-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathrelated.tex
77 lines (70 loc) · 7.42 KB
/
related.tex
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
\section{Related Work}
Close range interaction with surrounding objects or humans is an important research problem in character animation. This problem is challenging because it often involves potentially conflicting goals: maintaining intentional spatial constraints while avoiding unintentional contacts. Much research has been done on the challenge of handling contact and spatial constraints between body parts or objects \cite{Gleicher:1998:RMN,Liu:2006:CCO,Ho:2009:CMS,Kim:2009:SMM,Ho:2010:SRP}. Ho \etal \shortcite{Ho:2010:SRP} used an ``interaction mesh'' to encode the spatial relationship of interacting body parts. By minimizing the local deformation of the mesh, their method preserved the desired spatial constraints while reducing unintentional contacts or interpenetrations. In additional to the contact problem, close range interaction also demands sophisticated path planning. Previous work exploited inverse kinematics and motion planning techniques to generate motion that satisfies desired manipulation tasks in complex or cluttered environments \cite{Kallmann:2003:PCF,Yamane:2004:SAH}. A large body of robotics literature on the topic of motion planning for full-body manipulation is also highly relevant to the synthesis of close range interaction \cite{Harada:2003:PMH,Takubo:2005:PAO,Yoshida:2005:HMP,Nishiwaki:2006:MCS}. In this paper, dressing is also an example of close range interaction. Unlike most problems studied previously, dressing involves interacting with a unique object, cloth, which is highly deformable with frequent self-collisions.
% Depending on the applications, previous research developed techniques for different subproblems, such as collisions handling \cite{Ye:2012}, spatial constraints solving \cite{Ho:2010:SRP}, and path planning \cite{Kallman:2003,Yamane:2004:SAH,Bai:2012:SCO}. These existing techniques generated interesting animations beyond what simple forward simulation or keyframe interpolation could achieve. However, most methods assumed that the character interacts with humans or objects made of rigid bodies with a few exceptions. Ho and Kumura \shortcite{Ho:2009:CMS} introduced a technique to interact with deformable bodies using topology coordinates, in which the topological relationship of the character's body and the environment can be easily controlled. \shortcite{Jain:2010:SCT} proposed a simple
% - Spatial Relationship Preserving Character Motion Adaptation
% Path planning
% - Synthesis of Concurrent Object Manipulation Tasks
% - Planning collision-free reaching motions for interactive object manipulation and grasping
% - Synthesizing animations of human manipulation tasks
% Hand manipulation with soft objects
% - Coupling Cloth and Rigid Bodies for Dexterous Manipulation
% Robotic cloth manipulation
Researchers studying dexterous manipulation have developed control
algorithms to handle different types of manipulation, such as grasping
\cite{Pollard:2005:PBG,Kry:2006:ICS,Wang:2013:VHM,Zhao:2013:RRP}, finger
gaiting \cite{Ye:2012:SDH}, or rolling \cite{Bai:2014:DMU}. These methods
can successfully manipulate rigid bodies with various sizes and masses,
but it is not clear whether they can be extended to manipulating
deformable bodies, which typically have more degrees of freedom than rigid
bodies \cite{Wang:2012:manipulation}. In contrast to computer graphics, manipulating deformable bodies
has been addressed extensively in robotics. Researchers have demonstrated
robots manipulating cloth, ropes, cables, foam rubber, and sheet metal
\cite{Kosuge:1995:MFO,Wu:1995:AHC,Fahantidis:1997:RHF,Osawa:2007:UML,Cusumano:2011:BCD,Bersch:2011:BRC,Miller:2012:GAR}.
Our work is related to manipulation of cloth for folding laundry
\cite{Osawa:2007:UML,Cusumano:2011:BCD,Bersch:2011:BRC,Miller:2012:GAR} and assisted dressing of partially dressed, static mannequins \cite{Tamei:2011:reinforcement}.
However, due to the involvement of the human body, developing control
algorithms for dressing differs substantially from robotic manipulation of
cloth alone. In this paper, we do not address the problems related to
grasping and re-grasping, since this constitutes distinct challenges and
is actively being addressed by others in the robotics community.
% Papers that demonstrate dressing:
% - Character Motion Synthesis by Topology Coordinates: They generated keyframes in topology coordinates and interpolate motion. The character does not respond to the state of the clothes. No autonomous control. They demonstrated that a character stretching the arms out of a piece clothes wrapped around it using keyframes.
% - Harmonic Parameterization by Electrostatics
% - Clothing manipulation, Igarashi
A self-dressing virtual character has been previously demonstrated by a
few methods. Ho and Komura \shortcite{Ho:2009:CMS} introduced a technique
to interact with deformable bodies using topology coordinates, in which
the topological relationship of the character's body and the environment
can be easily controlled. They generated keyframe animation in topology
coordinates to demonstrate that a character is able to stretch her arms
out of a piece of clothing that is wrapped around her. To demonstrate the
effect of using electric flux for path planning, Wang \etal
\shortcite{Wang:2013:HPE} showed a virtual human putting on a sock and a
pair of shorts. In both cases, the clothes are already aligned with the
body parts and the character simply needs to pull them in the direction
indicated by the electric flux. While these methods hint at possible
solutions to the dressing problem based on movement of the character relative to the cloth, they have only been successfully applied to isolated situations and under many assumptions. In contrast, our
work designs a feedback controller such that the character can act
autonomously based on the state of the cloth and effectively achieve the goal in a diverse set of dressing situations with relatively few assumptions.
% Cloth sim interacting with rigids
% - Coupling Cloth and Rigid Bodies for Dexterous Manipulation
Although cloth simulation is a relatively mature research area, dynamic
coupling between cloth and rigid body systems still presents many
challenges. A variety of methods are proposed to handle two-way coupling
between deformable and rigid bodies
\cite{Jansson:2003:CDR,Sifakis:2007:HSD,Shinar:2008:TCR,Otaduy:2009:ICH,Miguel:2011:ESC,Macklin:2014:unified},
which can be potentially extended to rigid-cloth coupling. Otaduy et. al.
\cite{Otaduy:2009:ICH} solved contacts between cloth and rigid bodies by
implicitly solving a large mixed linear complementarity problem. Bai and
Liu \cite{Bai:2014:CCR} proposed a simpler coupling method which treats
existing cloth and rigid body simulators as black boxes without altering
the internal formulation of collision handling. Most recently, Macklin et. al. proposed a unified dynamic system in which all object interactions are solved as a collection of particle constraints. In our work, we directly
use the open source multibody simulator, DART \cite{Liu:2012:STM}, and the
cloth simulator, ARCSim~\cite{Narain:2012:AAR,Narain:2013:FCA}. ARCSim
treats the rigid bodies in the scene as objects with infinite mass and
only considers the contact forces from the rigid bodies to the cloth.
Since the type of clothes we consider in this paper are relatively
massless compared to the human character, ignoring the impact of cloth on
the character is a reasonable assumption. However, considering accurate
two-way coupling might be critical for dressing tighter clothing or
assistive dressing.