Comparison of different genotype encodings for simulated three-dimensional agents
Comparison of Di?erent Genotype Encodings
for Simulated3D Agents
Maciej Komosinski Adam Rotaru-Varga
Draft,September2001
Institute of Computing Science
Poznan University of Technology
Piotrowo3A,60-965Poznan,Poland
maciej.komosinski@cs.put.poznan.pl
Abstract
This paper analyzes the e?ect of di?erent genetic encodings used for evolving3D agents with phys-ical morphologies.The complex phenotypes used in such systems often require nontrivial encodings.
Di?erent encodings used in Framsticks—a system for evolving3D agents—are presented.These
include a low-level direct mapping and two higher-level encodings:a recurrent and a developmental
one.Quantitative results are presented from three simple optimization tasks(active height,passive
height,and locomotion speed).The low-level encoding produced solutions of lower?tness than the
two higher-level encodings under similar conditions.Results from recurrent and developmental encod-
ings had similar?tness values but displayed qualitative di?erences.Desirable advantages and some
drawbacks of more complex encodings are established.
Keywords
virtual creature evolution,morphological evolution,3D agents,genotype encoding,developmental encoding 1Introduction
A survey of the?eld[21]indicates that there are a number of recent studies of the evolution of simulated creatures equipped with realistic physical behavior[7,16,18].Most of these works can be traced back to the in?uential work of Karl Sims[20].When comparing such systems with other evolutionary systems,we can note that the use of a physical simulation layer implements a complex genotype–?tness relationship. Physical interactions between body parts,the coupling between control and physical body,and interac-tions during body development can all add a level of indirection between the genotype and?tness.The complexity of the genotype–?tness relationship o?ers a potential for rich evolutionary dynamics.
The most important element of the genotype-to-?tness relationship is the genotype-to-phenotype map-ping,or genotype encoding.There is no obvious simple way to encode a complex phenotype—which consists of a variable-size,structured body and a matching control system—into a simpler genotype. Moreover,an evolutionary algorithm can perform poorly when using a certain genotype encoding,and better when using others,for reasons not yet immediately obvious.The employed genotype encoding
can have a signi?cant e?ect on the performance of the evolution.This fact has been recently recognized by researchers who directed e?orts into developing more sophisticated encodings[2,3,10].The afore-mentioned evolutionary systems lack a common base for experimentation1,they use di?erent physical engines,evolutionary algorithms,and various approaches for genotype encodings.Such di?erences render a comparison aimed solely at the e?ect of genetic encodings di?cult.
In the present work we use a single system,Framsticks,as the context of our analysis of various genotype encodings.Framsticks is a realistic,three-dimensional simulation of agents and their interactions [12,16,15].We present the three encodings currently implemented in the system,which include a simple low-level encoding and two higher-level ones:a direct recurrent and an indirect developmental.The low-level encoding is the simplest,and is considered to be a special case,while the other two are more complex,having been designed to be more evolvable.We compare the performance of the encodings in there optimization tasks(passive and active height,velocity),in experiments which di?er only in the encoding used.
The organization of the article is as follows:Section2.contains a general discussion on the e?ect of di?erent encodings.Section3.and Section4.provide an overview of the Framsticks system and its three di?erent encodings,followed by experimental results in Section5.Conclusions and direction for further work are included in Section6.
2The Role of the Genetic Encoding
Most evolutionary simulation systems distinguish between the concepts of genotype and phenotype,and employ a mapping between the two(this mapping is the trivial identity mapping in simple GA cases). As an evolutionary system allows for a more complex phenotype space,a more complex genotype-to-phenotype mapping(or encoding)is called for to allow genotypes to concisely describe complex pheno-types.Given a phenotype space,an encoding does not automatically follow:it is possible to construct di?errent genotype spaces which map into the phenotype space,and even for one particular genotype space,it is possible to devise numerous mappings from it to the phenotype space.Does the selection of a particular encoding have a signi?cant e?ect on the outcome of the evolutionary search?While it is hard to devise an‘ideal’encoding,it is certain that some encodings perform better than others.The issue of genetic encodings has been?rst addressed in the context of GAs,with the binary vs.Gray coding being one speci?c example[17].These studies established that the genetic encoding can have a noticeable e?ect on the evolutionary search.
The task of evolution of physical agents is di?cult for several reasons.Single points in the phenotype space are complete creatures,with a structured morphology and control(each creature has a‘body’and‘brain’).There is a variable amount of information required to describe such an organism.The dimensionality of the search space is not?xed,and the space does not lend itself to a straightforward neighborhood de?nition.Some features of the phenotypes change continuously(e.g.,length of a body part),while others change discretely(e.g.,number of body parts).Furthermore,the evaluation of a single phenotype involves a lengthy physical simulation,which can amplify small changes in the phenotype and lead to large changes in?tness;the imposed?tness landscape is multi-peaked;and evaluations contain non-deterministic components.In order to deal with such a large and complex search space,adequate genetic encodings are called for.The large number of CPU cycles required for?tness evaluations poses an additional technical di?culty.
Let us consider the implications of a chosen genetic encoding for a phenotype space(which we consider as given,determined by the chosen simulation rules).Typically an encoding maps only to a subset of the phenotype space.Valid phenotypes exist which cannot be expressed by the encoding.An example in the Framsticks system is morphologies containing cycles:although they are valid phenotypes,the default encoding(recur)2cannot express them.The existence of phenotypes which are impossible to encode
means that entire portions of the search space are sealed o?from the search.This,nonetheless,might bene?t the search,if the pruned space is smoother or denser in high-?tness points than the entire space.
More importantly,the encoding de?nes the topology of the phenotype space.Genetic operators(which are encoding-speci?c)determine which genotypes—and the corresponding phenotypes—are neighbors. The topology is encoding-speci?c:suppose phenotypes F1and F2are encoded by neighboring genotypes G1and G2under encoding S,and encoded byΓ1andΓ2under encodingΣ.Even if G1and G2are neighboring(there is a mutation in S which turns G1into G2),Γ1andΓ2can be distant in the genotype space inΣ.For example,the recur genotypes‘XXXX’and‘XLXXX’are separated by one point mutation (insert‘L’),but the corresponding simul genotypes are separated by three mutations(one for each of the last three sticks).An encoding—together with its associated genetic operators—determines which phenotypes are neighboring,and also in?uences which phenotypes have a higher probability of being visited.
The phenotype space topology imposed by the encoding determines which parts of the space are easily accessible by the search.Although the evolutionary search operates on phenotypes,the genotype encoding indirectly in?uences the outcome of the search.The bias imposed by an encoding becomes evident in the case of a random search:two random searches with di?erent genotype encoding yield di?erent results,despite the identical phenotype space.A phenotype space has an inherent topology: phenotypes of similar?tness are‘close’to each other.The genotype encoding imposes another topology, by de?ning which phenotypes are close genetically(measured as the number of mutations separating the corresponding genotypes).Under a good encoding,these two topologies are more highly correlated [9].Unfortunately,this correlation is impossible to directly measure,because the vast complexity of the phenotype space and its?tness landscape.Our judgments of encodings are thus subjective,based on various theoretical and experimental observations.
Di?erent encodings have di?erent characteristics,and some are better suited for some types of evolu-tionary search.It is hard to objectively establish these qualities of the encodings,and it is doubtful that a single best one exists.This perspective was our motivation to expand the Framsticks system to support several encodings at the same time.
3The Framsticks World
The following sections contain an overview of the Framsticks system and its capabilities.The Framsticks system simulates a three-dimensional world populated by agents.Agents are composed of an articulated morphology(‘body’)and attached control system(‘brain’),which are described in turn.For a more detailed presentation,consult[1,12,16,15].
3.1Body
The bodies of the agents are composed of a set of interconnected simple elements called sticks;a stick consists of two material endpoints connected through a?exible rod.Sticks have various physical and biological properties(mass,stamina,assimilation,etc.).Articulations exist between sticks where they share an endpoint;the articulations are unrestricted in all three degrees of freedom(bending in two planes plus twisting).
A wide range of physical interactions between sticks are simulated:static and dynamic friction, damping,action and reaction forces,gravity,buoyancy(uplift pressure under water),and energy losses from deformations;some of these forces are shown in Figure1.The?nite element method is used for step-by-step simulation.Collisions are simulated between stick of di?erent agents,but not between the sticks of a single agent(for e?ciency reasons).
gravity
ground reaction friction
damping
elastic reaction
of joints Figure 1:Various kinds of forces considered in
the physical simulator of the Framsticks system.
3.2Brain
A Framsticks agent is also equipped with a control system,which is implemented by a network of arti?cial neurons .Some neurons are specialized into sensors and e?ectors ,for interfacing with the mechanical body.
Generic neurons are simple processing units,similar to the ones used in standard arti?cial neural networks.Every neuron has a variable number of weighted connections from other neurons,and several parameters which in?uence its function.A neuron output value is updated periodically.The updating rule is based on the standard sigmoid function:
o =2
1+e ?s t ·β?1(2)
s t =s t ?1+v t (3)
v t =v t ?1·λ+μ·(i t ?s t ?1)
(4)The additional tunable parameters are λand μ.For λ=0and μ=1we get (1).The parameters β,λand μcan all be under genetic control.
The rules for computing neuron activation values are deterministic,however,initial activations are set stochastically to small values.The reason for employing randomness is to discredit neural networks that rely extensively on speci?c initial patterns,and encourage those which rely more on the information present in the environment.The latter ones tend to produce more robust solutions.
The special neurons include e?ectors (muscles)and various sensors.E?ectors are muscles that can exert modulated forces at articulations they are attached to.Muscles exists in two varieties:bending and rotating .
(a)(b)(c)(d)
Figure2:Visualisation of e?ectors:(a)bending and(b)rotating,and
receptors:(c)equilibrium and(d)touch.E?ectors are drawn on the
?rst of the two endpoints they in?uence.
Sensors,also attached to sticks,include orientation and touch sensors.3An orientation sensor(denoted ‘G’,also called a gyroscope)measures the orientation of the stick relative to the horizontal plane.A touch sensor(denoted‘T’)reacts to a contact force at the end of the stick,and can detect contact with the ground(or lack of it).Both orientation and touch sensors are useful for building controls for locomotion and other behaviors.See Figure2.for an illustration of the Framsticks e?ectors and sensors.Additionally, see the Appendix for more detail on the capabilities of the system.
4Genetic Encodings in Framsticks
4.1Support for multiple encodings
There are multiple encodings supported by the Framsticks system,each with its own representation and operators.The system manipulates and transforms genotype strings in various representations,and ultimately decodes them into the internal representation used by the simulator.
Any creature can be completely described using a low-level representation,by listing all of its com-ponents and attributes.This representation can be treated as a special genotype encoding—special because it is a direct one-to-one mapping—which we call simul.Other higher-level encodings con-vert their representation into the corresponding simul version(possibly through another intermediary representation),as illustrated in Figure3.The reverse mapping of higher-level encodings is di?cult to compute,which is also true for biological phenotype encodings.As a consequence,in the general case it is not possible to convert a lower-level representation into a higher-level one(or a higher-level one into another higher-level one).Nonetheless,an approximate transformation is possible from devel genotypes to recur genotypes,but not the other way around.
Each encoding has its associated genetic operators(mutation,crossover,and optional repair),and a decoding procedure which translates a genotype into a simul(or another‘lower’)representation.A new encoding can be added relatively easily,by implementing these components,without the need to work with internal representations.The Framsticks system is accompanied by the Genotype Development Kit to simplify this process[15].
In this article we describe three encodings:the direct low-level,direct recurrent,and indirect develop-mental.There are other encodings which are currently under development(similarity-based,metabolism-based,etc.).The direct low-level encoding(denoted simul,from‘simulator’,elsewhere also referred to as‘f0’)is a universal encoding which directly describes any valid phenotype.The direct recurrent en-coding(recur from‘recurrent’,‘f1’),the original encoding in Framsticks,is signi?cantly more compact than simul,but still preserves a one-to-one mapping between genotypic and phenotypic parts.It uses higher-level rules to achieve compactness and transparency.The indirect developmental encoding(devel from‘developmental’,‘f4’)is similar to recur,but describes the process of creation of a creature,rather
Figure3:The architecture for multiple encodings.Solid
arrows indicate decoding of one representation into an-
other.Dashed arrow indicates an approximate transforma-
tion(with potential loss of information).
than its?nal form.Consequently,it supports some features that are results of interactions during the developmental process:most importantly,modularity.
4.2Direct low-level
The direct low-level,or simul,encoding describes agents exactly as they are represented in the simulator. This encoding is more of a direct representation than a proper encoding,but it is possible to use it as such. It does not use any higher-level features to make the genotype more compact or?exible,and because of this,it is expected that this encoding is not very well suited for evolution.Its useful characteristics are that it has a minimal decoding cost and that it is universal:every possible agent can be described using this encoding.These properties make it possible to use the simul encoding as an intermediary representation during the translation from other higher-level encodings.
A simul genotype consists of a list of descriptions of all the objects the agent is composed of:parts, joints,neurons,and neuron items(connections,sensors,e?ectors).Every description speci?es all the attributes of the object explicitly(except those which are equal to their default value).A sample genotype is provided in Figure4.All objects are implicitly numbered by their position in the list,and these absolute order numbers are used for later reference(e.g.,each neuron has a reference to the part it is attached to).The absolute reference numbers are global properties,and are a?ected by a change in the number or order of the objects.Generally the simul encoding does not impose any restriction on the phenotypes4, and it even allows morphologies with cycles(such as the one illustrated in Figure5.a).This is not true of the other two encodings.Further details of this format can be found in[13].
In order to use simul as a true encoding,both genetic operators are implemented.Point mutation of a genotype is straightforward:it either changes one attribute of one object,or(less frequently)removes an existing object or adds a new one.In either case,mutation a?ects exactly one element of the agent. The simul encoding does not o?er a straightforward method for crossover.Thus we based crossover on phenotypic geometry:both morphologies are cut in two parts using a plane randomly positioned in space,and the two halves from each of the agents are grafted together.Neurons follow the part they are attached to,and broken links are reconnected to recover lost functionality.Such an example is shown in
p:D=0
p:1,m=2,vol=2,D=0
p:2,m=3,vol=3,D=3
p:2.50017,-0.000170005,-0.865927,D=5
p:2.50017,0.000170005,0.865927,3,vol=3,D=7
p:3.50017,0.000170005,0.865927,D=9
p:2.00051,0.000340067,1.73215,D=11
j:0,1,dx=1,D=0
j:1,2,1.5706,dx=1,D=3
j:2,3,rz=-1.047,1,D=5
j:2,4,rz=1.047,1,D=7
j:4,5,rz=-1.047,1,D=9
j:4,6,rz=1.047,1,D=11
Figure4:A sample simul genotype.This genotype corre-
sponds to the phenotype shown in Figure7.b—recur geno-
type‘XRRX(X,X(X,X))’.Lines starting with a‘p:’represent
material endpoints,while lines starting with a‘j:’represent
rods joining two endpoints.The?rst two numbers after‘j:’
are the references for the two endpoints.
Figure5.In a special case,when two identical parents are crossed over,the resulting o?spring is identical to the parents.
4.3Direct recurrent
The direct recurrent,or recur,encoding was the?rst one employed in the Framsticks system.This higher-level encoding was designed so that genotypes are compact and robust in face of genetic operators. Being easily understood and manipulated by humans was also a consideration.
The details of the recur encoding had been covered in[12,14,16],so only a brief overview is given here.In a recur genotype,the component sticks of a morphology are described using a string as follows: each stick is represented by a letter‘X’,and two consecutive‘X’s represent two sticks joined together. If there is a stick joined to several other sticks,they are represented using the structure‘X(X...,X ...,X...)’.This is su?cient to represent any cycle-free stick topology,see Figure6.for some examples.
Meaning
R
Skew of branching plane
C
Length
F
Muscle strength
Table1:Modi?er symbols in recur.‘RRR’rotates the current branching plane by3·45?=135?;‘rr’by ?90?,etc.
Various attributes of sticks are speci?ed using modi?er letters.Modi?ers change the value of a certain
(a)(b)(c)
Figure5:A simul crossover example.(a)The two parent structures;shown with their cutting planes.(b)The separated parents;dashed lines are joints which are broken,and the two halves shaded in black are used in the child.(c)The child structure;dotted lines are newly created joints.
‘X’‘XX(X,X)’
(a)(b)
‘XXX(XX,X(X,X))’‘XXX(,,,XX(X(X,X),X(X,X,,)))’
(c)(d)
Figure6:Examples of genotypes describing bodies of increasing com-
plexity.(a)Single stick,(b)two joined sticks branching into single sticks,
(c)recurrent branching,(d)recurrent branching with some branches
missing.
property in a relative manner,starting with an implicit default value.Lowercase letters denote modi?ers which decrease,and capital letters denote modi?ers which increase the value of an attribute by a?xed factor.Repeated letters achieve a compound e?ect,for example,‘lX’represents a short stick,while ‘LLLX’a very long one.Furthermore,modi?ers are‘fuzzy’and do not only a?ect the?rst following stick, but later ones as well,with a decreasing weight.For example,the lengths of the sticks in the genotype ‘XLLXXX’are1.00,2.00,1.51,and1.255respectively.There are modi?ers for basic attributes(present also in the simul encoding),and some recur-speci?c ones,such as the curving angle between two sticks. Modi?ers and their meanings are listed in Table1.Related examples of genotype-phenotype pairs are presented in Figure7.
‘CXlXlXlX’‘XRRX(X,X(X,X))’‘CCXXX(XXX,XXX)’
(a)(b)(c)
Figure7:Examples of recur modi?ers.(a)Shortening and curvedness,
(b)rotation of branching plane by90?,(c)tendency for curving.
Neurons are represented by the symbols‘[...]’,inserted after the stick they are attached to. There are various parameters speci?ed for a neuron:type(‘|’for bending muscle,‘@’for rotating muscle, etc.),a comma-separated list of connections in the format‘:
The recur encoding has specialized genetic operators.Mutation either adds a new modi?er,a new stick or neuron,or deletes an existing modi?er,stick,or neuron,or changes the parameters of an existing neuron.Crossover operates on the genotype strings in a straightforward way:it swaps randomly isolated substrings among two strings.However,cut points are restricted to logical positions(i.e.,the multiple characters describing a single neuron are never separated).
The recur encoding has several properties resulting from its design features.Important properties are:
–Linkage between body parts is done implicitly,rather than using explicit references to parts(eg.,‘XX’implicitly creates a link between the two sticks).
–Linkage between neurons is done using relative rather than absolute numbering.Relative numbering is susceptible to disruptions only if the added(removed)neuron is between the two ends of the link,but resistant otherwise.The latter cases are more frequent,and links speci?ed by absolute numbering would be broken in such cases.
–Attribute changes propagate along the body structure.
‘X[0:5]’‘X[@1:3]X[G:-2]’‘XX[@0:0][|1:1,-1:2]X[-2:3]’
(a)(b)(c)
Figure8:Examples of describing linkage between body and control.(a)
Single stick and neuron,not connected functionally,(b)two sticks:one
with a rotating muscle,the other with a G receptor,(c)example of
a more complex NN and two kinds of neurons for the same joint.The
body shape,the neural network,and the genotype is shown in each case.
Note that the neural networks are shown with inputs arranged to the
left and outputs to the right,which results in a di?erent ordering of the
neurons than in the genotypes.
–Implicit default values and automatic constraints are used extensively(e.g.,if there is a link from
a sensor,there must be a sensor).
–Control elements(sensors,e?ectors)are described near the elements controlled.
Now we can list the characteristics of the recur encoding:
1.Non disruptive.Small changes to the genotype generally cause small changes to the phenotype.
The changes can propagate to multiple sites,but in a continuous and structured manner.
2.Crossover-friendly.Substrings retain at least part of their meaning when isolated and inserted into
another context.Disrupted references are?xed by automatic constraints and implicit rules.
3.Human-friendly.The relationship between a genotype and its phenotype is relatively easy to un-
derstand by a human user,so genotypes can be be analyzed and modi?ed by hand.
https://www.wendangku.net/doc/0016597411.html,plete.The genotype speci?es every aspect of the phenotype completely(albeit not all directly).
5.Minimal redundancy.There are no genotype parts with no in?uence over the genotype.
In summary,the direct recurrent encoding uses higher-level constraints,and is potentially more e?ec-tive in an evolutionary search than the direct low-level encoding.However,this conjecture needs to be tested empirically.
4.4Indirect developmental
The indirect developmental encoding describes an agent by specifying its developmental process rather than its?nal form.A preliminary description of this encoding has been included in[14].The encoding models a set of interacting phenotypic parts which execute actions speci?ed by the genotype.This encoding is inspired by the cellular developmental encoding of Gruau[7],but is extended to include the body as well.Developmental encodings have been applied to evolution of neural networks[11,23,8,4,19], and more recently to both neural networks and morphologies[5,6,2,3,10],and have been found to be
superior to direct encodings,by producing more structured and modular phenotypes.Extra complexity and unnecessarily large phenotypes are reported as disadvantages.
Using the devel encoding,creatures are built by passing through a developmental phase.A developing creature consists of a set of interconnected cells,which can be undi?erentiated or di?erentiated(sticks or neurons).Cells execute genetic codes which alter their properties,or create new cells through division. After a division the newly created cells execute di?erent codes(they di?erentiate).At this point the genetic codes fork,which is why the entire genotype is organized as a tree.Development of a creature starts out as a single undi?erentiated cell.As new cells are created,they follow their instruction in parallel.Undi?erentiated cells can mature into sticks(‘X’)or neurons(‘N’).Development halts when all cells mature[7,19].
Devel is similar to recur insofar as it uses similar genetic symbols(‘X’,‘[...]’,etc.).In recur the genotype is traversed,and the codes are interpreted by an external builder process,which creates the parts of the creature.In devel the codes are interpreted by the developing parts themselves.Recur codes consist of ones that generate new parts,and modi?ers that change properties of parts that are to be created.Devel also has analogous modi?ers,but these a?ect existing parts.
Figure9:A sample devel genotype represented as a tree.
The genotype is‘
recur genotype presented in Figure7.b.
A devel genetic code tree is represented as a string,generated by traversing the tree pre-order(for each node,?rst the node is described,then its subtrees).A division‘<’is followed by the codes executed by the parent cell,until the corresponding stop code‘>’,then by the codes executed by the daughter cell. If either of the codes is itself a tree,the same rule is applied recursively.This way the‘<’and‘>’codes also act like nested parentheses.An example is presented in Figure9:the genotype‘
The devel encoding supports repetition of the same code portion more than once.This is implemented by the repetition code‘#’,which has two subtrees(like‘<’):the?rst is the code which is repeated,and the second is the rest,executed after the repetitions.The‘#’code also speci?es the number of repetitions. The repeated subtree can contain an arbitrary number of codes,including divisions.In this case only daughter cells will continue the repetition,not both of them.
The repeated genotype portion can consist of a single node or a subtree of arbitrary size.If the repeated subtree includes only a modi?er,the e?ect of the modi?er will be enhanced,but no new parts will be created.If it includes a division(‘<’),multiple copies of the element will be created.The simplest case is the repetition of a single stick or neuron.The repeated portion can contain multiple division codes,and even nested repetition codes.For example,the genotype presented in Figure10.features a
repeated body segment consisting of three sticks.
Phenotype modularity produced through developmental repetitions can be contrasted with modularity produced through phylogenic gene duplication.During evolution a genotype portion encoding body substructures can be duplicated(especially by crossover).However,even if the resulting duplicated phenotypic structures are identical,they are no longer encoded by the same genetic codes,and thus will evolve independently.In the case of devel encoding,repeated identical body parts are encoded by shared genetic codes.This opens up the possibility of mutations which a?ect all copies simultaneously [19].Modularity is argued to be a useful property when evolving complex entities[7,2].However,the modularity in devel encoding is somewhat limited in that modules are identical,and cannot diverge or di?erentiate.
The devel encoding is accompanied by specialized genetic operators:mutation a?ects a single node in the genotype tree(it is similar to the recur mutation).Crossover isolates and swaps subtrees from the genotype trees,which is the standard method used in genetic programming.
The devel encoding is similar to the recur encoding in that it uses analogous genetic codes,and every devel genotype can be translated into an approximate recur genotype.Di?erences include support for modularity and the mechanism for propagating attribute values from one element to another.In devel, when a cell divides,the new cell inherits the attributes of the old cell.The e?ect is similar to how modi?er e?ects are propagated in recur.But the propagation-by-division is more?exible:for example,dividing neurons can duplicate their existing connections(see Figure11.for an example).This provides one way to compress information needed to describe a neural network.
Figure10:Example of the repetition operator in genotype
‘rr
Figure11:Example of multiplication of neural
links when a cell is divided.The devel genotype
is‘
ing recur is‘X[*:0][-1:10][-2:10][-3:10]’).
In conclusion,the devel encoding is similar to the recur encoding,but uses some features of a
developmental encoding.It supports segmentation and modular body layouts,features which are not supported by recur.However,devel is not a full-?edged developmental encoding,because its support for modularity is limited:repeated modules are always identical.(This is also true of the encoding used in[10],but not of the one used in[3].)
4.5Characteristics of the Three Encodings
In this section we attempt to summarize the characteristics of the three encodings,and chart the di?er-ences in Table2.We include the following characteristics in the comparison.Genotype complexity refers to the internal structural complexity of genotypes,and the degree of dependencies between genes5.The simul and recur encodings have medium complexity,while the devel encoding has somewhat higher complexity,due to the repetition structures.
Recur
Medium
Medium
High
Low
None
None
Low
5We use the word‘gene’loosely,referring to the smallest element of a genotype(a letter in the genotype string),or a set of adjacent elements(a substring of the genotype string).
The?rst two tasks required maximization of the average height of the agent,as measured by the geometric center of body parts.In order to limit evolution to static morphologies,in the?rst task we turned o?the simulation of neural networks.This limitation was removed for the second task,which opened up the possibility of movement to enhance?tness.The motion of a creature may increase the height on average,but morphology is still the dominant factor.The height maximization tasks are relatively simple compared to the full potential of the Framsticks system,but enables us to analyze results easily,just by looking at static images.
The third task was maximization of locomotion speed on land.This task required a coordinated set of body parts(limbs),and control structures(e?ectors,neurons,and usually sensors).Various results for locomotion using Framsticks have been reported previously in[16,14].
5.1System parameters
The evolutionary algorithm used was a steady-state GA,with the most important parameters listed in Table3.Genotypes were selected using tournament selection with size2.When a genotype was selected for reproduction,it was modi?ed in80%of the cases.If it was modi?ed,it was mutated(80% probability)or crossed over.If a genotype was not modi?ed,it was cloned,which served to lengthen the average lifetime of genotypes,to get?tness values sampled more than once.This was needed because of the nondeterministic nature of the simulation(random initialization of neuron states).The?tness of a genotype was de?ned as the average of the?tness values of the multiple individuals sharing the genotype.
These settings were a result of a tedious experimentation and adjustment process.For every combi-nation of genetic encoding and task we executed10?nal runs,for a total of3×3×10=90runs.
Value
Population size
20%
Crossing-over probability
64%
Initial distance from the ground
Setting Velocity
Performance measurement interval500
Simulation time after stabilization5000
Number of evaluations in a single run60,000
Table4:Parameters of?tness formula and the number of evaluations in a single evolutionary run.
Further settings are detailed in Table4.All values concerning time are in simulation steps,and values concerning distance are in simulation units.For comparison,the default stick length is1.0.Every evaluation was started with a stabilization period,during which no performance measurements were taken,and neurons were kept inactive.This was done in order to prevent the use of the potential energy resulting from the initial placement and position of the creatures.
5.2Results–Quantitative analysis
In a quantitative analysis the notion of best individual is important,but complicated by the fact that ?tness evaluations are non-deterministic.A?tness can depend on how many times a genotype had been evaluated.The more evaluations,the more stable and reliable is a?tness estimate.
Figure12:Best,average,and worst?tness values during an evolutionary run.
We reproduce a typical?tness pro?le from a velocity-oriented experiment.Figure12.plots various ?tness values against time.The middle line is the mean?tness of the population.The lines above (best(2))and below(worst(2))are the best and worst?tness values of genotypes evaluated at least two times.Lines best(1)and worst(1)are the best and worst?tness values of all genotypes(including those evaluated only once).As it can be seen,best(1)and worst(1)vary widely.Therefore we decided to de?ne the‘best’genotype as the genotype with highest mean?tness,evaluated at least two times(best(2)). Thus‘best’,‘highest’,etc.,are used with such meaning in the forthcoming discussion.
Charts in Figure13.summarize the?tness results for the three tasks and three genetic encodings. The bars show averages of best individuals taken from the10runs;standard deviations are also shown. In all three tasks,the worst was the simul encoding.Recur and devel were comparable.Statistically, the di?erence was important between devel and simul in task(a),and between recur and simul,and devel and simul in task(b).In task(c),no di?erences were statistically signi?cant6.
It may seem interesting that active height maximization did not yield better results than the passive one.One explanation is that static constructs alone were su?cient to produce solutions very close to physical limits.The?tness result values,taken together with attempts to manually construct agents (described in more detail below),indicate that the value of2.50for average height of the center is very hard to exceed.This is due to some physical limitations:the maximum stick length is2.00,and the high elasticity of sticks and joints makes them unable to bear large weights.The disadvantage of a moving design(e.g.,jumping)is instability,and apparently,in this regime the disadvantage was stronger than
0.51.01.52.02.5simul recur
devel
(a)(b)(c)
Figure 13:Best ?tness values found in the three tasks:(a)passive average
height,(b)active average height,(c)average velocity.Mean values are shown
with standard deviations.
the bene?t.Motion might be better,however,in increasing maximal height rather than average height (as in the case of a big leap followed by collapse).
5.3
Results –Qualitative analysis 5.3.1Height,passive agents
In this task three kinds of construction were typical,with variable intensity of branching:from antenna-like creatures through tree-and bush-like creatures (Figure 14).Typical simul solutions had a triangular base (90%of agents),which allowed for high stability and sti?ness.This shows the importance of cycles in the https://www.wendangku.net/doc/0016597411.html,ing cycles it is possible to convert some of the compressing forces into pulling forces,which are handled better by sticks.If cycles were not available,we would have expected even lower maximum ?tness values.
Using the simul encoding,one of the observed kinds of solutions was a number of sticks erected from a base.Such a structure allowed for a high position of the geometric center with high stability and sti?ness (Figure 14.d).With recur ,in 50%of the agents base points were joined not at ground level,but above (Figure 15.a,b,c).One of the exceptions was the bush-like agent with 4353parts (Figure 15.d).In the case of the devel encoding,the in?uence of modularity was observed in structures resembling a spiral,a chain,or a segmented backbone (Figure 16).
Although evolved creatures were stable during their normal evaluation period,in most cases they were knocked out of balance with minimal e?ort.Some minimal motion could usually be observed even in the case of passive agents,due to elastic forces and non-equilibrium initial conditions.In some cases agents ?ipped over spontaneously when simulated for times exceeding the length used during the evolution.This shows the extreme degree of adaptation to the given environment and peculiarities of the ?tness evaluation.
(a)(b)(c)(d)
Figure14:Representative best agents in passive height maximization task,simul encoding.
(a)(b)
(c)(d)
Figure15:Representative best agents in passive height maximization task,recur encoding.
(a)(b)(c)
(d)
(e)
Figure 16:Representative best agents in passive height maximization
task,devel encoding.
5.3.2Height,active agents
This task is similar to the passive height task,but with the possibility of using neural networks to generate movement.Many resulting structures were similar to ones from the previous task.In simul and recur, about40%of agents appeared to be moving purposefully;the rest were moving in a way that did not deteriorate their?tness or did not move at all(Figure17.a,b).A purposeful movement was usually stretching and straightening(Figure17.c),with an orientation sensor as the signal source.Among devel agents no purposeful movement was observed,but some interesting constructions emerged(Figure18).7
(c)
Figure17:Representative best agents in active height maximization
task.Recur encoding.
(a)(b)(c)(d)(e)
Figure18:Representative best agents in active height max-
imization task.Devel encoding.
5.3.3Velocity
In this task,evolved agents had small bodies(small weight),and used neural networks with e?ectors and receptors.This task is not so severely imitated by simple physical constraints as the height tasks,and there was a higher variety among the observed strategies.
Among agents evolved for velocity the most frequently encountered structure consisted of a few sticks, branched or bent(Figure20.b).The last stick was moved by a bending muscle,and used as a limb for pushing back.The branching on the other end stabilized the direction of the locomotion(if an agent
fell down after a jump,it turned over and got into the same orientation,due to the stabilizing limbs). Consecutive small jumps resulted in locomotion in one direction(Figure19.a).
Another popular solution was a construction with two pushing limbs,with a body perpendicular to the direction of the locomotion(Figure20.a).Usually only one limb had a muscle,and the other served as stabilization,moving passively,and helping sustain the direction of movement.
Apart from the common designs,many di?erent strategies could be observed,some illustrated here. The agent shown in Figure19.b.had three limbs(two pushing back,one pulling),the one in Figure19.c. used a triangular structure for pushing,and the one in Figure20.c.had a symmetrical body with two pushing limbs.
Since solutions had small bodies,there were no evident general di?erences between morphologies in the case of the three encodings.There was no room for segment repetitions and large scale modularity. Subjectively,the movement of recur and devel solutions appeared as somewhat biologically more plau-sible than simul movements.Neural networks,receptors,and e?ectors were very precisely tuned to the morphology,showing a tight coupling between morphology and control.
(a)
(b)
(c)
Figure19:Representative best agents in velocity maximization task.
Simul encoding.
5.4Comparison to human-designed agents
Designing agents by hand is a very complex process,in professional applications it requires planning and extensive knowledge about how the control system,receptors,and e?ectors work,as well as knowledge about the simulator.Designing neural networks for control by hand is especially di?cult and tedious.For this reason,human-built agents usually have lower?tness than agents produced by evolution.However, human creations are often interesting qualitatively.Human designs have such properties as explicit purpose,elegance,simplicity(minimum of means),and often symmetry and modularity.These features are opposed to evolutionary results,which are characterized by hidden purpose,complexity,implicit and very strong interdependencies between parts,as well as redundancy and randomness.
The di?cult process of designing neural networks can be circumvented by a hybrid solution:bodies can be hand-constructed,and control structures evolved for it.It is possible to turn o?evolution of body parts(using simul or recur).This approach can yield interesting creatures[1,12,14,15],often resembling creatures found in nature.
We tried to design agents by hand for the presented tasks,without much success.The locomotion task is simply too complex for a new good solution to be designed by hand in a reasonable amount of
螺纹通止规
螺纹通止规 定是:螺纹止规进入螺纹不能超过2.5圈,一般的要实际不得超过2圈,并且用得力度不能大,我们的经验是用拇指和食指轻轻夹持螺纹规以刚好能转动螺纹规的力度为准.力大了就相当于在使用丝锥或牙板了,那样规就用不了几次了. 螺纹通止规 螺纹通止规是适用于标准规定型号的灯头作为灯用附件电光源产品时候的设计和生产、检验的工具设备。 用途 一般用于检验螺纹灯头或灯座的尺寸是否符合标准要求，分别检验螺纹灯头的通规和止规尺寸或灯座的通规或止规尺寸。 工作原理 具体检验要求及介绍详见中国人民国国家标准：GB/T1483.1-2008或 IEC60061-3:2004标准规定容。 操作方法 具体检验要求及介绍详见中国人民国国家标准：GB/T1483.1-2008或 IEC60061-3:2004标准规定容。 通止规
通止规，是量规的一种。作为度量标准，用于大批量的检验产品。 通止规是量具的一种，在实际生产批量的产品若采取用计量量具（如游标卡尺，千分表等有刻度的量具）逐个测量很费事．我们知道合格的产品是有一个度量围的.在这个围的都合格，所以人们便采取通规和止规来测量． 通止规种类 （一）对统一英制螺纹，外螺纹有三种螺纹等级：1A、2A和3A级，螺纹有三种等级：1B、2B和3B级，全部都是间隙配合。等级数字越高，配合越紧。在英制螺纹中，偏差仅规定1A和2A级，3A级的偏差为零，而且1A和2A级的等级偏差是相等的等级数目越大公差越小，如图所示：1B 2B 3B 螺纹基本中径3A 外螺纹2A 1A 1、1A和1B级，非常松的公差等级，其适用于外螺纹的允差配合。 2、2A和2B级，是英制系列机械紧固件规定最通用的螺纹公差等级。 3、3A和3B级，旋合形成最紧的配合，适用于公差紧的紧固件，用于安全性的关键设计。 4、对外螺纹来说，1A和2A级有一个配合公差，3A级没有。1A级公差比2A级公差大50，比3A级大75，对螺纹来说，2B级公差比2A公差大30。1B级比2B级大50，比3B级大75。 （二）公制螺纹，外螺纹有三种螺纹等级：4h、6h和6g，螺纹有三种螺纹等级：5H、6 H、7H。（日标螺纹精度等级分为I、II、III三级，通常状况下为II级）在公制螺纹中，H 和h的基本偏差为零。G的基本偏差为正值，e、f和g的基本偏差为负值。如图所示：公差G H 螺纹偏差基本中径外螺纹f g h e 1、H是螺纹常用的公差带位置，一般不用作表面镀层，或用极薄的磷化层。G位置基本偏差用于特殊场合，如较厚的镀层，一般很少用。 2、g常用来镀6-9um的薄镀层，如产品图纸要6h的螺栓，其镀前螺纹采用6g的公差带。 3、螺纹配合最好组合成H/g、H/h或G/h，对于螺栓、螺母等精制紧固件螺纹，标准推荐采用6H/6g的配合。 （三）螺纹标记M10×1–5g 6g M10×1–6H 顶径公差代号中径和顶径公差代号（相同）中径公差代号。 通止规是两个量具分为通规和止规．举个例子：M6－7h的螺纹通止规一头为通规（T）如果能顺利旋进被测螺纹孔则为合格，反之不合格需返工（也就是孔小了）．然后用止规（Z）如果能顺利旋进被测螺纹孔2.5圈或以上则为不合格反之合格．且此时不合格的螺纹孔应报废，不能进行返工了．其中2.5圈为国家标准，若是出口件最多只能进1.5圈（国际标准）．总之通规过止规不过为合格，通规止规都不过或通规止规都过则为不合格。
感官动词和使役动词
感官动词和使役动词 默认分类2010-05-28 23:14:26 阅读46 评论0 字号：大中小订阅 使役动词，比如let make have就是3个比较重要的 have sb to do 没有这个用法的 只有have sb doing.听凭某人做某事 have sb do 让某人做某事 have sth done 让某事被完成（就是让别人做） 另外： 使役动词 1.使役动词是表示使、令、让、帮、叫等意义的不完全及物动词,主要有make(使,令), let(让), help(帮助), have(叫)等。 2.使役动词后接受词,再接原形不定词作受词补语。 He made me laugh. 他使我发笑。 I let him go. 我让他走开。 I helped him repair the car. 我帮他修理汽车。 Please have him come here. 请叫他到这里来。 3.使役动词还可以接过去分词作受词补语。 I have my hair cut every month. 我每个月理发。 4.使役动词的被动语态的受词补语用不定词,不用原形不定词。 (主)He made me laugh. 他使我笑了。 (被)I was made to laugh by him. 我被他逗笑了。 使役动词有以下用法： a. have somebody do sth让某人去做某事 ??i had him arrange for a car. b. have somebody doing sth.让某人持续做某事。 ??he had us laughing all through lunch. 注意：用于否定名时，表示“允许” i won't have you running around in the house. 我不允许你在家里到处乱跑。 ******** 小议“使役动词”的用法 1. have sb do 让某人干某事 e.g:What would you have me do? have sb/sth doing 让某人或某事处于某种状态，听任 e.g: I won't have women working in our company. The two cheats had the light burning all night long. have sth done 让别人干某事，遭受到 e.g:you 'd better have your teeth pulled out. He had his pocket picked. notes: "done"这个动作不是主语发出来的。 2.make sb do sth 让某人干某事 e.g:They made me repeat the story. What makes the grass grow?
高中美文摘抄600字
高中美文摘抄600字 本文是关于好词好句的，仅供参考，如果觉得很不错，欢迎点评和分享。 高中美文摘抄600字 1、瑟瑟的秋风，飘零的落叶，人有悲欢离合，月有阴晴圆缺，不变的在这萧瑟的季节里，内心突然忧伤起来。 2、树欲静而风不在，子欲孝而亲不在。所以，你懂了吗？ 3、原来，爱生命到灵魂深处，周围一切都是满满的爱，生意盎然。静静的我还要遗憾什么呢？虽然真正永恒的生命不存在，可是，如果你在对的时间，邂逅了花开的美丽；在对的时间，遇上了对的人；在对的时间，和另一个有着生命之初那份纯真的自己重逢。别时，仍有回忆，依旧怅然。笑人生，不要过得太慢，因为花会凋；谢也不要走得太快，因为那时的花还未开放。 4、“梦想”是我们对于美好事物的一种憧憬和渴望，在很多时候它可以成为一个人的信仰，并被渲染上浪漫的色彩。梦想它给予我们一个方向，一个目标，它能够使人伟大，人的伟大就是把梦想作为目标去执着的追求。 5、乱世也美丽，只要用心去寻找，也能够找到好的一面。所以，用客观的视角看世界，你会看到另一种风采，另一种美丽。 6、每一个人的人生都需要梦想。没有理想，人生就如荒漠，没有生气；没有理想，人生就如黑夜，没有光明；没有理想，人生就如
迷宫，没有方向！ 7、也许，有时候我们该静下心来拷问自己，这到底是怎么回事？ 9、坚强，其实是一种自然而然的生活状态。 10、尘世中没有哪一方净土能让人不受到伤害。想要的生活总在前方，总是差一段距离。 12、梦想是期望，是坚持！ 13、都说乱世出才子。婉约派词人李清照，爱国诗人杜甫，南唐后主李煜，因为国家的动荡，生活的坎坷，终写下了流传千古的名句。“寻寻觅觅，冷冷清清，凄凄惨惨戚戚。”这诉说了李清照孤凄悲凉的心声。“国破山河在，城春草木深。”这描绘出一幅都城野草茂盛的凄凉景象，这是杜甫眼中的“破碎山河。”“雕栏玉砌应犹在，只是朱颜改，问君能有几多愁，恰似一江春水向东流。”这是李后主的作为亡国之君，国破后的愁绪就像滚滚长江一般道不尽，说不完。 14、当每一天的坚强成为一种习惯，也许有一天突然撤去一种压力，我们连自己要做什么都搞不清楚，不是每个人都有足够的定力，一直坚持自己喜欢的事情。也许，坚强已成为大众化。 15、当我看到这样一段文字时，心酸了--爷爷拉着我说﹕“我很好，别担心，你自己出门在外注意身体，多吃点好吃的……”明明在饭桌上，我只看到一小盘咸菜孤独的摆着，眼前顿时模糊一片，那天是他的八十岁生日啊！祖父母们总是心疼自己子孙，看到子孙被挨骂，总是第一个出来搬父母小时候的事来招摇。祖父母们是我们一辈子最难报答的人，因为当我们有能力报答时，或许他们已离你而去。但是
美国历史与文化
《美国历史与文化》 结课论文 专业：化学工程与工艺 学号：041114116 姓名：杨乐
Columbo's influence on the American continent Columbo is a famous Spanish navigator, is a pioneer of the great geographical discovery. Columbus young is garden said believers, he so haunting had in Genoa made prison of Marco Polo, determined to be a navigator.1502 he crossed the Atlantic four times in 1492, discovered the American continent, he also became a famous navigator. On August 3, 1492, Columbus by the king of Spain dispatch, with credentials of Indian monarchs and emperors of China, led the three tons of Baishi of sailing, from Spain Palos Yang Fan of the Atlantic, straight towards to the West. After seventy days and nights of hard sailing, in the early morning of October 12, 1492 finally found the land. Columbo thought he had arrived in India. Later know that Columbus landed on a piece of land belonging to the now Balak America than the sea in the Bahamas, he was it named San Salvador. Columbo's discovery has changed the course of world history. It opens up a new era of development and colonization in the new world. At that time, the European population was expanding, with this discovery, the Europeans have settled in two new continents, there will be able to make a difference in the European economy and the resources of the mineral resources and raw materials. This discovery led to the destruction of the American Indian civilization. From a long-term point of view, there have been a number of new countries in the Western
螺纹通止规要求螺纹通规通
螺纹通止规要求螺纹通规通，止规止。 但是如果螺纹通规止，说明什么？ 螺纹止规通，又说明什么？ 我也来说两句查看全部回复 最新回复 ?wpc (2008-11-07 20:11:20) 在牙型正确的前提下螺纹通止规检测螺纹中径 ?lobont (2008-11-08 11:16:32) 对外螺纹而言，螺纹通规是做到中径上偏差，所以能通过就表示产品合格，通不过就表示螺纹做大了，要再修一刀； 螺纹止规做到中径下偏差，所以只能通过2～3牙，如果也通过，就表示外螺纹做小了，产品成为废品 ?qubin8512 (2008-11-18 15:36:05) 螺纹赛规与螺纹环规主要测量螺纹的中径。 ?datafield (2008-11-29 19:12:51) 检具不是万能的，只是方便而已。具体没什么的我有在哪本书上看过，是一本螺纹手册上的。 ?ZYC007 (2009-2-09 20:31:13) 在牙型正确的前提下螺纹通止规检测螺纹中径。 对外螺纹而言，但是如果螺纹通规止，说明螺纹中径大；螺纹止规通，又说明螺纹中径小。 ?WWCCJJ (2009-3-19 09:27:19) 检测的是螺纹的中径,螺纹检测规在检定时,也是检测其中径. ?tanjiren (2009-3-20 22:23:06) 螺纹通止规只能检测螺纹的作用中径,大径和底径等均无法准确测量出来. ?月夜(2009-4-01 21:47:13) 用来测量中径 ?丽萍(2009-4-02 10:11:41)
只能检测工件螺纹的中径 yg196733456 (2009-4-03 09:15:56)原来是测中径的知道了
感官动词的用法
感官动词 1.see, hear, listen to, watch, notice等词，后接宾语，再接省略to的动词不定式或ing形式。前者表全过程，后者表正在进行。句中有频率词时，以上的词也常跟动词原形。 注释:省略to的动词不定式--to do是动词不定式，省略了to,剩下do，其形式和动词原形是一样的，但说法不同。 see sb do sth 看到某人做了某事 see sb doing sth 看到某人在做某事 hear sb do sth 听到某人做了某事 hear sb doing sth 听到某人在做某事 以此类推... I heard someone knocking at the door when I fell asleep. (我入睡时有人正敲门,强调当时正在敲门) I heard someone knock at the door three times. (听到有人敲门的全过程) I often watch my classmates play volleyball after school. (此处有频率词often) （了解）若以上词用于被动语态，须将省略的to还原： see sb do sth----sb be seen to do sth hear sb do sth----sb be seen to do sth 以此类推... We saw him go into the restaurant. → He was seen to go into the restaurant. I hear the boy cry every day. → The boy is heard to cry every day. 2.感官动词look, sound, smell, taste, feel可当系动词，后接形容词。 He looks angry. His explanation sounds reasonable. The cakes smell nice.
关于励志美文摘抄600字
关于励志美文摘抄600字 篇一： 黝黑的天空裏，明星如棋子似的散佈在那裏。比較狂猛的大風，在高處嗚嗚的響。馬路上行人不多，但也不斷。汽車過處，或天風落下來，阿斯法兒脫的路上，時時轉起一陣黃沙。是穿著單衣覺得不熱的時候。馬路兩旁永夜不息的電燈，比前半夜減了光輝，各家店門已關上了。 兩人盡默默的在馬路上走。後面的一個穿著一套半舊的夏布洋服，前面的穿著不流行的白紡綢長衫。他們兩個原是朋友，穿洋服的是在訪一個同鄉的歸途，穿長衫的是從一個將赴美國的同志那裏回來，二人系在馬路上偶然遇著的。二人都是失業者。 你上哪里去? 走了一段，穿洋服的問穿長衫的說。 穿長衫的沒有回話，默默的走了一段，頭也不朝轉來，反問穿洋服的說： 你上哪里去? 穿洋服的也不回答，默默的盡沿了電車線路在那裏走。二人正走到一處電車停留處，後面一乘回車庫去的末次電車來了。穿長衫的立下來停了一停，等後面的穿洋服的。穿洋服的慢慢走到穿長衫的身邊的時候，停下的電車又開出去了。
你爲什麽不乘了這電車回去? 穿長衫的問穿洋服的說。穿洋服的不答，卻腳也不停慢慢的向前走了，穿長衫的就在後面跟著。 二人走到一處三叉路口了。穿洋服的立下來停了一停。穿長衫的走近了穿洋服的身邊，腳也不停下來，仍複慢慢的前進。穿洋服的一邊跟著，一邊問說： 你爲什麽不進這叉路回去? 二人默默的前去，他們的影子漸漸兒離三叉路口遠了下去，小了下去。過了一忽，他們的影子就完全被夜氣吞沒了。三叉路口，落了天風，轉起了一陣黃沙，比較狂猛的風，嗚嗚的在高處響著。一乘汽車來了，三叉路口又轉起了一陣黃沙。這是立秋的晚上。 篇二： 我生在江南一隅，在江南的温婉情怀里沉醉了二十多栽。江南周遭世事沧桑变迁，喧哗或耳语中流传的故事无法一一讲述得清，而我对江南的眷恋情怀却千年依旧! 伫立于江南的灵土，仰望冥冥天空，风烟俱净，澄澈得如一汪清水。漫漫岁月中流淌着江南水乡的清秀，江南古镇的恬静，江南雨巷的幽深，江南文杰的的灵韵 小桥，流水，人家，流溢在水墨江南里，看不明虚实，分不清究竟;水性的流淌中，滋润了多少文人豪杰的心胸。品读江南，凌波水韵，翰墨流芳。
美国历史与文化
浅谈美国历史 ——见证从蚂蚁到大象历程 引导语：中华文化源远流长，五千年的华夏文明留给我们太多的回忆。分久必合、合久必分；从繁荣昌盛到民族衰落；从压迫受辱到当家作主、从璀璨奇葩到复兴中华……可是，跨过大洋的彼岸，初出茅庐的美国却在近两百多年的历史跨度下完成了从蚂蚁到大象的历程，创造出美国独特的发展文化。今天的世界，“汤姆大叔”在全球“维护着世界和平”；好莱坞大片充斥着各大荧屏；NBA回荡在茶前饭后的娱乐中……两百多年来，美国历史一直都是民主制度的试验。早年被提出的问题如今持续被提出并获得解决；强大政府对抗弱小政府、个人权利对抗群体权利、自有资本主义对抗受到管理的商业与劳工以及参与世界对抗孤独主义。美国对于民主制度有很高的期待，而现实又是不如人意。然而国家经过妥协，已见成长与繁荣。在今天的发展过程中笔者认为有必要借鉴美国蚂蚁变大象的历程。 自从哥伦布发现新大陆之后，这片土地上开始了她不平凡的发展。十七世纪初，英国开始向北美殖民。最初的北美移民主要是一些失去土地的农民，生活艰难的工人以及受宗教迫害的清教徒。在殖民地时代，伴随着与北美洲原住民印第安人的长期战争，对当地印第安人的肆意屠杀，严重的劳力缺乏产生了像奴隶和契约奴隶这类的非自由劳力。万恶的黑奴贸易盛行起来。从1607年到1733年，英国殖民者先后在北美洲东岸建立了十三个殖民地。由于英国移民北美是为了追求自由和财富，如被迫害的清教徒和贫农。地方政府享受自治权。殖民地居民有比英人更广泛参与政治的机会和权利，培养了自治的意识和能力，所以他们相信社会契约中，政府是人民需要保护而得人民支持才组成的。在十八世纪中期,殖民地的经济，文化，政治相对成熟，殖民地议会仍信奉英王乔治三世，不过他们追求与英国国会同等的地位，并不想成为英国的次等公民，但是此时英法的七年战争结束，急于巩固领土，使向北美殖民地人民征租重税及英王乔治三世一改放任政策，主张高压手段。因此引发殖民地人民反抗，如“没有代表就不纳税”宣言、“波士顿惨案”、“不可容忍的法案”等。1775年4月在来克星顿和康科特打响“来克星顿的枪声”揭开美国独立战争的前奏。后来，这些殖民地便成为美国北美独立十三州最初的十三个州。 1774年, 来自12州的代表，聚集在费城, 召开所谓第一次大陆会议，希望能寻出一条合理的途径, 与英国和平解决问题，然而英王却坚持殖民地必须无条件臣服于英国国王, 并接受处分。 1775年，在麻州点燃战火, 5月，召开第二次大陆会议, 坚定了战争与独立的决心，并起草有名的独立宣言, 提出充分的理由来打这场仗，这也是最后致胜的要素. 1776年7月4日，宣告了美国的独立，1776年7月4日大陆会议在费城乔治·华盛顿发表了《独立宣言》。《独立宣言》开宗明义地阐明，一切人生而平等，具有追求幸福与自由的天赋权利；淋漓尽致地历数了英国殖民主义者在美洲大陆犯下的罪行；最后庄严宣告美利坚合众国脱离英国而独立。《独立宣言》是具有世界历史意义的伟大文献。完全脱离英国，目的是为‘图生存、求自由、谋幸福’，实现启蒙运动的理想。 1781年, 美军赢得了决定性的胜利， 1783年, 美英签订巴黎条约，结束了独立战争。这也充分展现出
NPT螺纹以及检测方法详解
N P T螺纹以及检测方法详 解 Prepared on 22 November 2020
一、目的：规范公司技术员，检验员，操作员对NPT螺纹的了解。 二、适用范围：适用于公司任何NPT螺纹类产品，参考资料为通用管螺 纹和国家标准GB/T12716-2011。 三、目录 1、NPT和NPTF介绍 2、螺纹技术参数参数讲解 3、NPT与NPTF加工工艺 4、NPT和NPTF的检测方法 四、内容： NPT和NPTF螺纹介绍 NPT 是 National (American) Pipe Thread 的缩写，属於美国标准的 60 度锥管 密封螺纹，用於北美地区，美国标准为13)通用管螺纹．国家标准可查阅 GB/T12716-2011。NPTF：美制干密封圆锥管螺。NPTF = National Pipe Thread Fine 称之为一般用途的锥管螺纹，这也是我们以前称之为的布氏锥螺纹。NPTF 螺纹称之为干密封式锥管螺纹，它连接密封的原理是在没有润滑剂或密封填 料情况下完全依靠螺纹自身形成密封，设计意图是使内、外螺纹牙的侧面、 牙顶和牙底同时接触，来达到密封的目的。它们两者的牙型角、斜度等指标 都是相同的，关键是牙顶和牙底的削平高度不一样，所以，量规的设计也是 不一样的。NPTF干密封管螺纹的牙形精度比NPT螺纹高，旋合时不用任何 填料，完全依靠螺纹自身形成密封，螺纹间无任何密封介质。干密封管螺纹 规定有较为严格的公差，属精密型螺纹，仅用在特殊场合。这种螺纹有较高 的强度和良好的密封性，在具有薄截面的脆硬材料上采用此螺纹可以减少断 裂现象。NPTF内、外螺纹牙顶与牙底间没有间隙，是过盈配合，而NPT螺 纹是过渡配合。NPTF螺纹主要用于高温高压对密封要求严格的场所。NPT
英语中感官动词的用法
英语中感官动词的用法 一、感官动词 1、感官动词(及物动词)有：see/notice/look at/watch/observe/listen to/hear/feel(Vt)/taste(Vt)/smell(Vt) 2、连缀动词(含感官不及物动词) be/get/become/feel/look/sound/smell/taste/keep/stay/seem/ appear/grow/turn/prove/remain/go/run 二、具体用法： 1、see, hear, smell, taste, feel,这五个动词均可作连系动词，后面接形容词作表语，说明主语所处的状态。其意思分别为"看/听/闻/尝/摸起来……"。除look之外，其它几个动词的主语往往是物，而不是人。 例如：These flowers smell very sweet.这些花闻起来很香。 The tomatoes feel very soft.这些西红柿摸起来很软。 2、这些动词后面也可接介词like短语，like后面常用名词。 例如：Her idea sounds like fun.她的主意听起来很有趣。 3、这五个感官动词也可作实义动词,除look(当"看起来……"讲时)只能作不及物动词外,其余四个既可作及物动词也可作不及物动词,此时作为实义动词讲时其主语一般为人。 例如:She smelt the meat.她闻了闻那块肉。 I felt in my pocket for cigarettes.我用手在口袋里摸香烟。 4、taste, smell作不及物动词时,可用于"t aste / smell + of +名词"结构,意为"有……味道/气味"。 例如:The air in the room smells of earth.房间里的空气有股泥土味。 5、它们(sound除外)可以直接作名词,与have或take构成短语。 例如：May I have a taste of the mooncakes?我可以尝一口这月饼吗？taste有品位、味道的意思。 例如：I don’t like the taste of the garlic.我不喜欢大蒜的味道。 She dresses in poor taste.她穿着没有品位。 look有外观，特色的意思，例：The place has a European look.此地具有欧洲特色。 feel有感觉，感受的意思，watch有手表，观察的意思。例：My watch is expensive.我的手表很贵。 6、其中look, sound, feel还能构成"look / sound / feel + as if +从句"结构，意为"看起来/听起来/感觉好像……"。 例如：It looks as if our class is going to win.看来我们班好像要获胜了。 7、感官动词+do与+doing的区别： see, watch, observe, notice, look at, hear, listen to, smell, taste, feel + do表示动作的完整性，真实性；+doing 表示动作的连续性，进行性。 I saw him work in the garden yesterday.昨天我看见他在花园里干活了。(强调"我看见了"
初中美文摘抄600字
初中美文摘抄 600 字 篇一：中学生美文摘抄大全 名篇中的经典 中学生美文摘抄大全 名篇中的经典 是的，我很重要。我们每一个人都应该有勇气这样说。我们的地位可能很卑微，我们的 身份可能很渺小，但这丝毫不意味着我们不重要。重要并不是伟大的同义词，它是心灵对生命 的允诺。——毕淑敏，《我很重要》 我送你 4 句话。其一，这世界上没有失败，只有暂时没成功。其二，改变世界之前，需 要改变的是你自己。其三，改变从决定开始，决定在行动之前。其四，是自己的决心，而不是 环境在决定你的命运。你不妨先改变自己的习惯，试着用友善的心态去面对周围的一切，你会 有意想不到的快乐。”——《知识窗》马国福 一个人要想使自己的人生有所造就，就必须懂得在关键时刻把自己带到人生的悬崖，给 自己一片悬崖其实就是给自己一片蔚蓝的天空啊。——《知音》王新文 没有坚冰，谁去认可红梅的烂漫？没有白雪，谁去判定青松的高洁？没有严寒，谁去仰 视乔木耸立？——《感悟春天》朱国良 不管是落花有意，还是流水无情，都是表现了时间之易逝的自然迹象，但是精神上的青 春，意志里的春色，进取中的春光，则当常驻于有志人和奋斗者的岁月年华里！——《感悟春 天》朱国良 阅读文学作品，是一种文化的积累，一种知识的积累，一种智慧的积累，一种感情的积 累。大量地阅读优秀的文学作品，不仅能增长人的知识，也能丰富人的感情。如果对文学一无 所知，而想成为有文化有修养的现代文明人，那是不可想像的。——赵丽宏《致文学》成功的 花，人们只惊羡她现时的明艳！然而当初她的芽儿，浸透了奋斗的泪泉，洒遍了牺牲的血雨。 ——冰心《繁星· 春水》 春天像刚落地的娃娃，从头到脚都是新的，它生长着。春天像小姑娘，花枝招展的，笑 着走着。春天像健壮的青年，有铁一般的胳膊和腰脚，领着我们上前去。——朱自清《春》 人最宝贵的是生命， 生命每个人只有一次。 人的一生应当这样度过： 当回忆往事的时候， 他不会因为虚度年华而悔恨，也不会因为碌碌无 为而羞愧；在临死的时候，他能够说：“我的整个生命和全部精力，都已献给了世界上最 壮丽的事业——为人类的解放而斗争。”——奥斯特洛夫斯基《钢铁是怎样炼成的》 我在朦胧中， 眼前展开一片海边碧绿的沙地来， 上面深蓝的天空中挂着一轮金黄的圆月。 我想：希望本是无所谓有，无所谓无的。这正如地上的路；其实地上本没有路，走的人多了， 也便成了路。——鲁迅《故乡》 每个人降生到这个世界上来，一定有一个对于他最合宜的位置。一个位置对于自己是否 最合宜，标准不是看社会上有多少人争夺它，眼红它，而应该去问自己的生命和灵魂，看它们 是否真正感到快乐。——周国平《最合宜的位置》
宗教对美国社会的影响
宗教对美国社会的影响 梁子毓英语学院英语137班 摘要：美国是当今世界上最发达的资本主义国家，同时又是发达国家中最具宗教色彩的国家，可以说美国的发展与宗教有着密不可分的关系。宗教从美国建国伊始至今对人们思想的影响是根本性的，这种影响绝不仅仅只是停留在道德方面，在政体的确立、民主制度的促进等方面都有着同样深远的作用。如今，宗教在美国的影响力有增无减，信仰上帝的人越来越多了，宗教在美国国家社会生活中的作用也越来越大。因此，研究宗教对美国发展的影响更有助于我们了解美国的社会现实和文化特征，可以使我们对宗教在美国社会的影响有更深刻的认识。 关键词：宗教美国社会影响 引言 美国文化的一大特征就是其宗教性。来自世界各地不同国家不同种族的移民带来了他们自己的宗教，使美国成为多宗教的国家。各种不同的宗教必然会对美国的社会产生影响，同样也融入了美国的文化。在众多宗教中，影响最大的是新教众多教派中的清教，清教在发展过程中，其影响超越了宗教领域，渐渐渗透到了美国宗教以外的政治、经济、文化领域，使美国政治、经济、文化都带有明显的清教主义特征。清教主义因而成为美国文化和社会形成过程中的重要因素。美国的文明都刻有明显的清教主义印记，清教主义文化也造就了独特的美利坚精神与文化。 一、基督教新教对美国历史的影响 美国虽然只有四百多年的历史，但却从一个英属殖民地逐渐成为世界上的头号强国。在这片土地上诞生了世界上第一部成文宪法，产生了世界上第一个民选总统，建立了三权分立的民主政体，这些都对世界历史产生了深远的影响。然而这些都与基督新教及其伦理道德有着密不可分的关系，这种关系甚至是决定性的。 早期殖民北美的新教1徒的新教信仰，构成了北美早期社会思想风潮的主调，也构成了以后美利坚的民族精神以及国家意识形态的基础。美国独立战争的发生，是因为早期移民北美的多数人都是新教教徒的缘故。由逃亡的清教徒们建立的美洲殖民地，在宗教上，一直与英国本土的宗教处于对立状态。美洲大陆的宗教主流为清教徒，英国的国教则为天主教与新教的混合体圣公会安力甘宗，安力甘宗作为英国国教一直是清教徒改革的对象。在美国独立战争及18世纪二十年代，英国本土和美洲殖民地发生了一场轰轰烈烈的宗教“伟大复兴”运动，这场运动在美国是新教教义的普及和强化运动，是一场彻头彻尾的新教教义在新大陆被强化的运动，这场运动最后导致了新教的进一步振兴，从而在思想上与英国国教彻底脱离了关系。宗教运动进一步促进了北美殖民地人群的主体意识，进一步加强了殖民地与英国本土的在宗教上和政治上的离心力，为独立战争做了思想和意识形态的准备。北美殖民地与英国本土上的宗教对立，对美国独立战争的爆发有着深远的影响。 对美国近代发展奠定基础的南北战争，实际上也有着深刻的宗教原因——美国清教对南方奴隶制的憎恶而引起的南北方对立。废奴主义一直是基督教教义的传统。在新教中这被理解为上帝爱世上的每一个人。这种思想成为西方人权思想的源头。既然上帝爱每一个人，个 1新教：新教（Protestantism）是由16世纪宗教改革运动中脱离罗马天主教的一系列新教派的统称，主流教派有路德宗、加尔文宗和圣公会。
通止规的用法及管理
通止规的用法及管理 1、止规 使用前：应经相关检验计量机构检验计量合格后，方可投入生产现场使用。 使用时：应注意被测螺纹公差等级及偏差代号与环规标识公差等级、偏差代号相同（如M24*1.5-6h与M24*1.5-5g两种环规外形相同，其螺纹公差带不相同，错用后将产生批量不合格品)。 检验测量过程：首先要清理干净被测螺纹油污及杂质，然后在环规与被测螺纹对正后，用大母指与食指转动环规，旋入螺纹长度在2个螺距之内为合格，否则判为不合格品。 2、通规 使用前：应经相关检验计量机构检验计量合格后，方可投入生产现场使用。 使用时：应注意被测螺纹公差等级及偏差代号与环规标识的公差等级、偏差代号相同（如M24*1.5-6h与M24*1.5-5g两种环规外形相同，其螺纹公差带不相同，错用后将产生批量不合格品)。 检验测量过程：首先要清理干净被测螺纹塞规油污及杂质，然后在环规与被测螺纹对正后，用大母指与食指转动环规，使其在自由状态下旋合通过螺纹全部长度判定合格，否则以不通判定。 3、注意事项 在用量具应在每个工作日用校对塞规计量一次。经校对塞规计量超差或者达到计量器具周检期限的环规，由计量管理人员收回、标识隔离并作相应的处理措施。 可调节螺纹环规经调整后，测量部位会产生失圆，此现象由计量修复人员经螺纹磨削加工后再次计量鉴定，各尺寸合格后方可投入使用。 报废环规应标识隔离并及时处理，不得流入生产现场。 4、维护与保养 量具（环规）使用完毕后，应及时清理干净测量部位附着物，存放在规定的量具盒内。生产现场在用量具应摆放在工艺定置位置，轻拿轻放，以防止磕碰而损坏测量表面。 严禁将量具作为切削工具强制旋入螺纹，避免造成早期磨损。可调节螺纹环规严禁非计量工作人员随意调整，确保量具的准确性。环规长时间不用，应交计量管理部门妥善保管。
感官动词的用法
1.感官动词用法之一：see, hear, listen to, watch, notice等词，后接宾语，再接动词原形或ing形式。前者表全过程，后者表正在进行。句中有频率词时，以上的词也常跟动词原形。 I heard someone knocking at the door when I fell asleep. (我入睡时有人正敲门) I heard someone knock at the door three times. (听的是全过程) I often watch my classmates play volleyball after school.(此处有频率词often) 若以上词用于被动语态，后面原有动词原形改为带to不定式： We saw him go into the restaurant. →He was seen to go into the restaurant. I hear the boy cry every day. →The boy is heard to cry every day. 2.感官动词用法之二：look, sound, smell, taste, feel可当系动词，后接形容词： He looks angry. It sounds good. The flowers smell beautiful. The sweets taste sweet. The silk feels soft. I felt tired. They all looked tired. 这些动词都不用于被动语态。如：The sweets are tasted sweet.是个病句。注意：如果加介词like,则后不可接形容词，而接名词或代词：
名家美文摘抄600字
名家美文摘抄 600 字 篇一：适合中学生美文摘抄大全 名篇中的经典 适合中学生美文摘抄大全 名篇中的经典 本篇收录了 80 段左右摘抄自名家适合中学生阅读的美文摘抄，全都是名篇中的经典，希 望大家能够喜欢。 是的，我很重要。我们每一个人都应该有勇气这样说。我们的地位可能很卑微，我们的身 份可能很渺小，但这丝毫不意味着我们不重要。重要并不是伟大的同义词，它是心灵对生命的 允诺。 ——毕淑敏，《我很重要》 我送你 4 句话。其一，这世界上没有失败，只有暂时没成功。其二，改变世界之前，需要 改变的是你自己。其三，改变从决定开始，决定在行动之前。其四，是自己的决心，而不是环 境在决定你的命运。你不妨先改变自己的习惯，试着用友善的心态去面对周围的一切，你会有 意想不到的快乐。” ——《知识窗》马国福 一个人要想使自己的人生有所造就，就必须懂得在关键时刻把自己带到人生的悬崖，给自 己一片悬崖其实就是给自己一片蔚蓝的天空啊。 ——《知音》王新文 没有坚冰，谁去认可红梅的烂漫？没有白雪，谁去判定青松的高洁？没有严寒，谁去仰视 乔木耸立？ ——《感悟春天》朱国良 不管是落花有意， 还是流水无情， 都是表现了时间之易逝的自然迹象， 但是精神上的青春， 意志里的春色，进取中的春光，则当常驻于有志人和奋斗者的岁月年华里！ ——《感悟春天》 朱国良 阅读文学作品， 是一种文化的积累， 一种知识的积累， 一种智慧的积累， 一种感情的积累。 大量地阅读优秀的文学作品， 不仅能增长人的知识， 也能丰富人的感情。 如果对文学一无所知， 而想成为有文化有修养的现代文明人，那是不可想像的。 ——赵丽宏《致文学》 成功的花，人们只惊羡她现时的明艳！然而当初她的芽儿，浸透了奋斗的泪泉，洒遍了牺 牲的血雨。 ——冰心《繁星· 春水》 春天像刚落地的娃娃，从头到脚都是新的，它生长着。 春天像小姑娘，花枝招展的，笑着走着。200 字美文摘抄 春天像健壮的青年，有铁一般的胳膊和腰脚，领着我们上前去。 ——朱自清《春》
优美文章600字摘抄片段
优美文章600字摘抄片段 【篇一：优美文章600字摘抄片段】 位置： > > > 文章内容 600字美文摘抄欣赏时间:2016-03-29 15:07来源:美文网点击:...次 作者：微尘陌上 精彩段落摘抄： 是谁，在一纸浅墨文字里安静等待？是谁，在一路时光里仆仆风尘？又是谁，在一叶轻舟里只影远行？ 那个人，是清浅岁月里最沉重的那一滴雨，在记忆的天空，季季滴落，季季来去。我流连在三月的雨季，看见永恒的天空已然由暗灰 变作了烟青，看见蛮荒的塬上渐渐滋生出生命嫩绿的芽子，写意成 一帘梵净山里早春的风景。一树茶花，朦胧烟雨，借着三月乍暖还 寒的春光，我分明看见，晶莹的泪珠两行，悄然跌落，是滑落过岁 月绸帕上的清冽水迹。 很多时候，我想，你一如深山里的山茶，你一如幽潭里的素莲，在 未至的花期里，幽隐着，而我，便在与你相约的时光里，对生活心 怀敬意，--期待，在那一个花期来临的日子，一睹你花开的样子。 如此，我退居于山野小村，营生着一颗素朴的心，在不计时长的小 村时光里，热爱着平凡的事情，热爱陌生人家的爽朗笑声，热爱市 井人生的无聊嬉戏，也热爱小猫小狗在日头下的慵懒困意，偶尔也 吃茶养花，偶尔也读书听雨，沉湎于日月星辰、花草田间、露水晨昏，在低温的日子，不急不缓，深情地活着！ 与你，相约，是一种清浅的禅意，因为，想你，是一种忧伤的美丽！初春的雨，总是会落进心里的某个地方，或者落进某一个不经意间 涌动的念里，天青的颜色，乍暖还寒时候，最难将息。就如看见细 雨飘零，总会令人无端想起旧年的黄梅子黄梅雨，还有一个遥远北 方的你。 来于大千，行于世间，或许，每一个人都会有极致的孤独，每一个 人都会有深眷的时刻，都有一个可以唯一与之相约的人，那个人， 或许在天边，或许在近前，而那一款独一的深情，却是极致到不悔，就算，真正能陪着自己走完人生的，只有时间最钟情 600字美文摘抄欣赏 作者：左寒ㄨ右影べ 精彩段落摘抄：
通止规的用法及管理
通止规的用法及管理 令狐采学 1、止规 使用前：应经相关检验计量机构检验计量合格后，方可投入生产现场使用。 使用时：应注意被测螺纹公差等级及偏差代号与环规标识公差等级、偏差代号相同（如M24*1.56h与M24*1.55g两种环规外形相同，其螺纹公差带不相同，错用后将产生批量不合格品)。 检验测量过程：首先要清理干净被测螺纹油污及杂质，然后在环规与被测螺纹对正后，用大母指与食指转动环规，旋入螺纹长度在2个螺距之内为合格，否则判为不合格品。 2、通规 使用前：应经相关检验计量机构检验计量合格后，方可投入生
产现场使用。 使用时：应注意被测螺纹公差等级及偏差代号与环规标识的公差等级、偏差代号相同（如M24*1.56h与M24*1.55g两种环规外形相同，其螺纹公差带不相同，错用后将产生批量不合格品)。 检验测量过程：首先要清理干净被测螺纹塞规油污及杂质，然后在环规与被测螺纹对正后，用大母指与食指转动环规，使其在自由状态下旋合通过螺纹全部长度判定合格，否则以不通判定。 3、注意事项 在用量具应在每个工作日用校对塞规计量一次。经校对塞规计量超差或者达到计量器具周检期限的环规，由计量管理人员收回、标识隔离并作相应的处理措施。 可调节螺纹环规经调整后，测量部位会产生失圆，此现象由计量修复人员经螺纹磨削加工后再次计量鉴定，各尺寸合格后方
可投入使用。 报废环规应标识隔离并及时处理，不得流入生产现场。 4、维护与保养 量具（环规）使用完毕后，应及时清理干净测量部位附着物，存放在规定的量具盒内。生产现场在用量具应摆放在工艺定置位置，轻拿轻放，以防止磕碰而损坏测量表面。 严禁将量具作为切削工具强制旋入螺纹，避免造成早期磨损。可调节螺纹环规严禁非计量工作人员随意调整，确保量具的准确性。环规长时间不用，应交计量管理部门妥善保管。
感官动词
感官动词的概念和相关考点 1、什么是感官动词？ 听觉：listen to、hear 视觉：look at、seem、watch 嗅觉：smell 触觉：feel、touch 味觉：taste 2、感官动词如何正确使用？ Tom drove his car away. →I saw him drive away. (全过程) 用法一：somebody did sth + I saw this I saw somebody do something. Tom was waiting for the bus. →I saw Tom waiting for the bus. (看不到全过程) 用法二：somebody was doing sth + I saw this I saw somebody doing something 练习： 一、句子翻译 1. I didn,t hear you come in. 2. I suddenly felt sth touch me on the shoulder. 3. I could hear it raining. 4. Listen to the birds singing. 5. Can you smell sth burning? 6. I found Sue in my room reading my letters. 二、灵活运用 1. I saw Ann waiting for the bus. 2. I saw Dave and Helen playing tenins. 3. I saw Clair having her meal. 三、选择最佳选项 1. Did anybody see the accident (happen/happening)? 2. We listen to the old man (tell/telling) his story from beginning to the end. 3. Listen! Can you hear a baby (cry/crying)? 4.—Why did you turn around suddenly? — I heard someone (call/calling) my name. 5. We watched the two men (open/opening) a window and (climb/climbing) through it into house. 6. When we got there, we found our cat (sleep/sleeping) on the table. 四、感官动词的被动语态 Oh，the milk is tasted strange.