Distributed Air Traffic Control
II: Explorations in a Test Bed
N ICHOLAS V. F INDLER1AND R ON L O2
A BSTRACT: Th s s the second of a two-part paper deal ng w th d str buted plann ng
for A
i r Traff
i
c Control. Three d
i
fferent organ
i
zat
i
onal structures have been
i
mplemented: the Local, Central
i
zed Arch
i
tecture, and the Locat
i
on Centered, Cooperat
i
ve Plann
i
ng System w
i
th one- and two-level Coord
i
nator-Coworker H erarch es. We present an n t al, s mpl f ed analys s of the speedup obta nable by us
i
ng the latter two organ
i
zat
i
onal structures. The veh
i
cle of our emp
i
r
i
cal stud
i
es, the D
i
str
i
buted A
i
r Traff
i
c Control Test Bed
i
s then
i
ntroduced. We d
i
scuss the des
i
gn and the results of a ser
i
es of exper
i
ments performed. We compare
i
n the test bed performance measures of the three systems us
i
ng the respect
i
ve organ
i
zat
i
onal structures. The compar
i
sons are made at d
i
fferent levels of traff
i
c dens
i
ty and problem s
i
ze,
i
n terms of commun
i
cat
i
on overhead and process ng t me needed for plann ng.
Keywords: D
i
str
i
buted Plann
i
ng and Problem Solv
i
ng; A
i
r Traff
i
c Control; System Recovery w
i
th Graceful Degradat
i
on; Locat
i
on Centered, Cooperat
i
ve Plann
i
ng System; Coord
i
nator-Coworker Control Structure; D
i
str
i
buted Scratch Pads.
List of Acronyms: ACT: A
i
r Traff
i
c Control, CCH: Coord
i
nator-Coworker H
i
erarchy, DAI: D
i
str
i
buted Art
i
f
i
c
i
al Intell
i
gence, DATC: D
i
str
i
buted A
i
r Traff
i
c Control, DPS: D
i
str
i
buted Plann
i
ng System, LCA: Local, Central
i
zed Arch
i
tecture, LCCPS: Locat on-Centered, Cooperat ve Plann ng System.
I NTRODU CTION
We present
i
n the second part of th
i
s paper the results of a ser
i
es of exper
i
ments compar
i
ng d
i
fferent organ
i
zat
i
onal structures: the Local, Central
i
zed Arch
i
tecture (LCA), and the Locat
i
on Centered, Cooperat
i
ve Plann
i
ng System (LCCPS) w
i
th one- and two-level Coord
i
nator-Coworker H
i
erarch
i
es (CCH). The latter two were
i
mplemented to control a D
i
str
i
buted Plann
i
ng System (DPS) for A
i
r Traff
i
c Control (ATC). The Coord
i
nator-Coworker parad
i
gm, d
i
scussed
i
n Part I of th
i
s paper,
i
s used as a gener
i
c metho
d of control. W
e have compared the respect
i
ve performances of the organ
i
zat
i
onal structures
i
n s
i
mulat
i
ng the A
i
r Traff
i
c Control env
i
ronment
i
n the D
i
str
i
buted A
i
r Traff
i
c Control (DATC) testbed. The compar
i
sons were made at d
i
fferent levels of traff
i
c dens
i
ty and problem s
i
ze n terms of commun cat on overhead and process ng t me needed for plann ng. (Note that the Local, Central
i
zed Arch
i
tecture can be v
i
ewed as a Coord
i
nator-Coworker
1Research Professor of Computer Science and Director of the Artificial Intelligence Laboratory, Computer Science Department, Arizona State U niversity, Tempe, AZ 85287-
5406
2At present Staff Member at the AT&T Bell Laboratories, Crawfords Corner Rd. Holmdel,
NJ 07733.
2
H erarchy of zero level. Further, the reason for stopp ng at two levels n the DATC
search tree i s that w i th the traff i c dens i ty preva i l i ng i n our exper i
ments, there was no need for ntroduc ng a flex bly chang ng number of levels.)
P R O C E S S I N G T I M E N EEDED FOR THE I N C R E M E N T A L S H A L L O W P L A N N I N G P ROCESS IN THE T HREE O RGANIZATIONAL S TRUCTURES S TUDIED
Incremental Shallow Plann i ng has been used to resolve potent i al i nc i dents i n the LCCPS. The process generates a DATC search tree. We have allowed n such trees up to two levels of branches. All branches need to be explored before the best plan-segment can be chosen. We call the development of each of such branches a task . A task lev el corresponds to the level of the branches the part cular task s responsi ble to develop. Let N 1 and N 2 be the number of the f i
rst- and second-level tasks, respect vely. The process ng t me needed to generate the DATC search tree w i th the Local, Central i zed Arch i
tecture can be calculated as
N 1N 2Σ t(T i ) + Σ t(T i,j )(1)
i=1 j=1where t(T i ) s the t me needed to complete task T i , and T i,j s the j -th level-two subtask of the level-one task T i .Let us now assume that there i s an unl i m i ted number of processors ava i lable.The process ng t me needed to generate the DATC search tree for LCCPS w th (1)
one- and (2) two-level CCH can be put as
N 2
Max { t(T i ) + Σ t(T i,j ) }
(2) j=1and Max { t(T i ) + Max {t(T i,j )}}
(3)
respect i vely. Here n i s the number of level-two subtasks of the level-one task T i .
W i th an unl i m i ted number of processors, relat i vely more parallel act i v i t i es i n the Coord i nator-Coworker structure become poss i ble w i th a two-level CCH, as can be seen i n the above formulae. Let us accept for the t i me be i ng the follow i ng s i mpl i fy i ng assumpt i ons:
?
the execut on t me for every task s dent cally equal to t ,
?the number of level-two subtasks of each level-one tasks i s equal to n (.e.,N 2 = n*N 1),
?the t i me needed for the coord i nator-select i on process i s negl i g i ble,
?the message-transm i ss i on-t i me for task-ass i gnment and result-report i ng requ i res negl i g i ble t i me, and
?every message i s responded to i mmed i ately after reach i ng i ts dest i nat i on.
3In th s case, the process ng t me needed for plann ng and the speedup obta ned by usi ng one- and two-level CCH i s shown i n Table 1.Archi tecture Processi ng Ti me Needed Speed-Ups Obtai ned LCA (N 1 + N 2)*t - LCCPS-1 ((N 1/N 2)+1*t N 1 - 1LCCPS-2 2*t ((N 1+N 2)/2) - 1Table 1 — Process ng t mes needed for plann ng and the speed-ups obta ned w th the three organ i zat i onal structures i n the i deal i st i c case when the s i mpl i fy i ng assumpt ons descr bed n the text hold However, n real l fe, none of the above assumpt ons hold exactly. Due to these compl i cat i ons, we cannot formulate analyt i cally the exact durat i on of the plann i ng processes. Therefore, we have des i gned and i mplemented a ser i es of exper i ments i n the DATC testbed. We have wanted to ga i n i n i t a better understand i ng i n a quant i tat i ve manner of the i mpact of each contr i but i ng factor to the overall effect i veness of the three organ i zat i onal structures stud i ed.A TESTBED TO STUDY DISTRIBUTED PLANNING IN THE AIR TRAFFIC CONTROL ENVIRONMENT
The DATC testbed i s wr i tten i n VAX LISP, a full vers i on of Common LISP,runn ng on a VAX-11/780 under the VMS operat
ng system. (In further stud es, we i ntend to use a cluster of workstat i ons wh i ch was not ava i lable at the t i me the work reported here was performed.) The ult i mate goal of creat i ng a testbed i s to prov i de a powerful and effect i ve s i mulat i on env i ronment for emp i r i cal stud i es on ATC wh i ch i nvolve several d i fferent chosen organizational structures for d i str i buted plann i ng.
It i s useful to cons i der mult i ple a i rcraft behav i or at two levels. The lower level i s nav i gat i onally or i ented; the h i gher level i s concerned w i th i nc i dent el i m i nat i on and i s strongly i nfluenced by the i mposed organ i zat i onal structure. For i nstance, when an a i rcraft dec i des to descend from a certa i n alt i tude, i t follows certa i n act i ons — a lower-level behav i or. A h i gher-level behav i or i s exh i b i ted when an a i rcraft detects a potent i al i nc i dent. Such h i gh-level behav i or character i zes the i nteract i ons between the a i rcraft and the consequences of such i nteract i ons. It also controls the d i str i buted plann i ng processes w i th i n the g i ven structure, such as the respons i b i l i ty delegated to some spec i f i c part i c i pat i ng a i rcraft.It i s plaus i ble to assume that every a i rcraft i
n the testbed uses the same planner to control i ts fl i ght and to ensure i ts nav i gat i onal safety. Th i s planner w i ll be called the d i str i buted planner. It i s also assumed that each a i rcraft i s able to perform sens i ng, look-ahead, potent i al i nc i dent detect i on, plan generat i on and execut i on, commun i cat i on, and i f appropr i ate, negot i at i on. Furthermore, all a i rcraft are categor i zed i
nto one of three d i fferent classes — namely, superjet (short for superson i
c jets), jet an
d p r o p
e l l e r . The classes have d i fferent
4performance character i st i cs, such as cl i mb speed, max i mum alt i tude, max i mum speed, etc. S i m i larly, the runway status has d i fferent mean i ngs w i th the three classes of a i rcraft, wh i ch i s cons i dered i n the Process Control Structure. Every a rcraft s supposed to have a flight plan and knows the a priori fl ght plan of each part i c i pat i ng a i rcraft. The a i rcraft attempt to follow the i r fl i ght plans as f i led.Due to i nteract i on w i th other a i rcraft and unforeseen c i rcumstances, they must
dev i ate from the i r fl i ght plans at t i mes. In do i ng that, they execute commands generated by vari ous planni ng processes. These command are unknown to the others unless they are requested through commun cat on.
Each a i rcraft undergoes var i ous phases dur ng ts fl ght. The phases are take-off, ascend i ng, cru i s i ng, descend i ng, hold i ng, approach preparat i on, approach i ng,and land ng. For each phase, the a rcraft has a temporary goal (e.g., ascend to a spec f ed alt tude, cru se at a g ven speed, etc.) and some expected behav or.There are three types of worlds n the testbed. The Real World (RW) reflects the s i tuat i on i n the global a i rspace conta i n i ng all the part i c i pat i ng a
i rcraft and also mon i tors the performance of an organ i zat i onal structure. The assoc i ated knowledge base i s rather r i ch. It i ncludes all a i rcraft's fl i ght parameters, fl i ght plans and the set of commands that can be ssued. Note that no nc dents occur n th i s world. (They must be detected and resolved by the a i rcraft i nvolved, i n advance.)Every a i rcraft ma i nta i ns the i
mage of a Simulated World (SW), wh i ch reflects i ts surround i ng env i ronment on i ts radar scope. It i s obta i ned by extract i ng the relevant i nformat i on from the Real World d i rectly.Each ai
rcraft creates and uses vari ous Extrapolated Worlds (EW) to ?detect d i screpanc i es between Extrapolated World and Real World s i tuat i
ons at the appropr i ate t i me po i nt,
?look ahead, and
?detect potent i al i nc i dents.
The Extrapolated World prov i des the a i rcraft an 'estimated future state' b y extrapolat i ng from the current s i tuat i on over t i me. (The extrapolat i on of the fl i ght paths can be both strai ght and curved li ne.)
The testbed can be run n e ther graph c or non-graph c mode. There are three worlds i n the graph i c mode. A three-d i mens i onal perspect i ve v i ew i s g i ven of the a i rspace from a part i cular pos i t i on.
In the Real World mode of presentat i on, the b i gger, square-shaped v i ewport i s used to show the Real World. The i mage consi sts of a gri d whi ch li es at the ground level, the mode of presentat on — Real World (RW) — s nd cated at the upper left corner, world t i me at the upper r i ght corner, and a set of a i rcraft i n the a i rspace.Each a i rcraft i s shown by a symbol appropr i ate to i ts type. A l i ne, cons i st i ng of exclamat i on marks, connects the a i rcraft and the gr i d. The po i nt where th i s l i ne i ntersects the gr i d shows the exact X and Y coord i nates of the a i
rcraft. Next to
5
th i s po i nt, there are three p i eces of i nformat i on: the a i rcraft ID, the current alt tude n feet d v ded by 100 (rounded), and the current speed n knots per hour
(rounded). All act ve a rcraft's IDs, the phase they are n, and the r current alt tude and speed ('a' and 's', respect vely) are shown n the trac ng (smaller, rectangular)v ewport.
In the S i mulated World mode of presentat i on, the S i mulated World i s d i splayed i n bas i cally the same format, w i th three except i ons. F i rst, the a i rcraft generat i ng the SW, i s shown i n the upper left corner along w i th the mode of presentat i on.Second, the a i rspace i s l i m i ted to the radar range of that a i rcraft. Th i rd, the
trac i ng v i ewport shows the current phase of the target a i rcraft,
i ts surround i ng a i rcraft and status i nformat i on i nd i cat i ng whether i t cont i nues the look-ahead or starts a new look-ahead process. In the s i tuat i on dep i cted, the a i rcraft i n quest i on does not have a plan yet and, therefore, t needs to look ahead a full Look-Ahead T i me per i od.In the Extrapolated World mode of presentat i on, the Extrapolated World i s
d splayed aga n n about th
e same format. The only d fference s that the a rspace
i s s i gn i f i cantly larger for the purposes of extrapolat i on.The testbed i s i mplemented on a un i processor. The d i ff i culty of s i mulat i ng the d i str i buted plann i ng process i s overcome ma i nly by the message passing capab i l i t i es prov i ded i n the testbed. S i m i larly to the objects i n an object-or i ented programm i ng env i ronment, each a i rcraft i n the testbed i s capable of send i ng and
rece i v i ng messages, and i s respons i ble for i ts own behav i or. D i fferent modules of the a i rcraft kernel (see F i gure 1 of Part I) are act i vated when a message i s rece i ved. These modules i n turn may tr i gger the commun i cat i on-un i t of the kernel to send further messages that w i ll act i vate the modules of other a i rcraft's kernel.
The handl i ng of t i me i n our un i processor-based testbed i s cr i t i cal. The DATC
testbed has to
?record the durat i on of each act i v i ty i n terms of actual (real) t i
me,
?ma i nta i n the correct current Real World t i me from the perspect i ve of each a i rcraft, and
?i ntegrate the above two funct i ons i n the s i mulat i on of mult i ple a i rcraft act i v i t i es.
The testbed prov i des several fac i l i t i es for the above. A funct i on called processor-time returns the CPU t i me elapsed s i nce the testbed started. Know i ng the beg i nn i ng and end i ng processor t i me of a part i cular act i v i ty, one can calculate i ts durat i on.
The testbed also ma i nta i ns a so-called aircraft world time for each a rcraft. It i s the Real World t i me from i ts perspect i ve, and i s recalculated and ass i gned to the respect ve a rcraft at each of the follow ng testbed stages :
?the Real World has just been updated,
?the a i rcraft i n quest i on has just f i n i shed i ts current "percept i
on phase",
6
?the a rcraft has just processed a message.
Before an a rcraft perce ves or processes a message, the Real World s updated
appropr
i ately to reflect the correct s
i
tuat
i
on. After the a
i
rcraft has f
i
n
i
shed
i
ts
current percept
i on phase or the process
i
ng of a message, a new a
i
rcraft world
t i me
i
s ass
i
gned to
i
t. Th
i
s reflects the t
i
me taken for the perce
i
v
i
ng or message
process ng act v ty, and helps n the bookkeep ng of actual t mes.
When an a
i rcraft sends a message, a t i m e-s t a m p
i
s calculated and
i
s
assoc
i ated w
i
th the message,
i
nd
i
cat
i
ng
i
ts send
i
ng t
i
me. Its value
i
s the sum of
the current a
i rcraft world t
i
me of the sender and the durat
i
on of the current
a i rcraft act
i
v
i
ty so far.
T HE D ESIGN OF THE T ESTBED AND THE R ESU LTS OF E MPIRICAL I NVESTIGATIONS
The compar
i son between the performances of the three arch
i
tectures stud
i
ed
i
s
i n terms of commun
i
cat
i
on overhead and process
i
ng t
i
me needed for the plann
i
ng process, at d
i
fferent levels of traff
i
c dens
i
ty and problem s
i
ze. We have prepared 18 scenar i os wh i ch d i ffer from each other e i ther i n the number of part i c i pat i ng a
i
rcraft (traff
i
c dens
i
ty) or
i
n the number of branches of the DATC search tree (problem s ze) or n the number of levels of CCH (arch tecture). We have employed three d
i
fferent traff
i
c dens
i
t
i
es, two problem s
i
zes and three arch
i
tectures. The follow
i
ng convent
i
on
i
s used for referenc
i
ng a scenar
i
o. The reference cons
i
sts of s
i
x characters
i
n the form:
DdSsAa
The cap
i
tal letters D, S, A are used to
i
nd
i
cate traff
i
c dens
i
ty, problem s
i
ze and arch
i
tecture, respect
i
vely. The lower case letters d, s and a stand for numbers. Table 2 shows the possi ble values and thei r meani ng.
i
i
i
i
Table 2 — The meani ng and possi ble values of the parameters
d, s and a accord ng to the scenar o reference convent on The problem of s
i
ze one (S1) means that there ex
i
st two level-one and four level-two tasks dur
i
ng the
i
nc
i
dent resolut
i
on process. (Note that there are at most four tasks ava
i
lable
i
n th
i
s problem. Th
i
s
i
s because we allow level-two branches to be developed only after all level-one branches are explored.) The
7problem of s ze two (S2) means that there ex i st three level-one and s i x level-two tasks dur ng the nc dent resolut on process. (The above cho ces were made after a careful cons i derat i on of comput i ng t i me and memory requ i
rements. In future stud i es, us i ng a cluster of workstat i ons, such l i m i tat i ons w i ll be of much less concern.)We assume a baud rate (the number of b ts per second that can be transm tted)of 9600 and that i t requ i res 10 b i ts to transm i
t a character. Thus, the transm ss on t me needed for a message m i s equal to (message-length(
m )*10)/9600We defi ne message-delay-in-response as the di fference between the message-response-time and the message-receipt-time .
When the number of relevant a rcraft ncreases, t should have a pos t ve effect on systems based on the arch tectures w th a one- or two-level CCH (A1 and A2,respect i vely). Namely, the extra processors work i ng i n these cases speed up the plann ng process by mak ng better use of the ava lable process ng power. The value of the speedup i n us i ng the respect i ve arch i tectures A1 and A2 can be calculated accord i ng to the follow i ng formulae:
(t(P A0)/t(P A1)) - 1(5)(t(P A0)/t(P A2)) - 1 (6)
Here t(P A0), t(P A1) and t(P A2) are the t mes needed for the Incremental Shallow Plann i ng process i n the Local, Central i zed Arch i tecture (A 0) and i n the LCCPS arch i tectures w i th one- and two-level CCH, respect i vely.The result of the compar i sons, i nvolv i ng test runs of 18 scenar i os, shows that
the arch i tectures A1 and A2, n general, are more eff c ent than A0. However, for two scenar i os (D 2S 1A 1 and D 2S 2A 1), each w i th two part i c i pat i ng a i rcraft, the arch i tecture A 1 d i d not fully ut i l i ze the ava i lable processors. More i mportantly and surpr i s i ngly, arch i tecture A 2 d i d not perform better than arch i tecture A 1 i n almost every scenar i o except when only two a i rcraft part i c i pate. Clearly, the management of the ava lable processors plays an mportant role. In the trace dump of the s i mulat i
on outputs generated by the DATC testbed, we have found that the delay-i n-response to messages i s the major cause for th i s phenomenon, wh i ch i s expla i ned next.
An a i rcraft can process the messages rece i ved only after hav i ng f i n i shed what i t i s currently do i ng (execut i ng a task, process i ng a message, etc.). Thus, i f some i mportant "upstream" plann i ng act i v i t i es depend on the process i ng of such messages, the whole plann i ng process i s ser i ously affected. We have found two types of s i tuat i ons contr i but i ng to extended delays-i n-response. F i rst, when a coord i nator i s execut i ng a task, all task-request i ng messages must be temporar i ly suspended. Th i s places the task requesters (the ava i lable processors) i n a wa i t i ng state. Second, the conf i rmat i on process of a coord i nator w i
ll also come to a halt
8 when the nom nee s execut ng a task. Th s also slows down the plann ng process cons derably.
In order to m
i n
i
m
i
ze the probab
i
l
i
ty of occurrence of the f
i
rst type of
s i tuat
i
on, we have added more
i
ntell
i
gence to the a
i
rcraft processors. We have
der
i ved several heur
i
st
i
cs for the coord
i
nators to follow. One such heur
i
st
i
cs for a
level-zero coord
i nator
i
s: do not request a task unless all the other a
i
rcraft have
been assi gned tasks.
Note that th
i s heur
i
st
i
c cannot be appl
i
ed to the resolut
i
on of f
i
rst-level
i nc
i
dents s
i
nce several coord
i
nators would then be wa
i
t
i
ng for the same reason at the same t
i
me, result
i
ng
i
n deadlocks. We not
i
ce when the number of relevant a
i
rcraft
i
s larger than the number of f
i
rst-level branches, some or all of the level-one coord
i
nators must wa
i
t so that other ava
i
lable processors may be able to respond as well. The number of level-one coord
i
nators to wa
i
t
i
s the least of the number of f
i
rst-level branches and the d
i
fference between the number of the relevant a
i
rcraft and the number of f
i
rst-level branches. Let us call th
i
s m
i
n
i
mum value n. We have caused the level-zero coord
i
nators to requ
i
re n level-one coord nators to wa t. However, such wa t s only temporary and s n order to avo d deadlocks. Its durat
i
on should also be relat
i
vely short for reasons of eff
i
c
i
ency.
We have found no obv ous cure for the second type of s tuat on contr but ng to extended delays n response. We could e ther let the processors be nterrupt ble or el
i
m
i
nate the nom
i
nat
i
on process needed for the resolut
i
on of level-one
i
nc
i
dents. We have chosen the latter method
i
n our stud
i
es because the number of coord nators requ red for the two problem s zes are small (3 for S1 and 4 for S2). We have thus mod f ed the algor thm so that when a coworker detects a new level-one
i
nc
i
dent,
i
t becomes a self-appo
i
nted coord
i
nator for that
i
nc
i
dent
i
n the arch
i
tecture w
i
th a two-level CCH.
We can also not
i
ce two m
i
nor causes for some
i
neff
i
c
i
ency
i
n plann
i
ng. F
i
rst, when a coworker wants to request a task from a group of coord nators at a certa n level
i
t works for,
i
t uses a scheme of equal-pr
i
or
i
ty
i
n select
i
ng the task-source to request from. If the coworker processor
i
s also
i
n
i
ts own task-sources-l
i
st (.e., the processor has the role of both the coord nator and the coworker), t may end up w
i
th request
i
ng a task from another coord
i
nator. Th
i
s results
i
n the generat
i
on and process
i
ng of external messages w
i
th task-requests and task-ass
i
gnments and,
i
n turn, decreases the average processor ut
i
l
i
zat
i
on, slow
i
ng down the plann ng process tself. We have, therefore, added a rule for the coworker n request ng tasks; namely, always request a task from tself whenever there s a cho
i
ce (
i
nternal messages take no transm
i
ss
i
on t
i
me).
The other m
i
nor cause for
i
neff
i
c
i
ency
i
n plann
i
ng we have found
i
s that even when a g
i
ven coord
i
nator knows that
i
t has no other task to be executed,
i
t w
i
ll not automat
i
cally not
i
fy the other a
i
rcraft about th
i
s s
i
tuat
i
on. Th
i
s may create many cycles of task-request and no-task-ava
i
lable messages,
i
ncreas
i
ng the
i
dle t
i
me of the task requesters. We have solved th
i
s problem by mak
i
ng the
9
coord
i nator not
i
fy the other a
rcraft as soon as t s go
i
ng to ass
i
gn the l a s t
unass
i gned task ("no-other-task-ava lable" message).
We have made the appropr ate mod f cat ons and run 12 out of the 18 scenar os
aga
i n (the scenar os nvolv ng arch tecture A0 are not affected by the added
heur st cs). We have found that by a careful message pass ng management w th the
arch
i tectures A1 and A2, the average delay-n-response s lower and the relevant
a i rcraft become a more effect ve and eff c ent team for problem solv
i
ng. Tables 3
through 6 show the correspond ng results.
In Table 3, we f
i nd that the total number of external messages generated and
processed
i s lower for the arch
i
tecture A2.
Table 3 — The average delay-n-response of a s ngle message
for scenar os nvolv ng A1 and A2
Table 4 — The length of the ncremental shallow plann ng
process for scenar os nvolv ng A1 and A2
The average delay-n-response (n seconds) of a s ngle message s lower n the second group of test runs than i n the fi rst one.
Table 5 — The percentage of speedup obtai ned by usi ng
one- and two-level CCH for scenar os
10W i th the program mod
f cat ons descr bed, the process
ng t me (n seconds)requ i red for the Incremental Shallow Plann ng Process reduces drast cally. Such mprovements are shown to be even more deci si ve i n Table 6. Furthermore, because of the careful sequenc ng of messages that conta n task requests, the processor ut i l i zat i on rate also i ncreases.
Table 6 — The average degree of processor ut l zat on dur ng the incremental shallow planning process over all relevant aircraft Table 4 to 6 are causally related to each other. As noted before, by a careful message pass i ng management for the arch i tectures A1 and A2, the average delay-i
n-response to a s i ngle message over all scenar i os w i ll be shorter. Th i s has been shown by the data collected n the second group of test runs. We have found the average degree of processor ut l zat on and the percentage of speedup to be h gher over all scenar os. The length of the ncremental shallow plann ng process has also become much lower. F gure 4 shows the percentage of speedup n a graph cal form for problem s zes S1 and S2, respect vely.D2D3D4112D2D3D4
F I
G . 4 — The percentage of speedup obta i ned by us i
ng a one- and two-level Coord i nator-Coworker H i erarchy (A 1 and A 2) at three d i fferent a i r traff i c dens i t i es (D1, D2 and D3) for problems of si ze 1 (S1) and si ze 2 (S2).
From the above two d agrams, we judge A1 to be a better choi ce i f the number of f i rst-level branches i s equal to the number of relevant a i rcraft (as i n scenar i os D2S1 and D3S2). The add i t i onal message pass i ng act i v i ty i n connect i ng level-two tasks w i th level-one coworkers has a negat i ve effect because there are no extra processors ava i lable for tak i ng level-two tasks. When the number of f i rst-level branches s less than the number of relevant a rcraft by two , A2 becomes a better cho i ce (as i n scenar i o D 4S 1) because the extra processors employed i n the plann i ng process can now take care of the add i t i onal message pass i ng act i v i ty.However, i f the d i fference between the number of level-one branches and the number of relevant a i rcraft i s only o n e (as
i n scenar i os D 3S 1 and D 4S 2), the add i t i onal message pass i ng act i v i ty reduces the overall system performance.
F i nally, i f the number of f i rst-level branches i s greater than the number of relevant a i rcraft, A 2 i s a better cho i ce (as i n scenar i o D 2S 2) because A 2 i s a more flex i ble control structure than A1 — the processors can change roles more freely i n i t.The overall conclus ons are ?i t i s feas i ble to use LCCPS for d i str i buted A i r Traff i c Control,
?by a careful message pass i ng management for arch i tectures w i th a one- or two-level CCH, the relevant a i rcraft become an effect i ve and eff i c i ent problem solv ng team, and ?wh i ch the better arch i tecture i s depends on the d i
fference between the
number of relevant a i rcraft and the number of f i rst-level branches i n the DATC search tree 3.Unfortunately, due to comput i ng bottlenecks, we have not able to collect s i mulat i on data i nvolv i ng a h i gher number of part i c i pat i ng a i rcraft and to prov i de more i nformat i on about the th i rd statement above. (For some of the most demand ng scenar os, we have used almost 9 megabytes of core memory, and over
1400 L sp funct ons have been employed.) It s, however, qu te certa n that when a cluster of workstat i ons i s used for s i m i lar stud i es, the computat i onal bottlenecks would d sappear and s tuat ons w th much h gher traff c dens t es can be analyzed.C ONCLU SIONS
The research reported i n th i s two-part paper covers the des i gn and i mplementat i on of a DPS, the Location Centered, Cooperativ e Planning System (LCCPS), for DATC. The results of exper i ments character i ze the effect i veness of d i fferent organ i zat i onal structures. The part i t i on i ng of nodes i s demand-dr i ven.Groups of a i rcraft, i dent i f i ed w i th the i r locat i on, are organ i zed accord i ng to a spec i f i c structure to resolve potent i al confl i cts — hence the name Locat i on Centered, Cooperat i ve Plann i ng System. The DATC testbed i s used not only to d i splay the lower-level nav i gat i onal aspects of fly i ng but also to perform the 3In more general terms, each first-level branch is created to solve a partitioned subproblem. Thus, with a given architecture and a certain number of processors,partitioning of the given problem also plays an important role for efficient group p l a n n i n g.
control act
i v
i
ty for d
i
str
i
buted plann
i
ng. Cruc
i
al fac
i
l
i
t
i
es have been
i
dent
i
f
i
ed
and
i mplemented to make the testbed as user-fr
i
endly as poss
i
ble. The fac
i
l
i
t
i
es
ava lable n t can also be employed n bu ld ng a general-purpose testbed for DAI research.
We have developed organ
i zat
i
onal structures w
i
th one- and two-level
Coord
i nator-Coworker H
i
erarchy. The
i
r performance has been compared w
i
th that
of the Local Central
i zed Arch
i
tecture (wh
i
ch can be v
i
ewed as a zero-level
Coord
i nator-Coworker H
i
erarchy). The compar
i
son
i
s
i
n terms of commun
i
cat
i
on
overhead and the effect
i veness of confl
i
ct resolut
i
ons. It has been shown
emp
i r
i
cally that one- and two-level Coord
i
nator-Coworker H
i
erarch
i
es prov
i
de
i
ncreased process
i
ng eff
i
c
i
ency and capab
i
l
i
t
i
es,
i
mproved flex
i
b
i
l
i
ty and rel
i
ab
i
l
i
ty, and lower process
i
ng costs.
In summary, the results of our stud
i
es fall
i
nto the follow
i
ng three categor
i
es of ach
i
evement:
?the des
i
gn and
i
mplementat
i
on of mechan
i
sms to deal w
i
th the problems of connect
i
on, commun
i
cat
i
on, uncerta
i
nty and coherence,
?the des gn and mplementat on of a general-purpose testbed for DAI research,?the qual
i
tat
i
ve and quant
i
tat
i
ve evaluat
i
on of the feas
i
b
i
l
i
ty of delegat
i
ng plann
i
ng respons
i
b
i
l
i
t
i
es to processors on each
i
nd
i
v
i
dual a
i
rcraft
i
n a DATC reg
i
me.
The stud
i
es have
i
mproved our understand
i
ng of D
i
str
i
buted Art
i
f
i
c
i
al Intell gence, n general, and have demonstrated the mportance of DAI n the world of DATC, n part cular.
F
i
nally, the follow
i
ng future research d
i
rect
i
ons may be po
i
nted out:?add ng a pr or ty level to each message so that more urgent messages can be processed earl
i
er,
?add ng more ntell gence to the Message Process ng Module so that messages result ng n h gher product v ty of the whole system can be responded to earl er,?enabl ng some of the plann ng processes to be nterrupted so that messages can be responded faster, wh ch n turn mproves processor management,?study
i
ng the effects of no
i
se
i
n message transm
i
ss
i
on,
?study
i
ng the effects of lett
i
ng coord
i
nators choose an organ
i
zat
i
onal structure
i
n wh
i
ch the number of levels of the Coord
i
nator-Coworker h
i
erarchy depends on the si tuati on, and
?
i
mplement
i
ng a mult
i
processor vers
i
on of the DATC testbed for s
i
mulat
i
on purposes.
These stud
i
es w
i
ll have to be done
i
n us
i
ng a cluster of workstat
i
ons to el
i
m
i
nate computat
i
onal bottlenecks. It
i
s hoped that our present results w
i
ll pave the way to future mplementat ons of a LCCPS for A r Traff c Control.
A PPENDIX I. R EFERENCES
Cammarata, S., McArthur, D., and Steeb, R. (1985). "Strateg i es of cooperat i on i n d i str i buted problem solv i ng." Proc. of the Ninth IJCAI Conf., Los Angeles, CA, 767-770.
Chambers, A. B., and Nagel, D. C. (1985). "P i lots of the future: Human or computers?" Comm. of the ACM , 28, 1187-1199.
Ch i en, R. T. (1982). "Art
i f i c i al Intell i gence and human error prevent i on study i n ATC systems." Final Report , Department of Transportat i on and Federal Av i at i on Adm i n i strat i on, DOT-FA79WA-4360.Cork i
ll, D. D., and Lesser, V. R. (1983). "The use of meta-level control for coord i nat i on i n a d i str i buted problem solv i ng network." Proc. of the Eighth IJCAI Conf., Karlsruhe, West Germany, 748-756.
Dav i s, R., and Sm i th, R. G. (1983). "Negot i at i on as a metaphor for d i str i buted
problem solv i ng." Artificial Intelligence . 20, 63-109.F i ndler, N. V., and Lo, R. (1986). "D i str i buted plann i ng i n a i r traff i c control."Journal of Parallel and Distributed Computing . 3, 411-431.Hunt, V. R., and Zellweger, A. (1987). "Strateg i es for future a i r traff i c control systems." IEEE Computer , 20, 19-32.Lesser, V. R., and Cork i ll, D. D. (1981). "Funct i onally accurate, cooperat i ve d i str i buted systems." IEEE Trans. on Systems, Man and Cybernetics , SMC-11, 81-96.
McArthur, D. R., Steeb, R., and Cammarata, S. (1980). "A framework for d str buted
problem solv ng." Proc. of the AAAI Conf., Stanford, CA, 181-184.Sm i th, R. G. (1977). "A framework for D i str i buted Problem Solv i
ng." Proc. of the Fifth IJCAI Conf., Cambr dge, MA, 836-841.
Steeb, R., Cammarata, S., Hayes-Roth, F. A., Thorndyke, P. W., and Wesson, R. B.(1981). "D i str i buted i ntell i gence for fleet control." R e p o r t R-2728-ARPA, The RAND Corp., Santa Monica, CA.
Swetram, G. F. Jr. (1981). "A prel i m i nary character i zat i on of the AERA man-machi ne i nterface for ATC computer replacement." MITRE Corp. Working Paper , WP-81W00135, McLean, VA.
Thorndyke, P. W., McArthur, D., and Cammarata, S. (1981). "Autop lot a d str buted planner for ai r fleet control." Proc. of the Sev enth IJCAI Conf., Vancouver, Canada,171-177.
Wesson, R. B. (1977). "Plann ng n the world of the a r traff c controller." Proc. of the Fifth IJCAI Conf., Cambr dge, MA, 473-479.
Wi ener, E. L. (1985). "Beyond the steri le cockpi t." Human Factors , 27, 75-90.
三星通关秘籍! 终于通关了!说说我的经验. 以下是我的阵容:
可见我的阵容并不强大,两个T都没满级,DPS也相当不给力.唯一特色是控比较多.这也是我能顺利通关的最重要的原因. 另外很多人也许会吐槽我的副T凤凰,之所以把它放在这个位置,是因为他的治疗技能,很多时候可以拯救世界.以下是他的技能: 凤凰冲击2级:攻击敌方全体目标,造成145%威力伤害,有几率触发集火 烈火精灵2级:召唤小凤凰,为自己以及队伍中血量最少的2名队友恢复25%血量 烈日炙烤0级:这个你要是学了,就废了. 下面请我的主力英雄隆重登场! 第一控:小小 山崩2级:攻击前排和中间群体敌人,造成120%威力伤害,并造成敌方眩晕,持续2秒 说明:此招一出,除了boss以外的所有野怪都被晕2秒! 每回合可释放次数:2次 第二控:流浪剑客 风暴之锤2级:攻击敌方中间竖排敌人,造成85%威力伤害,并造成敌方眩晕,持续2秒 说明:此招一出,boss和面前的2个野怪被晕2秒! 每回合可释放次数:8次 (15级的蓝色流浪剑客,每次耗蓝只有907) 第三控:萨尔 动能力场1级:攻击敌方所有目标,造成60%威力的伤害,并降低目标移动速度40%,持续5秒,有几率触发集火
说明:此招一出,所有怪减速40%,和流浪剑客的晕完美配合,很多时候怪都没有出手的机会! 每回合可释放次数:4次 既然控制这么给力,那么如何将控制的效率发挥到极致呢? 重要的有3点: 1. 要有速度! 以下是我的装备,可供大家参考. 值得一提的是那件”破损的远祖战鼓”.是可以和普通远祖战鼓完美叠加的,包括被动加速和主动加速,群体加速30有木有!
Galaxy S6/S6E完全精简攻略 1、【绝对不能删】删除后系统无法正常运行的应用 /system/app下: BadgeProvider(应用程序脚标服务,删除后不停报错) InCallUI(通话过程服务,删除了无法使用任何通话**能,也无法挂断电话) SimCardMgr(双卡管理服务,删除了无法正常通讯) /system/priv-app下: DefaultContainerService(应用容器,系统基础服务) ExternalStorageProvider(存储器,系统基础服务) LogsProvider(通话记录服务,删了不能打电话) MtpApplication(USB连接服务,删了连不了电脑,还会报错) SecMediaProvider(存储器,系统基础服务) SecSettings2.apk(设置,系统基础服务) SecSettingsProvider2.apk(设置,系统基础服务) InputDevices(输入服务,删了自己想吧) SecContacts_L_Phone_FLAGSHIP_CHN(联系人,删了无法管理联系人) SecContactsProvider(联系人存储,删了无法存储联系人) SharedStorageBackup(共享存储备份,系统基础服务) SystemUI(系统UI,系统基础服务) Telecom(电话,删了没信号) TeleService(通讯基础服务,删了没信号) 简单点说,以上东西就是S6上最小的系统,包括数据连接在内的核心通讯功能全部正常。缺点就是众多**功能缺失,但如果你就打电话发短信,只用几个固定应用,不拍照不用蓝牙,且事先已经安装好常用APP(如微信、浏览器什么的),那么也就够用了,适合疯狂追求最简的人们。这里需要提醒的是,我列出的只是文件夹形式存在的App,在/system/app目录下还有一个文件名很长的单独文件(忘了名字,明天查查再补充),那个千万不能删,删了直接就无法引导了。 2、【强烈建议不要删】可以删、但删除会导致较为严重后果的应用: /system/app下: Bluetooth(顾名思义,删了就用不了蓝牙) NfcNci(NFC服务,删了NFC就挂了,不过这个用的人应该很少) mcRegistry(删除后wifi开关巨慢、不能存储密码且系统性能会下降) PackageInstaller(应用安装服务,删了安装不了应用) PacProcessor(应用处理服务,删了安装不了应用) SecHTMLViewer(HTML浏览服务,很多应用需要调用这个才能正常现实HTML内容) WebViewGoogle(新版HTML浏览服务,和上面那个功能一致,但使用的是Webkit核心,很多新应用调用的是这个)
愤怒的?鸟:?约?冒险--全三星图?攻略 三星要点: 1.尽量?最少的?鸟来完成任务 2.攻击?标应该为障碍物的薄弱点,譬如?柱 3.利??些附加的障碍物来达到摧毁其他建筑物的?的 S M U G G L E R'S D E N 1-1 第?关很简单,你所使?的是三个红??鸟;只需调整?向击中右部的中间?柱就可以了 1-3 如下图,射击的?标瞄准障碍物的下部。同时要让你的红??鸟在最后击中最右边的笼?,然后让整个障碍物失去平衡往左边下落。??个?鸟完成任务即可三星
1-5 这?关也是,让红??鸟摧毁左边的笼?;然后让其下落的时候击中右边的障碍物下部来让右边的建筑物失去平衡。 1-6 这?关会出现新的?鸟,蓝??鸟。这种?鸟的威?在于,它们?次可以射出三只?鸟,你可以让它们各个突破,击中?系列?标,但最理想的状态是,让它们三者近距离出击同?个?标,发挥最?攻击?。 射出之后再点击?次?标左键即可分为三只?鸟
这?关很简单,按照如下的路线在半程散射即可完成本任务 1-7 这?关要在快接近?标的时候散射,同时要确保三个?鸟有?少?个击中中间障碍物的最下部;让其失去平衡从?完成任务
1-8 这?关有相当的技巧,路线如下。两个?鸟攻击下?的障碍物,然后?个?鸟将最右上的?头击落从?让其下落把所有的笼?砸开 1-9 这?关还是散射,注意要保证三个?鸟能够击中障碍物的下部
1-10 这?关的?的就是让你彻底摧毁前?两个建筑物,然后让其剩余的物件如同多?诺?牌?样把后?的建筑物推到完成任务 1-11 本关会开启新的黄??鸟。这种?鸟虽然极其神速,但碰到?壁和玻璃墙时的攻击?仍然有限。?这种?鸟进攻?头壁垒简直像捅破?层纸?样好?,这样你就可以闪电袭击敌?,让它们葬?于堡垒的废墟之中。?且在发射之后你还可以按?次?标左键进?加速攻击
魔灵召唤三星魔灵最佳选择及培养攻略 魔灵召唤三星魔灵魔灵召唤三星魔灵怎么选择培养,想了解一下更多有关魔灵召唤三星魔灵魔灵召唤三星魔灵怎么选择培养最新的攻略及资讯吗?下面就由小骨来为大家带来魔灵召唤三星魔灵魔灵召唤三星魔灵怎么选择培养的推荐 魔灵召唤三星魔灵最佳选择及培养建议。今天一起来看下大神给小白们的建议吧,尤其是前期可选择的三星魔灵的相关讲义,和一些中后期有潜力的三星魔灵推荐,魔灵召唤这款游戏非常的有内涵,需要小伙伴们深入的去体验才能更好的游戏,属于慢热经久耐玩的游戏类型。 火系: 1,火犬神 火犬为什么是新手第一选择狗粮队长,除了它的技能优势,被动是消灭敌人后获得额外攻击次数。也是因为相对容易升技能,得到渠道也简单,火山可以刷到,水犬,风犬前面地图就有,很容易满技能,付文不好可以带沙漠,付文好就可以带火山,当然你有好的胎4.胎5就另说了,不然就是平民第一狗粮队长选择。数据请自己查找,一般五星犬神,付文一般的带沙漠3,付文好带普通火山,六星好付文可带地狱沙漠,或者困难火山!注意普通火山速度需要110以上。另外犬神,竞技场,爬塔等其他地方表现也很好!付文,猛攻刀刃,2.46攻击,或者2.6攻击,4爆率,堆点速度! 2,火死神 这个也是狗粮队长,但它是中后期,因为前期比较难得到,全靠抽取,并且技能难升,而且付文不好时候并不一定比犬神快!但技能,付文,上6星,带困难,地狱火山都是一把好手!付文参考犬神!选择升5,有兴趣才6! 3,火地狱火 这个在遗迹有掉,也可以作为狗粮队长,但它的优势在于2技能满后,群体90%破甲,是前期三星唯一一个高概率群破甲!中期没水海盗,水偷,也可用他代替,3技能群晕一回合也不错!如果
全系别增益整理与简析 恶魔系: 实用性:★★★★★ 作用:全员加180护甲。赤裸裸的土豪系,180护甲给谁都好用! 野兽系: 实用性:★★★☆☆ 作用:全员加182力量。看似实用的属性,给了四个奇葩的前排壮汉… 巨人系: 实用性:★★★★★ 作用:全员加180魔抗。同样是前排巨汉,比野兽系的几个强悍太多。 斧系: 实用性:★★☆☆☆ 作用:全员加180力量 这是一个极其混乱的系别,宙斯也算斧系… 巨魔系: 实用性:★★★☆☆ 作用:提供生命回复。巨魔这个曾经的竞技场主角,能不能翻身?我感觉有点悬。 飞行系: 实用性:★★★★☆ 作用:全员加810法强。这是一个版面输出很高的团队技能,唯一的缺点是很难组合。
鱼人系: 实用性:★★★★☆ 作用:全员加180敏捷。非常好的一个技能,只可惜大鱼人连个闪避都没有,还站位那么靠前… 近卫系: 实用性:★★★★☆ 作用:全员加3960生命值。五小强中的三个着实火了一阵,相信奇迹相信梦想! 刀剑系: 实用性:★★★☆☆ 作用:全员加180点物暴。唯一能输出的仅有剑圣和火猫,且都难堪主T 位置,遗憾! 机械系: 实用性:★★★★☆ 作用:全员加27%的攻速和施法。有人说飞机是逗比,没错,但这确实是个好技能啊。(等待伐木机和发条登场咯) 西行系: 实用性:★★★☆☆☆ 作用:全员加72点魔抗。猴子强,则西行强,反之亦然。猴子和刚被的觉醒决定了这支队伍的未来走向。 弓箭系: 实用性:★★★★★☆ 作用:全员加36.4命中。受益最大的自然是一姐,信一姐得永生! 总结: 英雄的系别构成是游戏非常重要的一个环节,因为仅仅需要两个同系别的英雄就能形成不错的收益,相当多了一件非常不错的高级装备。而这些属性的获得方式也比洗练来得更为简单和划算
水太深了,防骗必看!史上最全的三星S3 i9300鉴定攻略-购买指南(超详细) 这是我个人买9300手机时参考用的,推荐 大家下载 首先公布两个容易造成误导的网站,这两个网站可以查询手机的型号和主板,查询销售地无效,在这两个网站上面查不到也不代表是假的: https://www.doczj.com/doc/6514070257.html,/ https://www.doczj.com/doc/6514070257.html,/?page=analysis&sub=imeinr 另外,看串号第几位来识别版本的方法就更离谱了,10年前就不可靠了。 言归正传: 先看主机: 一:看外观,成色怎么看就不用细说了,你懂的。以下主要讲三点:1:标签:看S/N码。S/N 码第4、5位分别代表出厂年份和月份。8代表08年,9代表09年,A代表10年,B代表11年,C代表2012年。第5位代表月份。如图这台手机,C7就说明是2012年7月份出厂的(电池上的S/N 码同理)。另外说明:港版的标签是中文的, ps:最新产的港版白色,电池仓标签,已经不是made by sumsang,而是直接“三星电子(株)韩国制造”,如图
2:看螺丝标:港版有蓝色螺丝标,写着原装标贴,撕毁无效。如图
3:看螺丝,估计有人问我发个螺丝干嘛??其实我是回答你们最关心的问题“如何鉴别翻新机”,以三星为例,你们把手上的机子,随便拆一个螺丝出来看,如果螺丝上面有上图那样有油漆式的痕迹,那么恭喜你,是全新机,如果是全光滑的表面,那么是被换过的螺丝,翻新几率99%,这时候你们会质疑了,这么一个螺丝可以断定,我可以告诉你,这是行业内最直接最简单的鉴别方式,市场回收机都是这样检验的,信不信,你就拆一个螺丝看看,如果是光滑的,那么建议你去拆机验证,如果不是我说的,我可以帮你出拆机费,好了,我又曝光了一个行业内幕,大家可以试试,就一个螺丝,对手机没影响。如图
《使命召唤6现代战争2》特别任务单人3星全关卡心得攻略 前言: 本人是个FPS苦手,基本不太玩FPS,大部分FPS都会晕,COD4,COD5也会晕, 但COD6的特别任务太好玩了,欲罢不能(COD6居然完全不晕了,哈哈) 写了点心得,希望帮助跟自己一样苦手或更苦手的玩家,别半途而废 特别任务单人3星看似很难但都有方法过去的哈哈,我都过了,你也可以的 只要用心,一定能过的,诀窍一个字-----“心”....呵呵 ----------------------------------------------------- 提示: 先玩1星难度熟悉流程寻找套路,再攻3星会好打很多 每关开始研究地图,研究地形,研究敌人的规律 多利用屏幕左上的小地图,敌人开枪时会显示位置 有时可以跳一跳,打几枪引敌 地雷是好帮手,插在敌人必经路看不见的拐角有大用 只有一条命,所以慢慢摸索,以静制动,敌人会经常来找你或绕后偷袭你的 ----------------------------------------------------- 注意: 为了不影响你自己的过关乐趣,最好先不要看,一旦看了将严重影响你通关的乐趣 游戏的乐趣在于自己攻克难关,达成3星时的兴奋与激动,所以慎重观看 实在苦手,实在过不去再看,不然你会后悔的,知道方法后将缺失了很多乐趣 ----------------------------------------------------- ■1-1-★★★训练关 多练练,边跑边打,用手枪m9,或mp5好点,熟练是关键 ----------------------------------------------------- ■1-2-★★★高台防守耍耍狙 第1波敌人杀到留下最后1个,然后下去四周拐角处埋点地雷,高台上也埋点 每波开局后的控制飞弹是关键,尽量炸多的人和一开始来的车,车上会下人 时刻注意是否有敌人靠近(用显示敌人位置的枪) ----------------------------------------------------- ■1-3-★★★巴西小巷好多人 开局拿到左前方凳子上有带显示敌人位置的枪是关键,稳扎稳打,注意会有敌人包抄绕后 某些地方埋点地雷防绕后敌人和狗 ----------------------------------------------------- ■1-4-★★★雪地谜踪 开局走右上方躲过第1波敌人,其后简单,有些地方可以爬在地上过去不会被发现,不难
愤怒的?鸟:星战版——图?全三星 昨天,R o v i o再次为我们奉献上了《愤怒的?鸟》最新版《星球?战》,操作?法还是?样,但各种鸟类有了新的功能,披上了星际?战的华丽外?,红鸟会挥舞光剑、黄鸟能发射激光等,当然,猪猪们也都进化了,不再是单独的被动挨打了,炮塔能主动向?鸟发动攻击了,?鸟们将?临着?次巨?的挑战,下?是?编为?家准备的全三星图?攻略。 如果你想在电脑上玩的话,请下载《愤怒的?鸟星球?战》P C版,已经破解和汉化。 《愤怒的?鸟星球?战》汉化破解版: h t t p://w w w.962.n e t/y o u x i/12414.h t m l 1-1: 我们按照图中?编所画的虚线来调整发射线,调得差不多后就发射攻击对?的建筑,接着建筑倒塌下来会把“?绿猪”给压扁了! 1-2: 根据图中虚线的位置,飞向星球猪下?的建筑?晶。然后,建筑全部垮塌下来,正在嚷嚷地太空猪也随着掉落并消失了。
1-3: 按照图中所?的发射线调得差不多后就发射攻击对?的建筑,接着建筑倒塌下来会把“?绿猪”给压扁了! 1-4:
根据图中所标?的虚线位置,我们点击愤怒的?鸟向左拉后,对准建筑下?的柱?,愤怒的?鸟飞了过去,将柱?撞破后,又碰到了炸药箱,轰的?声,太空猪们被消灭了。 1-5: 根据图中所?的虚线位置,我们将??太空?鸟往左拉对准前?的?晶块,在快要到达?晶块钱,再次点击?晶块(如上图所?),??太空?鸟会使?冲击波技能将这些零碎的建筑物全部冲击到?猪?上并将它们消灭。 %{p a g e-b r e a k|1-5|p a g e-b r e a k}%
三星换货攻略 前言:随着大家对东亚换货的熟悉,相信大家都对四星书和二星书的重要性很清楚了,而一星书更适合在东南亚就近根据状态刷食品(番薯)和调味料(蜂蜜、砂糖、罗望子),只有西洋書要在欧洲刷货过来,大家应该也都很熟悉了。而大家都比较忽略三星书的作用,这从三星书的市场价格也能看出来。这里就把三星书的进货和换货路线详细的列举出来供大家参考。 一、欧洲进货篇 1、阿姆刷蜻蜓球or都柏林刷玻璃工艺品,可以拉去云台山、汉阳、浦项、长崎等港 口,这个是大家三星书最常规的用法,这里不多说了。 2、加来刷葛布:这个要求有个法国刷货号。葛布可以拉去杭州,无状态可以5船满比; 也可以拉去堺和江户,这三个城市通常纺织品库存都很低。 3、热那亚双刷古代美术品和大理石像。为什么会提到这个,因为这个是美术品,大家 都知道美术品库存通常都在低位,古美可以在朝鲜和台湾,大理石像可以通吃华南。 4、欧洲进货到此为止,没有列出去贝鲁特刷马刀是因为东地里面太远了,到热那亚相 信已经是一些人的极限了。 二、东南亚进货篇 东南亚可以用三星?是的,没错。这一部分才是本文的重点,熟悉这一部分将对你驰骋南蛮贸易如虎添翼。 有状态时:1、安文炮弹对云台山、杭州、泉州、堺水灾。改版前炮弹是中枢才有,现在不用中枢就可以买到了,投资100万就能看到。经实测在云台山可以4船满比古墨。 2、占碑短剑、马斯喀特弯刀对朝鲜水灾。弯刀基础兑换率是短剑的一倍,如果在印度遇到朝鲜水灾时,可以考虑过去刷。 3、嘉定生丝对各地造船所需,注意去除纤维高库存城市。大部分人造船所需都选择工业制品,而很少选择纤维。经实测生丝可以达到满比兑换。 4、占碑石像对各地经济繁荣。除堺美术品自动高位,其它城市经济繁荣时如果美术品很低,未避免撞车都可以将石像作为一个备选项,基础兑换率在50/20左右。 无状态时:1、泗水金工对朝鲜。基础兑换率为50/30,汉阳工艺品低位时,可以一船满比,其它4船不会低于4:1兑换率。 2、洛布里丝绸布料对杭州。基础兑换率在50/22左右,因为杭州纺织品常年低位,经实测可以一船3:1,其它几船不低于5:1。
三星品牌重塑全攻略(doc22) 【案例提示】 在电子行业进展中,提及在近几年品牌价值提升最快的公司,无疑人们想到的是韩国三星!这家公司成立于1969年,早期业务要紧以生产廉价产品为主,在西方人心目中三星曾经是代表着“低价位,低质量、仿制品”的二三流公司。1990年代,三星施行品牌重塑策略,全力打造三星一流品牌,到现在的三星代表“时尚、高档、技术领先,e化”的全球领导性的品牌公司的蜕变,仅用了36年的时刻! 时至今日,作为主品牌的三星,其要紧业务范畴包括半导体、数字媒体、通讯网络及数字应用业务,同时在各个领域上三星品牌都有其鲜亮的产品个性,在高端市场上立足,引领时尚潮流。从1990年代的重新塑造三星品牌形象到推出一流产品品牌,从一个经历亚洲金融危机,负债1 70多亿美元,处在破产边缘的公司,到2000年,品牌价值为52亿美元,居世界43位;直至2005年10月27日出版的美国《商业周刊》刊登的品牌价值排行榜上,三星的品牌价值以149亿美元位居“2005年全球100个最有价值品牌”第20位,同时首次超过了它多年的宿敌——索尼(108亿美元,第28位)。三星在电子市场上走出了一条全新的重振品牌之路,并借此跻身全球闻名企业。三星的品牌价值在过去6年增加了186%(如下图所示),制造了增速最快的纪录。 三星是如何摆脱逆境,迅速提升其品牌价值的?它成功的要领在哪里?我们又能从中得到哪些启发呢?针对这些咨询题本文通过对三星的详细研究,首次从品牌价值的角度,揭示了其品牌价值迅速提升的全过程。 攻略一:确定重塑品牌的战略部署 背景:打造强势品牌前的三星状况 在上世纪八十年代末至九十年代初,三星制造的微波炉堆积成山,不得不打折处理,结果降低了自己的品牌声誉,在美国消费者心目中留下二流甚至三流的品牌印象。从而使得三星代表着“低价位,低质量、仿制品”。
开心消消乐全关卡三星通关攻略大全无论走带繁华的都市,还是拥挤的小巷,都可以看到经典三消手游开心消消乐的影子,足以证明这款游戏在玩家们心目中的地位。开心消消乐一共为玩家们提供了数百个不同难度的关卡,想要获得三星甚至是四星的好成绩真是难上加难,不过有了小编精心整理的开心消消乐攻略大全,可以帮你打通关。 关于开心消消乐游戏中的四星攻略,小编想告诉大家的是,四星关卡都是隐藏关,难度摆在那里,只有10关,分别是第7关、第20关、第25关、第45关、第108关、第112关、第118关、第119关、第144关、第227关。剩下的关卡最好的成绩都是三星哦。 四星关卡目标分数 第7关50000分 第20关10000分 第25关12000分 第28关25000分 第35关40000分 第38关50000分 第40关170000分 第43关110000分 第45关200000分 第58关225000分 第98关200000分 第108关105000分 第112关96000分 第118关71000分 第119关50000分 第144关170000分 第163关130000分 第198关380000分
第232关380000分 第227关190000分开心消消乐攻略第1关: 清除足够数量目标青蛙,还是非常简简单的,初次入门关卡的难度还是非常低的,游戏的基本入门时,也有简单的系统教程,只要跟着教程一步一步来,即可完成目标任务,就难度来说还是非常的简单的,只要清除10个青蛙就可以过关了,但是对于初次接触游戏的新手玩家来说还是有点困难的。 首先明确本关卡的任务:清除10个青蛙,那么小伙伴们要注意啦,我们的首要目的就是清除青蛙,那么就必须要围着青蛙来消除~ 方法很简单:只需要移动相邻的两个动物,组合成3只同等属性的小动物就可完成消除~ 开心消消乐攻略第2关: 首先~进入每个关卡需要消耗5点的精力。相信这点小伙伴们都了解,如果精力用完了,可是不能开始的哦,所以要珍惜
玩具塔防攻略三星通关心得分享讲解 玩具塔防攻略三星通关心得分享讲解,在游戏中想要获得三星的哈,那么玩家们的生命是一点都不能够丢失了,所以整体的难度还是有的,玩具塔防攻略三星的要求对于玩家们来说,往往需要玩家们打上好几次才能够属性改关卡的一些特点,然后才能够做到一些针对性的部署。 道具的使用: 1、修理工:防御塔将免费自动的修复,持续的时间为60秒; 2、弹幕攻击:强大的弹幕攻击将摧毁基地附近所有的敌人,将在12个关卡之后可用; 3、复兴:一次使用可让毁掉的防御塔复活,将在7个关卡之后可用; 4、压碎:用手指压碎敌人,持续时间60秒,将在17个关卡之后可用。 这些道具可用在关键的时候,帮助我们度过难关,但是价格也比较的贵,玩家们在使用的 时候,需要酢情考虑,否则会影响到后面的防御塔的升级。 防御塔升级: 游戏中的防御塔一共有着四种类型,分别为步枪兵(20$)、火焰喷射兵(100$)、加农炮(500$)和防空炮(120$),其中的防空炮只能够用来对付天空中的飞机,对于地面的部队,则不造成伤害,而加农炮的攻击速度比较的慢,又来对付装甲车有着奇效,而步枪兵和火焰喷射兵则是主要用来对 于步兵和轻甲车的,其中火焰喷射兵属于AOE伤害。 阵型布置: 每一个关卡都有着一个特定的地形,玩家们利用好地形上面的优势,再来布防防御塔,可以让我们的火力更加的击中,另外,玩家们最好将防御塔分开来不妨,这样就不会造成子弹的流逝, 从而浪费我们的火力。另外对于每一个关卡中的敌兵的种类,玩家们需要了解清楚,好做到提前准备好我们的防御塔,做到有针对的进行防御。 为你提供最新最全的 资讯及 ,想查找更多关于:玩具塔防攻略三星通关心得分享讲解的内容,你可使用站内搜索功能哦 ! 除了上面的这些小技巧之外,玩家们在每一个关卡结束之后,玩家们可以再我们布防的防御塔中选取收回,这样我们在下次布防的时候,不仅更加的便宜,而且升级不需要经验。 百度攻略&游戏多提供 1