Modeling The Process of Animation Production

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/246546382

Modeling the Process of Animation Production

Article in International Journal of Automation Technology · April 2013

CITATIONS READS

0 7,331

6 authors, including:

Naohiro Shichijo Masao Sugi


Hitotsubashi University The University of Electro-Communications
28 PUBLICATIONS 331 CITATIONS 77 PUBLICATIONS 807 CITATIONS

SEE PROFILE SEE PROFILE

Tatsunori Hara Jun Ota


The University of Tokyo The University of Tokyo
129 PUBLICATIONS 908 CITATIONS 643 PUBLICATIONS 5,190 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Naohiro Shichijo on 29 May 2014.

The user has requested enhancement of the downloaded file.


Modeling the Process of Animation Production

Paper:

Modeling the Process of Animation Production


Hisato Nakanishi∗ , Naohiro Shichijo∗∗ , Masao Sugi∗∗∗ ,
Taiki Ogata∗ , Tatsunori Hara∗ , and Jun Ota∗

The University of Tokyo
5-1-5 Kashiwanoha, Kashiwa, Chiba 277-8568, Japan
E-mail: ota@race.u-tokyo.ac.jp
∗∗ National Institute for Science and Technology Policy

3-2-2 Kasumigaseki, Chiyoda-ku 100-0013, Japan


∗∗∗ The University of Electro-Communications

1-5-1 Chofugaoka, Chofu-shi, Tokyo 182-8585, Japan


[Received February 18, 2013; accepted April 8, 2013]

In this study, a work process model for animation pro- tions [3]. People in Japan’s domestic animation indus-
duction is developed. Interviews with animation stu- try have been trying to keep up the high quality of their
dio workers and surveys of their working diaries and works and have completed them on schedule. However,
shot progress charts are conducted to create a frame- in order to be more competitive international, further im-
work for the work processes. The work hours neces- provements in the quality management and productivity
sary for each operation of the work project are ex- will be necessary, and not only will the skills of each an-
pressed using a gamma function. Workers are clas- imator but also the productivity of the animation process
sified into two categories, managers and animators, will need to be improved. In order to improve produc-
and their behavioral rules are modeled. To establish tivity, the process used in executing their animation pro-
a rule for the determination of the operation sequence duction work needs to be clarified. This process involves
order for an animator to perform multiple work op- multiple workers and multiple work flows to follow, and
erations, four typical dispatching rules, often used in the relations between these workers and work flows are
a scheduling theory, are selected, and the one that is very complicated. Revealing these relations and making a
closest to actual data is regarded as the rule that the quantitative model of them will be of vital importance to
animator actually follows. As a result of the analysis, it the process.
is found that the Earliest Due Date (EDD) and SLACK
are closest to actual animators’ behavioral rules. Man-
agers are also found to move the sub-deadline (a dead- 1.2. Previous Studies
line of each operation) up about 30% earlier than the
Many studies have been done with a view to improv-
actual deadline. The knowledge we obtain is that, in
ing the productivity of animations and with an emphasis
order to realize better working conditions, animators
on supporting animators in order to reduce the labor of
should determine their operation sequence order by
drawing animations. DeCarlo et al. proposed a method
using SLACK, and managers should designate sub-
of line drawing 3D shapes [4]. Todo et al. proposed a
deadlines by keeping a balance between the preven-
method of modifying the shade and shadow of 3D CG
tion of deadline violation and the reduction of over-
images to animation-like drawings [5]. The technology
time work.
of deforming and combining two images and obtaining
an intermediate image between them is called morphing,
and software that performs 2D vector image morphing has
Keywords: animation production, scheduling, job analy-
already been developed [6]. The above studies were done
sis
to support the technical aspects of animation drawing but
not to examine how the work process of animation pro-
duction should be clarified, controlled, or improved.
1. Introduction More general studies to improve the behavior of work-
ers and to enhance their productivity, studies focusing
1.1. Background mostly on industrial production, have been made in the
Animations produced and broadcasted in Japan are field of industrial engineering. Barnes [7] proposed a
highly rated not only domestically but also by foreign method of making efficient work procedures for anima-
communities [1, 2]. The Ministry of Economy, Trade, tors based on work analyses presented in Refs. [8] and [9].
and Industry has proposed a strategy, “Cool Japan,” which In Toyota’s Just In Time production (JIT) strategy [10],
aims to promote the overseas expansion of Japan’s strate- production volume is finely tuned in each process to re-
gic industry, i.e., its cultural industry including anima- duce inventory. The work is standardized by modeling the

Int. J. of Automation Technology Vol.7 No.4, 2013 439


Nakanishi, H. et al.

waste in each process. Cheng et al. proposed a work per- 2. Work Performance Framework for
formance model for nurses [11, 12]. This study assumes Animation Production
that each nurse has his or her own dispatching rule for
the work sequence based on a given work sheet, and has In this section, the survey method is first introduced,
derived a rule that can explain the relation between the and then the framework for the work process is described.
work sheet and what the nurse actually did. They also For the description of the framework, from a viewpoint
proposed a new scheduling method of generating the work of a production scheduling model in manufacturing, work
sequence from work sheet on the basis of a simulated an- processes of the target work and machines (workers in
nealing technique. This method shares the same interest the present case) involved in the processes are explained.
with the present study with respect to the work process Next, the necessary working hours for each work process
analysis, but it does not directly focus on the efficiency of and the behavioral rules (the task-assignment rule of pro-
the work process of animation production. duction managers for assigning work processes to anima-
In short, no studies previously done have analyzed the tors, and the dispatching rule for animators to determine
work process of animation production to make a quantita- the processing order of assigned work processes) are ex-
tive model of it. plained.

1.3. Purpose of the Study 2.1. Survey Methodology


With this situation in mind, the present study aims to To make a descriptive framework of the work process,
develop a quantitative model of the work process of ani- we interviewed a total of 23 workers at an animation
mation production and to propose a new method of per- studio producing television programs and surveyed their
forming work based on the developed model. To these working diaries. Their shot progress chart was also sur-
ends, the following two challenges need to be overcome. veyed. The details of the survey are given below.
(1) It is difficult to examine a framework to be used as
• Interview: The workers were asked to describe, in
a base for modeling various aspects of the anima-
general terms, animation production, including the
tion production, such as what processes the anima-
type and frequency of troubles as well as the knowl-
tion production work consists of, what processes in-
edge that each member possessed.
dividual animators take charge of, and what behav-
ioral rules they follow in their work. • Working diary survey: The workers were asked to
(2) It is difficult to clarify the details of the above behav- record their daily work for two weeks.
ioral rules. Because the worker members follow the • Shot progress chart survey: A shot progress chart
rules unintentionally, expressing it explicitly is not is a table that describes the progress of the work; it
easy. will be explained in detail in section 3.5. The charts
To solve these problems, the present study takes the fol- of two programs were examined.
lowing steps. Basic actions taken to overcome prob-
lem (1) are interviews of animation studio workers, sur- 2.2. Framework Configuration
veys of their working diaries, and surveys of shot progress
charts. In addition, the framework of the work process 2.2.1. Process and Work Operation
is developed by taking account of the behavioral rule of The survey made it apparent that the work of animation
animation studio workers, as this behavioral rule is nec- production consists of the stages and processes shown in
essary for the successful performance of their duties. For Fig. 1. Each rectangular box in Fig. 1 represents a work
problem (2), if the above survey cannot clarify the detail process. Although some are the same processes as those
of the workers’ behavioral rule, possible rules will be ex- described in Ref. [13], the processes in this diagram are
tracted comprehensively based on the studies of produc- more detailed. A stage, indicated by surrounding dashes,
tion scheduling problem, and the most likely behavioral is a set of processes. The stages in Fig. 1 have the follow-
rule will be identified by comparing the work conditions ing meanings.
of each rule to actual work process data.
In section 2, a framework of the work process is shown. • Layout: Original background arts, original rough
In section 3, work performance rules of animation studio drawings, and an animation exposure sheet are cre-
workers are extracted for the modeling of the work pro- ated from a storyboard.
cess. An appropriate work performance method is pro-
• Key animation: Key animation drawings are cre-
posed in section 4, and conclusions and future prospects
are given in section 5. ated from the original background arts and the origi-
nal rough drawings.

• Inbetween: Inbetween drawings are created from


the key animation drawings and the animation ex-
posure sheet.

440 Int. J. of Automation Technology Vol.7 No.4, 2013


Modeling the Process of Animation Production

Fig. 1. Animation production process.

Table 1. Workers of each process.

Stage Layout Key animation Management


Process Making Check Revision Making Check Revision Management
Key Episode Animation Key Episode Animation Production
Worker
animator director director animator director director manager

• Ink-and-paint: Painted frames are created from the such as when a worker decided to redraw something. It is
inbetween drawings and color palettes. noted that “retake” means the work going back to an up-
stream stage in Fig. 1 (e.g. from “key animation” to “lay-
• Background: Background arts are created from the out”), and should not be confused with “revision” process
original background arts. in each stage. Retakes will not be considered in this study.
• Compositing (Photography): An animation is cre- It was also found that the animation studio contracted
out the stages “inbetween,” “ink-and-paint,” and “back-
ated from the storyboard, background arts, painted
frames, and the animation exposure sheet. ground” to subcontractor studios, so the studio checked
and controlled the progress of each stage, but did not over-
It was also found that there were processes additional to see the processes in these three stages in detail. Although
those in Fig. 1. Specifically, there is always a process be- the management of the subcontractor studios is an impor-
tween every two consecutive processes, a “management tant issue, the present study is not directed to this prob-
process,” in which drawing images are collected at the end lem; it focuses only on the stages of “layout” and “key
of the preceding process and redistributed to the work- animation,” which the animation studio subjectively work
ers for the next process. The animation production works on and which largely affect the efficiency of the entire
are carried out in the following way. If making an ani- project.
mated TV program is regarded as the whole work, each
episode corresponds to a single product. (For a TV pro-
gram of 12 episodes, the production work would consist 2.2.2. Workers Implementing the Work Process
of 12 products, for example.) Each product (episode) is The work categories of the workers who take care of
then composed of “shots,” a shot being a series of image each of the abovementioned processes are shown in Ta-
frames taken by means of continuous camera work or a ble 1. The processes of the same name in the stages “lay-
single camera placement. The number of shots in each out” and “key animation” are usually carried out by the
episode depends on various factors, such as the length of same workers.
the episode, the story of the episode, and the director’s In Table 1, there are four kinds of workers: Key anima-
preference for camera work. There are usually several tors, episode directors, animation directors, and produc-
hundred shots in each episode. tion managers. The former three have common character-
The abovementioned work processes are performed for istics that they are directly involved in making animation
each shot. In this paper, a process for a shot is referred to frames. Hereafter in this paper, key animators, episode di-
as an “operation.” rectors, and animation directors are comprehensively re-
According to the working diaries, retakes, mentioned ferred to as “animators.” In contrast to animators, pro-
in Ref. [13], were done only in a limited number of cases, duction managers take charge of maintaining the project

Int. J. of Automation Technology Vol.7 No.4, 2013 441


Nakanishi, H. et al.

vided into shots and work processes, and information


on overall deadline from a higher organization.

(b) The manager specifies the sub-deadline of each oper-


ation, i.e., the date each process in each shot should
be finished, and assigns each operation to animators
(key animators, episode directors, and animation di-
rectors) so that the workload of each staff is bal-
anced. The role of the animators in each process is
shown in Table 1.

(c) The animators conduct operations assigned. They re-


port the manager when an operation is complete. If
there are other operations to be processed, return to
(b). If no animators have unprocessed operations,
proceed to (d).
Fig. 2. Required time distribution and estimated cumulative
distribution function. (d) The production of the animated program is complete.
In the above (b), note that each sub-deadline corre-
progress and are not directly involved in drawing anima- sponds to each operation, and it should not be confused
tion frames. Production managers are called “managers” with the final deadline of an entire episode. It was found
hereafter. that the scheduling was not made in a centralized man-
ner as done in ordinary scheduling problem solutions
2.2.3. Time Required for Each Process but rather by autonomous decentralized decision-making
based on each worker’s behavioral rule. There were two
In this section, the time required for each work process major rules for managers as follows:
is estimated from the working diary data. Since the time
varies from worker to worker for the same process, it is • Operation assignment rule:
appropriate to assume that the time required for each pro- How to assign operations to animators?
cess is a random variable.
The time required for each work stage is estimated from • Sub-deadline determination rule:
the working diary and plotted in a cumulative probability By when should the operation be finished?
graph. The time necessary for a key animator to finish
The rule animators use to conduct their operations are the
“making” process of “layout” stage was calculated from
following two:
their working diaries and plotted as crosses in Fig. 2. By
assuming that the plotted data can be approximated with • Operation suspension decision rule:
a gamma distribution [14], a cumulative distribution func- How much operations does an animator do within
tion of required time is obtained and presented as a curve a day? Specifically, when does an animator stop
in Fig. 2. The probability density function of the gamma his/her work?
distribution, p(X = t), can be expressed with parameters
k(> 0) and θ (> 0) in the following form, • Operation sequence order determination rule:
In what order does an animator carry out multiple
t k−1 e−t/θ
p (X = t | k, θ ) = . . . . . . . (1) operations assigned?
Γ (k) θ k
The framework of the work process of animation produc-
Here, Γ (k) is a gamma function. k and θ , which max-
tion can thus be determined. It has the following charac-
imize the likelihood, are obtained by numerical calcula-
teristics if compared to general business or manufactur-
tion. For Fig. 2, we obtained k = 3.5 and θ = 5.2 × 10−1 .
ing.
These parameters were also calculated for the other pro-
cesses in the same manner. Since the time required for the • Animation production work is performed by many
“management” process is extremely short, it was set to 0 workers and less automated. It is strongly dependent
in this study. on the implicit knowledge of individual workers and,
in particular, on their scheduling ability for the as-
2.2.4. Behavioral Rule signment of operations and the determination of the
As a result of the survey, it was found that animation sequence of operations.
production work was performed by exchanging informa-
• Unskilled workers often work as managers and as-
tion in the manner shown in Fig. 3.
sign operations to animators. (This was found in the
(a) A manager (production manager) receives a group interviews.) Therefore, their ability greatly affects
of operations, in which an animation program is di- overall work efficiency.

442 Int. J. of Automation Technology Vol.7 No.4, 2013


Modeling the Process of Animation Production

Fig. 3. Production process with focus on workers.

• The working time of animators varies with the situ- 1, 2, . . ., Ns ) is the shot number, and p (p = 1 (lay-
ation. This is because working hours are extended out making), 2 (layout check), . . . , 6 (key animation
during busy times. This fact is not taken account of revision)) is the process number.
in ordinary scheduling problems. (Ordinary schedul-
• The time required for each operation t j is determined
ing problems assume constant available time for all
machines.) randomly according to the required time distribution
given in section 2.2.3.
• Although there is a kind of core working period like
• Since the animators and managers does not know the
usual flextime scheme, working hours policy in an-
imation production industry is relatively loose, and actual working hours, estimated value t˜j given ran-
the variation of working periods among animators domly within the range of ±20% around the above
is large. Therefore it is difficult to predict working true value, is used for various purposes.
hours of animators. • The animators are divided into Ng groups. This
means that each group takes care of a total of Ne /Ng
episodes. Each group has WKA key animators, WED
3. Estimation of Behavioral Rules episode directors, and WAD animation directors.

In this section, the behavioral rules discussed in sec-


3.2. Work Assignment Rule
tion 2.2.4 is again discussed but in greater detail. The rule
of operation assignment (section 3.2), rule of sub-deadline Taking the interviews into account, it was considered
determination (section 3.3), and rule of operation suspen- adequate to make the following rule: “Operations should
sion (section 3.4) are identified based on the interviews. be assigned to animators proportional to the reserve mar-
The rule of operation sequence order determination (sec- gin of their working abilities.”
tion 3.5) is obtained by comparing the results of simula- Algorithm 1 shows a manager’s algorithm that assigns
tions performed under rule candidates and the actual work operations that are the work processes ρ and ρ + 3 (ρ =
process (shot progress chart in section 2.1) and by finding 1, 2, 3) of shots 1, 2, . . ., Ns in episode ε to animators.
the most suitable rule. Here, the order date of ε is expressed as Eε . Oε denotes
a set of operations in episode ε that is not assigned to
any animator. It is noted that |Oε | = Ns × N p = 6Ns holds
3.1. Formulation
at the initial condition. OKA ED AD
ε , Oε , and Oε are the sets
The work process of animation production is formu- of operations that should be conducted by key animators,
lated in the following manner. The work process of ani- episode directors, and animation directors, respectively.
mation production is formulated in the following manner. Therefore OKA ε + Oε + Oε ≡ Oε holds. Li means a set
ED AD

• The entire work consists of Ne episodes. An order of operations assigned to animator i but not finished yet.
is placed for one of the episodes every Nw weeks. j.e, j.s, and j.p in algorithm 1 present the three values
The work for each episode has to be finished within of an operation j = (e, s, p). j(p ← n) denotes an opera-
D days after the order date. The number of shots in tion in the same shot of the same episode with operation
each episode is fixed at Ns . There are 6(= N p ) work j, but the process number n is different from that of j. For
processes in the work, as shown in Table 1: Layout example, for two operations j1 = (e, s, p) = (ε , σ , ρ1 ) and
making, layout check, layout revision, key animation j2 = (e, s, p) = (ε , σ , ρ2 ), the relationship of these two can
making, key animation check, and key animation re- be written as j2 = j1 (p ← ρ2 ) and j1 = j2 (p ← ρ1 ). It is
vision. noted that j1 (p ← ρ1 ) ≡ j1 and j2 (p ← ρ2 ) ≡ j2 .
In algorithm 1, for convenience, we denote W1 = WKA,
• Each work operation is labeled by j = (e, s, p) where W2 = WED , W3 = WAD and O1 = OKA ε , O2 = Oε , O3 =
ED

e (e = 1, 2, . . ., Ne ) is the episode number, s (s = AD


Oε .

Int. J. of Automation Technology Vol.7 No.4, 2013 443


Nakanishi, H. et al.

Fig. 4. Sub-deadline determination.

When j . p = 2 (Layout check)


D j shall be set one day earlier than the case with j.p = 3
(see below and Eq. (4)). The one-day difference between
j.p = 2 and j.p = 3 is due to necessary hours for the lay-
out check process.
Dj = ∑ t˜j − 1 . . . . . . . . . . . (3)
j ∈Li( j)
j  .p=1

When j . p = 3 (Layout revision)


D j shall be determined as the expected date for a key ani-
mator who takes charge of the corresponding shot to finish
all of the assigned operations with j.p = 1 (i.e. expected
date for him to start operations with j.p = 4).
Dj = ∑ t˜j . . . . . . . . . . . . (4)
j ∈Li( j)
“SetSubDeadline( j)” in algorithm 1 means that the j  .p=1
manager determines the sub-deadline of operation j based When j . p = 4 , 5 (Key animation making/check)
on the sub-deadline determination rule, which is ex- D j shall be determined by bringing forward the deadline
plained in the next section. by as many days as is necessary for the subsequent pro-
cesses.
3.3. Sub-Deadline Determination Rule
D j = Eε + D − (6 − j.p) . . . . . . . . (5)
This rule was also identified based on the interviews.
At first, D j , the absolute sub-deadline (in words, unpost- When j . p = 6 (Key animation revision)
ponable sub-deadline) for the operation j is determined The absolute deadline of episode ε , i.e., final process date,
according to the current work volume of the animators has already been given.
who are involved in this operation and to the final deadline D j = Eε + D . . . . . . . . . . . . (6)
of an entire episode ε (i.e. (Eε + D)). How to determine
D j is detailed below. The actual sub-deadline d j that a manager requests of
an animator is determined based on the abovementioned
When j . p = 1 (Layout making) absolute (unpostponable) deadline D j using a parame-
D j shall be determined as a middle day between the abso- ter α which represents the degree of advancement of the
lute sub-deadline for the fourth process (i.e. key anima- deadline schedule.
⎛ ⎞
tion making process) of the same shot and the work hours
necessary for the operation j’s corresponding animator to dj = ∑ t˜j + α ⎝D j − ∑ t˜j ⎠ . . . (7)
finish all the operations already assigned to him. j  ∈Li( j) j  ∈Li( j)
⎛ ⎞
1 Figure 4 explains what this equation expresses. α = 1
D j = ⎝Eε + D − 2 + ∑ t˜j ⎠ . . . . (2) indicates no advancement of the deadline schedule (i.e.
2 j  ∈L
i( j) the requested sub-deadline is identical to the absolute sub-
deadline), and α = 0 indicates full advancement. For ac-
Here, i( j) means the animator i who is assigned to the
tual use, α is determined by the method explained here-
operation j.
inafter.

444 Int. J. of Automation Technology Vol.7 No.4, 2013


Modeling the Process of Animation Production

3.4. Work Suspension Rule


Table 2. Dispatching rules as candidates for operation se-
Using the information obtained in the interviews, we quence order determination rule.
created the following model of work suspension.
Necessary time taken account of
• Daily work usually finishes within the animators’ No Yes
standard daily work hours.

taken account of
Sub-deadline
• If an animator’s to-do list contains an operation No FCFS SPT
which might not be finished by the specified sub-
deadline, i.e., if the sub-deadline cannot be met, the
animator shall work overtime, although the working Yes EDD SLACK
hour limit should not be exceeded.

3.5. Operation Sequence Order Determination


• EDD: Earliest Due Date. Operations in Li are con-
Rule
ducted in the order of sub-deadline for each opera-
This rule is, in a sense, implicit, and it is therefore dif- tion.
ficult to extract it from the interviews or working diaries.
As was done in the studies by Cheng et al. [11, 12], data • SLACK: Operations in Li are conducted in the as-
on actual work progress (i.e., shot progress chart, from cending order of “slack time” S j , expressed in the
which one can find what shot was finished on what day) following formula.
and simulation results of comprehensively-selected oper-  
ation sequence order determination rule candidates (sec- S j = d j − t˜j + (current time) . . . (8)
tion 3.5.1) are compared, and the one that gives the small-
est difference between the actual data and the simulation
results is employed as the actual rule (section 3.5.2). 3.5.2. Identification of Operation Sequence Order
Determination Rule Based on Simulations
3.5.1. Listing of Possible Operation Sequence Order Which rule candidate would be closest to actual data is
Determination Rule determined by simulations. The following are assumed to
The problem of determining the sequence order of mul- reproduce the actual situation.
tiple operations is called the scheduling problem [15, 16]. • The total number of episodes Ne = 12, order place-
One of the common methods for solving such problem ment time interval Nw = 4 (weeks), and the number
is to apply a dispatching rule that uses current time and of shots in each episode Ns = 250.
work information to determine the next operation. Since
dispatching rules are relatively simple and intuitive, this • The number of animator groups Ng = 4. As a result,
study assumes that animators determine the order of op- the number of episodes that each group takes care of
erations based on a certain dispatching rule. Under the is 12/4 = 3 (episodes). For each group, the num-
dispatching rule, the score or priority of each operation ber of episode directors WED = 1, and the number of
is calculated, and the operation list is sorted with respect animation directors WAD = 2.
to the score or priority. The first operation ranked in the
sorted list is the one that animators should perform next. • Standard working hours per day hstd = 8 (hours),
Several methods of calculating the score or priority of and upper limit of working hours allowed hlim = 16
each operation have been proposed. The methods em- (hours).
ployed in this study are the following. The methods • The shot progress chart is used as actual data. As
for calculating scores for dispatching rules are classi- mentioned in section 2.1, the shot progress chart is
fied into 2 × 2 = 4 categories according to the conditions a table that records the progress of the work a di-
(a) whether the working hours necessary for an operation ary. It gives the date and the number of shots finished
is taken account of and (b) whether the sub-deadline is on that day. The ratio (progress rate) of the finished
taken account of, and a typical rule in each category is shots in each process to the total number of shots is
selected as a rule candidate (Table 2). The details of se- calculated from the shot progress chart and used for
lected rule candidates are described below. experiments.
• FCFS: First Come, First Served. Under this rule, The following values could not be obtained from the in-
operations in Li (the to-do list for the animator i) are terviews or working diaries and are therefore treated as
conducted in the order of them being added to Li . variables to be derived.
• SPT: Shortest Processing Time. Under this rule, op- • Final deadline is D days after the date
erations in Li are conducted in the ascending order of when work on each episode is started
estimated time necessary for each operation. (D = {20, 25, 30, . . ., 95, 100}).

Int. J. of Automation Technology Vol.7 No.4, 2013 445


Nakanishi, H. et al.

Table 3. Combination of variables that best matched actual data.

Case 1 Case 2
WKA 5 5
R EDD/SLACK SLACK/EDD
D 85 60
α 0.3 0.3

• Each group has WKA (WKA = {5, 6, 7, . . . , 20}) key


animators.

• The parameter α that represents advancement of the (a) Case 1 (Solid lines: Actual data for Layout, Dashed lines:
deadline schedule for managers (see section 3.3) is Simulation results)
given by α = {0.1, 0.2, . . . , 0.9, 1.0}.

• Four candidates for the operation sequence order de-


termination rule for animators are shown in Table 2
(R = {FCFS, SPT, EDD, SLACK}).
In order to find the best combination of the variables
D, WKA, α , and R, the difference in the progress rate on
a given day between the simulation and the actual data
for a selected combination of the variables is calculated.
(The progress rate is set to 0 on the day the work begins.)
The sum F of the calculated differences (i.e., sum of the
difference in the progress rate over days and processes) is
derived from the following formula.
Np Nd  
1  d d
F=
Np ∑∑ Π
 p − Π̃ p . . . . . . . (9)
p=1 d=0

Here, (b) Case 2 (Solidlines: Actual data, Dashed lines: Simulation


results)
N p : Number of processes to be analyzed (= 6)
Fig. 5. Simulation results compared with actual data.
Nd : Maximum number of days to finish designated work
(should be set to a sufficiently large number)

Π pd : Progress rate of the process p on d-th day (actual • EDD and SLACK were found to be appropriate
data) as rules for the animators to determine the order
of assigned operations. In each column of Ta-
Π̃ pd : Progress rate of the process p on d-th day (value ob- ble 3, the combination of the variables that gives a
tained by simulations) smaller F appears on the left. Therefore, in case of
F is calculated for all combinations of parameters D, “EDD/SLACK,” EDD comes first and SLACK next.
WKA, α , and R to find the combination with the smallest However, there is not much difference between the
F. value F in EDD and in SLACK. This indicates that
animators determine the operation sequence order
with emphasis on the sub-deadline. In other words,
3.5.3. Simulation Results the working conditions of animators can be con-
We performed simulations for two animation programs trolled considerably by adjusting the sub-deadline.
with shot progress charts available. The combination of • α is set to a relatively small value, 0.3. This means
the variables that best matched the actual data is shown in that the sub-deadlines managers tell to animators are
Table 3, and the progress rates of the two cases are shown ahead of the “actual” limit dates to a significant de-
in Fig. 5. In one of the two cases (called case 1), actual gree. As a result, animators have to work overtime
data were available only for three processes of the layout from the beginning.
stage. We therefore set N p = 3 in Eq. (9) for this case.
The results are discussed below. • It was found from the subjective impressions of re-

446 Int. J. of Automation Technology Vol.7 No.4, 2013


Modeling the Process of Animation Production

lated animators that the values of WKA and D ob-


tained in Table 3 were quite realistic.

Because of the above facts and because the actual data


and the simulation results in Fig. 5 are mostly in line, the
results shown in Table 3 should be highly valid. It can
therefore be considered that the validity of the model pro-
posed in sections 3.2 to 3.5 has been confirmed.
The premise of the knowledge obtained in this study is
that the work hours for the target work vary considerably,
although the work hours could be a somewhat rough esti-
mate. If the working hours could be accurately estimated,
a centralized scheduling method widely used at factories
would be more appropriate. On the contrary, if no in-
formation about working hours were available, a simpler
method such as FCFS, with decentralized manner would Fig. 6. Cumulative probability density distribution of work
be adequate. It was also shown that even when the in- hours per day of a worker responsible for layout and key
terviews or other surveys could not collect sufficient data, animation.
the data could be supplemented to allow for efficient anal-
ysis by integrating the abovementioned simulation analy-
sis and the results of the interviews, working diary survey,
and shot progress chart survey. to about 0.3 as in the previous section.
α = 1 is preferable for shortening overtime work hours
and α = 0.3 for minimizing deadline violation. In other
4. Proposal of More Appropriate Style of Work words, these two values are in a trade-off relationship, and
Execution it is difficult to set α to meet the two conditions. Both
conditions can be improved if the value WKA (number of
The current situation was analyzed in the discussions key animators) is increased. In an actual situation, WKA
in the preceding sections. In this section, to deepen the is first set to the minimum number necessary, and then
discussion, we propose a method of specifying the sub- the two evaluation functions (overtime work and deadline
deadline (i.e., specifying α ) and an appropriate rule for violation) are compared to set an appropriate value of α .
determining the operation sequence order for animators. To close this section, we discuss the relation of the
Overtime work from an animator’s viewpoint and produc- present study to automation technology. An important
tivity and deadline violation from a manager’s viewpoint point of the present study is that, even for jobs with many
will be discussed below. manual operations, such as service-sector jobs, a quanti-
Since the daily working hours of an animator can be tative model could be developed by using a scheduling
calculated by using the model proposed in section 3, over- model, one of the core knowledge of automation tech-
time work and slack time during the animation produc- nology. Another important point is that, with this model,
tion period can be estimated. A simulation was per- one can find which part in the work process could be im-
formed by setting (WKA, R) = (5, SLACK) and (α , D) = proved, as shown in the present study, and which manual
{(1, 30), (1, 90), (1/3, 30), (1/3, 90)} to calculate work operation should be automated in the future.
hours per day of key animators. The work hours were av-
eraged for each group and plotted in Fig. 6. It was found
that workload varied more from day to day in the schedule 5. Conclusions and Future Prospects
with α = 1/3 than in the schedule with α = 1. This dif-
ference in variation is more significant for D = 30 than for In this study, we have built a work process model for
D = 90. These tendencies were also observed for anima- animation production. It has been found through a com-
tors involved in other operations. It was therefore found prehensive analysis that EDD and SLACK are appropriate
that the sub-deadlines should be set to the actual “limit” as the rules for determining the order of multiple opera-
dates without moving forward in order to prevent varia- tions for animators. It has also been found that the sub-
tions in the workloads of animators. deadlines that managers tell to animators are set to dates
Next, productivity and deadline violation are discussed 30% earlier than the actual final deadlines. A method of
from the viewpoint of management. Simulations with var- reducing workloads and increasing on-time work comple-
ious sets of D, WKA, α , and R showed that, among the op- tion has been proposed based on the model developed. It
eration sequence order determination rules in R, SLACK has therefore been shown that, with this work model, var-
was the one that minimized the number of shots com- ious proposals could be made using quantitative data.
pleted after the deadline. As a result, SLACK was found The knowledge obtained in the present study could be
to be preferable as an operation sequence determination applied to work processes in which variation in work-
rule for animators. It was also found appropriate to set α ing hours is relatively large, even though work hours can

Int. J. of Automation Technology Vol.7 No.4, 2013 447


Nakanishi, H. et al.

be roughly estimated. From the standpoint of method-


ology, the proposed modeling method, which utilizes in-
terviews, working diary surveys, and shot progress chart Name:
surveys and which is based on a scheduling model, is Hisato Nakanishi
a highly general method. This is because interviews or
other surveys can supply necessary information appropri- Affiliation:
The University of Tokyo
ately, even when not all information is available. Brief Career:
Possibilities for future study include improvements in 2012 Graduate School of Engineering, The University of Tokyo
the precision of the work process model and widening of
the range of the processes incorporated in the model.

Acknowledgements Name:
Naohiro Shichijo
This study was partially supported by the “Project of personnel re-
cruiting and training in the content industry (Project of basic skill
Affiliation:
improvement of animators)” (Ministry of Economy, Trade, and Foresight Center, National Institute for Science
Industry, 2010) and by the Waseda Institute for Advanced Stud- and Technology Policy (NISTEP)
ies.

References:
[1] J. Krikke, “Computer graphics advances the art of anime,” IEEE Address:
Computer Graphics and Applications, Vol.26, Issue 3, pp. 14-19, 3-2-2 Kasumigaseki, Chiyoda-ku 100-0013, Japan
2006. Brief Biographical History:
[2] T. Ueno, “Japanimation and Techno-Orientalism: Japan as the Sub- 1999-2010 Interfaculty Initiative in Information Studies, The University of
Empire of Signs,” Documentary Box, Vol.9, 2003. Tokyo
[3] Promotion of the Cool Japan Strategy and the Creative Tokyo 2010-2012 Waseda Institute for Advanced Study, Waseda University
Project, Available: http://www.meti.go.jp/english/policy/mono 2013- National Institute for Science and Technology Policy (NISTEP)
info service/creative industries/pdf/121016 01b.pdf., [accessed Main Works:
Feb. 7, 2013]
• S. Liu and N. Shichijo, “Japanese corporate strategies on location choices
[4] D. DeCarlo, A. Finkelstein, S. Rusinkiewicz, and A. Santella, “Sug- of R&D activities in china: identifying the role of knowledge spillovers,” J.
gestive Contours for Conveying Shape,” ACM Trans. on Graphics,
Vol.22, Issue 3, pp. 848-855, 2003. of Enterprising Culture, Vol.18, No.2, pp. 167-191, Jun. 2010.
[5] H. Todo, K. Anjyo, and T. Igarashi, “Stylized Lighting for Cartoon • Y. Baba, M. Yarime, and N. Shichijo, “Sources of Success in Advanced
Shader,” The J. of Computer Animation and Virtual World, Vol.20, Materials Innovation: The Role of “Core Researchers” in
Issues 2-3, pp. 143-152, 2009. University-Industry Collaboration in Japan,” Int. J. of Innovation
[6] Live2D, http://www.live2d.jp/, [accessed Aug. 2, 2012] Management, Vol.14, Issue 2, pp. 201-219.
[7] R. M. Barnes, “Motion and Time Study: Design and Measurement • Y. Baba, N. Shichijo, and S. Sedita, “How do collaborations with
of Work,” Wiley, 1980. universities affect firms’ innovative performance? The role of ‘Pasteur
[8] F. B. Gilbreth, “Motion Study,” D. Van Nostrand, 1911. scientists’ in the advanced materials field,” Research Policy, Vol.38,
[9] F. W. Taylor, “The Principles of Scientific Management,” Harper pp. 756-764, 2009.
and Bors., 1929. Membership in Academic Societies:
[10] Y. Sugimori, K. Kusunoki, F. Cho, and S. Uchikawa, “Toyota Pro- • The Japan Society for Science Policy and Research Management
duction System and Kanban System Materialization of Just-in-time (JSSPRM)
and Respect-for-human System,” Int. J. of Production Research, • The Academic Association for Organizational Science (AAOS)
Vol.15, No.6, pp. 553-564 1977. • Academy of Management (AoM)
[11] M. Cheng, H. Itoh Ozaku, N. Kuwahara, K. Kogure, and J. Ota, • Association of Computational Machinery (ACM)
“Nursing Care Scheduling Problem: Analysis of Inpatient Nurs-
ing,” The Society of Instrument and Control Engineers Trans. on
Industrial Application, Vol.8, No.7, pp. 54-67, 2009.
[12] M. Cheng, H. Itoh Ozaku, N. Kuwahara, K. Kogure, and J. Ota,
“Dynamic Scheduling in Inpatient Nursing,” Int. J. of Automation
Technology, Vol.3, No.2, pp. 174-184, 2009.
[13] S. Kamimura, “Encyclopedia of Animation Basics,” Graphic-sha,
2009. (in Japanese)
[14] H. Hironaka, “Encyclopedia of Mathematical Sciences,” 2nd (Ed.),
Maruzen, 2009. (in Japanese)
[15] J. Weglarz, “Project Scheduling: Recent Models,” Kluwer, Algo-
rithms, and Applications, 1999.
[16] M. Pinedo, “Scheduling: Theory, Algorithms, and Systems,” Pren-
tice Hall, 2001.

448 Int. J. of Automation Technology Vol.7 No.4, 2013


Modeling the Process of Animation Production

Name: Name:
Masao Sugi Taiki Ogata

Affiliation: Affiliation:
Associate Professor, Department of Mechanical Research Associate, Research into Artifacts,
Engineering and Intelligent Systems, Graduate Center for Engineering (RACE), the University
School of Informatics and Engineering, The Uni- of Tokyo
versity of Electro-Communications

Address: Address:
1-5-1 Chofugaoka, Chofu-shi, Tokyo 182-8585, Japan 5-1-5 Kashiwanoha, Kashiwa, Chiba 277-8568, Japan
Brief Biographical History: Brief Biographical History:
2003-2007 Project Research Associate, Graduate School of Information 2009- The University of Tokyo
Science and Technology, The University of Tokyo 2009- Project Researcher, Tokyo Institute of Technology
2007-2009 Project Assistant Professor, Information and Robot Technology Main Works:
Research Initiative, The University of Tokyo • “Temporal Co-creation between Self and Others with Multi-sensory
2009-2010 Project Researcher, RACE (Research into Artifacts, Center for Inputs,” The Advanced Engineering Informatics, Vol.20, pp. 321-333,
Engineering), The University of Tokyo 2006.
2010-2011 Assistant Professor, Department of Mechanical Systems Membership in Academic Societies:
Engineering, Graduate School of Engineering, Tokyo University of • The Japan Society for Precision Engineers (JSPE)
Agriculture and Technology • The Society of Instrument and Control Engineers (SICE)
2011- Associate Professor, Department of Mechanical Engineering and • Society for Serviceology (SfS)
Intelligent Systems, Graduate School of Informatics and Engineering, The
University of Electro-Communications
Main Works:
• Y. Tamura, M. Sugi, T. Arai, and J. Ota, “Attentive Deskwork Support
System,” J. of Advanced Computational Intelligence and Intelligent
Informatics, Vol.14 No.7, pp. 758-769, 2010.
• M. Sugi, Y. Shiomi, T. Okubo, K. Inoue, and J. Ota, “A Solution for 2D Name:
Rectangular Cutting Stock Problems with 3-Stage Guillotine-Cutting Tatsunori Hara
Constraint,” Int. J. of Automation Technology, Vol.4, No.5, pp. 461-468,
2010. Affiliation:
• M. Sugi, H. Nakanishi, M. Nishino, Y. Tamura, T. Arai, and J. Ota, Associate Professor in Research into Artifacts,
“Development of Deskwork Support System using Pointing Gesture Center for Engineering (RACE), The University
Interface,” J. Robotics and Mechatronics, Vol.22, No.4, pp. 430-438, 2010. of Tokyo
• M. Sugi, M. Cheng, M. Yamamoto, H. Ito, K. Inoue, and J. Ota, “Online
Rescheduling in Semiconductor Manufacturing,” Int. J. of Automation
Technology, Vol.4, No.2, pp. 184-197, 2010.
Membership in Academic Societies:
• The Institute of Electrical and Electronics Engineers (IEEE)
Address:
5-1-5 Kashiwanoha, Kashiwa, Chiba 277-8568, Japan
• The Robotics Society of Japan (RSJ)
• The Society of Instrument and Control Engineers (SICE) Brief Biographical History:
• The Japan Society for Precision Engineering (JSPE) 2009- The University of Tokyo
Main Works:
• T. Hara and T. Arai, “Encourage non-designer’s design: sustainable
value creation in manufacturing products and services,” CIRP Annals –
Manufacturing Technology, Vol.61, Issue 1, (ISSN 1660-2773),
pp. 171-174, 2012.
• T. Hara and T. Arai, “Simulation of product lead time in design
customization service for better customer satisfaction,” CIRP Annals –
Manufacturing Technology, Vol.60, Issue 1, (ISSN 1660-2773), 2011.
• T. Hara, T. Arai, and Y. Shimomura, “A CAD system for service
innovation: integrated representation of function, service activity, and
product behavior,” J. of Engineering Design, Special issue on PSS, Vol.20,
No.4, pp. 367-388, http://dx.doi.org/10.1080/09544820903151715, 2009.
Membership in Academic Societies:
• The Japan Society of Precision Engineering (JSPE)
• The Japan Society of Mechanical Engineering (JSME)
• The Society for Serviceology (SfS)
• The Information Processing Society of Japan (IPSJ)

Int. J. of Automation Technology Vol.7 No.4, 2013 449


Nakanishi, H. et al.

Name:
Jun Ota

Affiliation:
Professor in Research into Artifacts, Center for
Engineering (RACE), The University of Tokyo

Address:
5-1-5 Kashiwanoha, Kashiwa, Chiba 277-8568, Japan
Brief Biographical History:
1989-1991 Nippon Steel Corporation, Kanagawa, Japan
1991- The University of Tokyo
1996-1997 Visiting Scholar at Stanford University, Stanford, CA, USA
Main Works:
• Z. Liu, H. Kamogawa, and J. Ota, “Motion Planning for Two Robots of
an Object Handling System Considering Fast Transition Between Stable
States,” Advanced Robotics, Vol.26, Issues 11-12, pp. 1291-1316, 2012.
• Y. Kung, Y. Kobayashi, T. Higashi, and J. Ota, “Motion Planning of Two
Stacker Cranes in a Large-Scale Automated Storage/Retrieval System,” J.
of Mechanical Systems for Transportation and Logistics, Vol.5, No.1,
pp. 71-85, 2012.
• T. Mizuno, M. Sakura, M. Ashikaga, H. Aonuma, R. Chiba, and J. Ota,
“Model of a sensory-behavioral relation mechanism for aggressive
behavior in crickets,” Robotics and Autonomous Systems, Vol.60, Issue 5,
pp. 700-706, 2012.
Membership in Academic Societies:
• The Japan Society for Precision Engineers (JSPE)
• The Robotics Society of Japan(RSJ)
• The Society of Instrument and Control Engineers (SICE)
• The Japan Society of Mechanical Engineers (JSME)
• The Institute of Electrical and Electronics Engineers (IEEE)

450 Int. J. of Automation Technology Vol.7 No.4, 2013

View publication stats

You might also like