation (Graphics Series).

Charles River Media, Inc.

Eshelby, J. D. (1957). The determination of the elastic ﬁeld

of an ellipsoidal inclusion, and related problems. Pro-

ceedings of the Royal Society of London. Series A,

Mathematical and Physical Sciences, 241(1226):376–

396.

Fei, Y. R., Batty, C., Grinspun, E., and Zheng, C. (2018). A

multi-scale model for simulating liquid-fabric interac-

tions. ACM Trans. Graph., 37(4):51:1–51:16.

Fisher, R. A. (1954). Statistical methods for research work-

ers; 20th ed. Oliver and Boyd, Edinburgh.

Gunn, C. (2006). Collaborative virtual sculpting with haptic

feedback. Virtual Reality, 10(2):73–83.

Irving, G., Teran, J., and Fedkiw, R. (2004). Invertible ﬁnite

elements for robust simulation of large deformation.

In Proceedings of SCA, pages 131–140.

Ju, T., Schaefer, S., and Warren, J. (2005). Mean value coor-

dinates for closed triangular meshes. In SIGGRAPH,

page 561–566.

Mandal, A., Chaudhuri, P., and Chaudhuri, S. (2022). In-

teractive physics-based virtual sculpting with haptic

feedback. Proc. ACM Comput. Graph. Interact. Tech.,

5(1).

McNeely, W. A., Puterbaugh, K. D., and Troy, J. J. (1999).

Six degree-of-freedom haptic rendering using voxel

sampling. In Proceedings of the 26th Annual Con-

ference on Computer Graphics and Interactive Tech-

niques, SIGGRAPH ’99, pages 401–408.

uller, M. and Gross, M. (2004). Interactive virtual mate-

rials. In Proceedings of Graphics Interface 2004, GI

’04, pages 239–246.

O’Brien, J. F., Bargteil, A. W., and Hodgins, J. K. (2002).

Graphical modeling and animation of ductile fracture.

ACM Trans. Graph., 21(3):291–294.

Ortega, M., Redon, S., and Coquillart, S. (2007). A six

degree-of-freedom god-object method for haptic dis-

play of rigid bodies with surface properties. IEEE

Transactions on Visualization and Computer Graph-

ics, 13(3):458–469.

Otaduy, M. A. and Lin, M. C. (2005). Stable and responsive

six-degree-of-freedom haptic manipulation using im-

plicit integration. In World Haptics Conference, pages

247–256.

Patkar, S. and Chaudhuri, P. (2013). Wetting of porous

solids. IEEE Transactions on Visualization and Com-

puter Graphics, 19(9):1592–1604.

Schraad, M. W. (2014). A Theoretical Approach to the Cou-

pled Fluid–Solid Physical Response of Porous and

Cellular Materials: Dynamics, chapter 6, pages 127–

152. John Wiley & Sons, Ltd.

Scriven, L. E. (1994). Porous media: Geometry and trans-

port by pierre m. adler, butterworth-heinemann, stone-

ham, ma, 1992, 544 pp. AIChE Journal, 40(2):380–

381.

Tang, M., Manocha, D., Otaduy, M. A., and Tong, R.

(2012). Continuous penalty forces. ACM Trans.

Graph., 31(4):107:1–107:9.

Tang, M., Manocha, D., and Tong, R. (2010). Fast

continuous collision detection using deforming non-

penetration ﬁlters. In Proceedings of I3D, pages 7–13.

ACM.

Union, I. T. (2013). ITU-R the double-stimulus continuous

quality-scale. https://www.itu.int/dms pubrec/itu-r/

rec/bt/R-REC-BT.500-13-201201-I!!PDF-E.pdf. Ac-

cessed: 2021-04-17.

GRAPP 2023 - 18th International Conference on Computer Graphics Theory and Applications

160

Xu, H. and Barbi

c, J. (2017). 6-dof haptic rendering us-

ing continuous collision detection between points and

signed distance ﬁelds. IEEE Transactions on Haptics,

10(2):151–161.

Yoon, Y. J. and Cowin, S. C. (2009). The elastic moduli

estimation of the solid-water mixture. International

Journal of Solids and Structures, 46(3):527 – 533.

Zilles, C. B. and Salisbury, J. K. (1995). A constraint-based

god-object method for haptic display. In Proceedings

1995 IEEE/RSJ International Conference on Intelli-

gent Robots and Systems, volume 3, pages 146–151.

Real-Time Physics-Based Mesh Deformation with Haptic Feedback and Material Anisotropy

161