the humble servant

[Fulvio Romano]

I forgot something important, the images of the patterns and drawings, as well as much of the arguments, are taken from the book "industrial rhbotics" of shavicco - Sicilian. at the faculty of robotics this book is friendlyly called "the bible" :wink:.

 

[Fulvio Romano]

an example of robot "soffice"

 

[MBT]

an example of robot "soffice"

is also included in the "hard-ware" category or rather in the "soft-ware" category :tongue::biggrin:

 

[Fulvio Romano]

a few posts ago I said that the carrier laying of the terminal organ is composed of a part of position and one of orientation:

where q=(q1, q2, ..., qn) is the vector of the joint variables and x = [p, fi] = [(x1, x2, x3), fi) è il vettore posa (= posizione + orientamento) nello spazio operativo. Perché non esplicito il vettore orientamento fi? Beh...lo vedremo tra un po', non è così ovvio.

I have not specified in practice what is the carrier fi that represents the orientation. Well, let's see more detail now.
the purpose is to describe a complete trajectory that combines two points in space, depending on time. If it is rather intuitive to generate a position trajectory between two points, such as a segment, or a circumference arc, it is not so intuitive to generate a trajectory, for example "line" between two directions.
rotation matriceswe have seen, although not in formal detail, the rotation matrices. by means of a rotation matrix it is possible to take a described vector with respect to a reference tern, and to describe it with respect to another tern, rotated with respect to the first. Therefore a rotation matrix can be considered the description of the orientation of one terna compared to another.
a rotation matrix in the Euclidean space, has three rows and three columns, for a total of nine elements, obviously not all independent. being an orthogonal matrix (operates an isometry in a tensoral space) only three of its elements are independent. It means that if we want to use a rotation matrix for the description of an orientation, we should manage, point by point, nine elements of which three are calculated by hourly equations, and add to these six congruence equations that maintain the orthogonal matrix at every point.
It is not a good deal, in fact the rotation matrices are not used for this purpose, because of the strong redundancy of information inherent in their structure.
angoli di eulerowe saw that the parameters for the description of the orientation are three. It is then spontaneous to say that the best solution is a so-called "minimum description" or a three-parameter description. It's not like that, but we'll see why.
the most widespread representation to three parameters is called description by "eulero angles".
This representation consists in describing a orientation as succession of three successive rotations around three axes.
as the axes are three and the successive rotations must be three, we have 27 possible combinations, but from these we must eliminate all those in which two consecutive axes coincide. 12 different tares of eulous angles remain. to describe a orientation through three parameters and it is necessary to know with what terna has been described.

perhaps it is here the case of remembering that for rotations not infinitesimal, the tensor field that describes them does not enjoy the commutative property. if ruoto of alpha around x, of beta around y and of range around z I reach a thirst. if the three rotations occur in a different order you will get in general a different order. to understand it just take a parallelepipedo and make a test, for simplicity with all rotations of 90°.

one of the most famous tarts is the so-called "rpy" which is usually used to describe the arrangement in the space of aircraft and hulls. "rpy" stands for:
roll, that is the roll angle, the angle around the longitudinal axis
pitch, i.e. the beaking angle, around the cross axis
Yaw, get the angle of landing around the cross axis.

See case the orientation of a spherical wrist can be described by associating the r-p-y parameters to the 4-5-6 axes. comfortable, huh? Unfortunately, no...

Unfortunately, even the corners of eulero are little used in robotics for a series of structural problems related to the particular mathematical structure. Many robots expose the set expressed in eulous angles, but in reality the controller for their calculations uses one of the two systems described after.
What are the problems? let's list them.
- first of all the angles of eulero present a series of singularity of representation. This is a more serious problem than the already described singularities of the Jacobin. In fact the singularities of the Jacobin are the mathematical correspondent of physically existing cinematic singularities. must be managed, of course, but those singularities are physically present in the cinematic structure of the manipulator. in the case of the corners of eulero instead these singularities (called "representation") are linked exclusively to the mathematical structure of the chosen representation. therefore are not the limits of the manipulator, but only the numerical limits introduced by the specific method used for calculations. In fact the singularities relative to a certain terna are not present in the other tares, this has made it want in the past to use more tares with singularity of representation not aligned, and to jump from one to another to "scan" the problems. This method has the undisputed disadvantage of having to write and solve equations to "join" the descriptions at the points where you decide to sew them, complicating even more the method.

- they have a rather unstable mathematical structure. in particular around corners close to 0° and 180° where the breasts and the things cancel or become very small.

- they are mapped in the operating space counterintuitively. It is difficult to explain without seeing what you are describing, but we imagine that we want to describe a rotation around a single axis, other than those used for representation. for example a 45° axis compared to the three coordinated plans. How could you do that? Intuitively the rotation could be done precisely around this axis. Very simple, isn't it? instead a linear trajectory realized by a representation of the angles of eulero describes this simple rotation in a much more complex way. you will see the object rotate around itself along several axes with a "bad" movement, counterintuitive and with non-homogeneous speed, then, of course, at the end of the trajectory, with the required arrangement.
anglethen we overcome the difficulties of representation through the corners of eulero. we must give up a "minimum" representation of only three parameters. "let's pawn" introducing a four-parameter representation, then redundant, which will need a congruence equation, but with this we gain a number of advantages.

the first representation to four parameters, the simplest, is that called "axis / angle". If I have two sets in space I always have the possibility to identify a single rotation, around a precise axis, which can describe the second regarding the first.
If as parameters I use this axis' guiding things, and the angle around it I have to rotate, I made bingo.
I have only one singularity, if the angle roof of rotation is null, I have no way to know about which axis should be carried out. little bad, because it is not a singularity of representation, but a physical singularity. If I have a null angle, then no rotation, it will be enough to give mathematical consistency by arbitrarily choosing an axis.
the congruence equation to add, of course, is that the sum of the squares of the managerial things is equal to one.

the advantages are remarkable. we solved the problems of singularity of representation, we have an intrinsically linear mathematical structure because based on a single rotation and on the projections of a vector on the coordinated planes; therefore intrinsically more stable. also disappears the problem of the mapping of the trajectory in the operating space we had with the angles of eulero. a rotation around an axis is now performed in the most natural way possible, or just as rotation around this axis.

the only drawback of this representation, in addition to that of the congruence equation in more, is an ambiguity. a tit rotation around r, and one of -teta around -r actually concise. It is an ambiguity, it must be managed, but much better than managing singularities.

 

[Fulvio Romano]

unit quaternationthe axis/angle representation is an excellent solution to the problem of the description of the setup, but we take a step further and discover a new representation to four parameters. not within the mathematical detail, but I would like to give just some news on a mathematical concept that remained completely unknown until the fifth year of university.
from a numerical point of view the use of this method is very simple. I define:

eta = cos(teta / 2)
epsilon = sin(teta / 2) * r
(epsilon is a vector, as r is a verse, what identifies the rotation axis)

seems simple. the equation of congruence becomes:

eta^2 + epsilon_x^2 + epsilon_y^2 + epsilon_z^2 = 1

from which the term "quaternion (the first member) unit".

we solved the problems of the axis/angle because the definition of this quaterna of numbers makes a rotation of roof around r and one of -teta around -r give origin to the same quaternion. therefore the ambiguity of the axis/angle representation is lost because a given rotation identifies a unique quaternion.

very simple, but in reality the term "quaternion" hides a much deeper meaning. the stability of representation through unitary quaternion is actually inherited from the algebra of quaternions that has a mathematical formalism that puts us safe from "holy" related to the specific representation.

quaternions were discovered, or invented, depending on the meaning that one wants to give to these words, by the hamilton mathematician in 1843. like every mathematical structure, it is "invented", then you begin to study it and you "discover" that has properties that are to be studied, analyzed, understood. This is why in mathematics the terms "discover" and "invent" are not enough in my opinion to explain what actually happens.

the "quaternion" is an extension of the complex number, which in turn is an extension of the real number. real numbers can be represented as a straight, a size.

the complex number can be represented on a two-dimensional plane, and has a real part and an imaginary part. the imaginary part is formed by a real number multiplied by "i" the imaginary unit that has the property that its square is equal to less one.
complex numbers are widely used in various fields, especially for the description of undulatory phenomena, from electromagnetic waves to the response of dynamic systems. the reason for this "comodity" is that the complex number can be represented in Cartesian form (real part and imaginary part), but also in polar form (module and phase). There is an important theorem that binds the representation in form and phase of a complex number with a description made by an exponential function.
This means that by expressing an oscillatory law by form and phase, it is possible to study it in the field of complex numbers. being this a field, you have behind all the mathematical formalisms, theorems, the rules that give formal robustness to the whole system.

the quaternion can be represented in a four-dimensional hypercube, and has a real part and a vector part, the latter consisting of three imaginary components. each imaginary component is formed by a real number multiplied by units "i", "j" and "k". each of these units has the square equal to less one, and the multiplication operator has the same effect of the vector product in an euclideous space. (ii) i^2 = k^2 = -1, and i*j = k, j*k = i and k*i = j.
It is interesting to note that having defined the multiplication operator in this way, if i*j = k will also be j*i = -k, that is the field (which is actually a ring, and not a field...) of the quaternions does not enjoy the commutative property. Yeah, we were expecting that, right? exactly like rotations in the Euclidean spaces.

quaternions are not very used, and not much known. are used above all, as we have seen, in the study of transformations (for example rotations) in the euclide spaces to more dimensions. their use is useful because having a complete algebra behind them make possible the use of a mathematical formalism already ready. their use in control schemes makes the calculation more stable.

 

[africagia]

Hey, give me time, I haven't gotten to control yet. we have arrived at the cinematics, we still have to talk about direct dynamics, reverse dynamics, independent controllers to the joints, centralized.... we are preparing analyses ii and we arrived at the algebra...:tongue:

Okay! I'll wait for you to get to control!! :finger:

 

[Fulvio Romano]

We finally start talking about control.

robot control is a fascinating science in its complexity. before entering the details however it is important to begin to understand what you mean by "control" and "automation".
"automation" means a set of technologies, called "control systems", aimed at making as much autonomous machines and processes as possible, industrial and non-industrial. the term "automation", according to a theory whose sources I can no longer find (dnaction!), comes from a transcription error. the term appeared in the early 1950s when for the first time the concept of machine can move independently. to make the concept of automatic movement was used the neologism "auto-motion" which by a transcription error became "automation".
with automation therefore means a technology whose purpose is to give some kind of autonomy to a machine, whether it is autonomy of movement, reasoning, analysis or deduction.

autonomy is made with control techniques. also here is the case of spending two words to understand what is meant by the term "controlling", because in the common jargon this word can have other meanings. often by "controlling" is in fact meant to "verify".
If the baby has a fever, Mom says: "We control temperature," but in reality it should say more correctly "we measure temperature", because control assumes an action aimed at changing a magnitude so that it takes on the desired values. always with the same example, if with the thermometer misuro the temperature, I decide that it is too high and I give the child an antipyretic so that it descends below a definite threshold a priori, here is that the whole of these actions can be correctly defined "temperature control".

 

[SHIREN]

We finally start talking about control.

robot control is a fascinating science in its complexity. before entering the details however it is important to begin to understand what you mean by "control" and "automation".
"automation" means a set of technologies, called "control systems", aimed at making as much autonomous machines and processes as possible, industrial and non-industrial. the term "automation", according to a theory whose sources I can no longer find (dnaction!), comes from a transcription error. the term appeared in the early 1950s when for the first time the concept of machine can move independently. to make the concept of automatic movement was used the neologism "auto-motion" which by a transcription error became "automation".
with automation therefore means a technology whose purpose is to give some kind of autonomy to a machine, whether it is autonomy of movement, reasoning, analysis or deduction.

autonomy is made with control techniques. also here is the case of spending two words to understand what is meant by the term "controlling", because in the common jargon this word can have other meanings. often by "controlling" is in fact meant to "verify".
If the baby has a fever, Mom says: "We control temperature," but in reality it should say more correctly "we measure temperature", because control assumes an action aimed at changing a magnitude so that it takes on the desired values. always with the same example, if with the thermometer misuro the temperature, I decide that it is too high and I give the child an antipyretic so that it descends below a definite threshold a priori, here is that the whole of these actions can be correctly defined "temperature control".

Hello fulvio
a question a misleading moment
but control and controller have the same autonomy
thank you 1000 :smile:

 

[Fulvio Romano]

a control scheme therefore is formed mainly by the following blocks, as seen in the figure:
- a process to control
- a "set point" or a value you want to get from the process
- a "controller", or an entity that from the set point is able to generate inputs (u) for the process to be controlled, so that its outputs are as close as possible to the desired ones, or to the setpoint.

so far the control is said in "open ring", or "fire control". the controller in fact generates outputs on the basis of the setpoint assuming that the system to control does what he expects. for this type of schemes the controller must obviously somehow know the system to control. must therefore necessarily be based on model.

Let's make an example. we are in the car, we impose the setpoint with two variables. the first that the speed is 50km/h, the second that follows the paved road. Suppose we know perfectly the car and the road, we can, with closed eyes, govern the accelerator and the steering wheel so as to remain in a small around 50km/h and more or less in the center of the road. a deep knowledge of both the road and the car is necessary, but it is a possible task.
What if a side vent comes? Does the car roam? How much? Wouldn't it be better to open your eyes?

 

[Fulvio Romano]

Okay, let's open our eyes, so let's change the control scheme, like in the second figure. now we have:
- always the process to control, the car
- always set point, i.e. follow the road, with speed set
- always a controller
but now it adds
- a "retrotion" or a measure of the y output of the system to check that back
- a node sumer. in this node enter the setpoint and the y output of the system to control, and exit [setpoint - y], that is a magnitude that measures the distance of y from the setpoint, what I have from what I would like to have, that is the error of control.

this type of control is called "feedback", or "powering (feed) backwards (back)", or "negative retroaction"; retroaction because the measure of y back into the scheme, and negative because it does it with a minus sign.

we realize that the control scheme is much more effective. it is no longer necessary to know perfectly the car and the road because as soon as you accelerate a little too the error will become negative and will remove gas to the car, vice versa if it slows too much will become positive and will increase the acceleration. the car now can also be driven by a newbie who has never climbed before. not only, but when the so-called side vent arrives, the car will wave, the mistake will warn this bandage and will compensate it. Will he do it in a timely manner? Well, that depends on the controller, but the important thing is that you do it. it is said that control is "robust" compared to the disorder, because it is able to reject it.

the world around us is zeppo of systems with negative retroaction. from the endocrine system to the labyrinth of the inner ear that does not fall to the ground.

 

[Fulvio Romano]

to understand a little more practically what a controller is, we see the so-called "pid" or controllers with proportional-integral-derived action. a scheme of principle is visible in the image below.

We said that the controller's task is to take the setpoint in, or the desired value, and to give out what must be the input of the process so that the output of the same is as close to the desired one.
proportionalthe simplest idea is to get a mistake as a difference between the desired output and the real one, and multiply this error by a constant.
back to the car example. I could decide to open the butterfly valve of an alpha angle = k * and, where 'is' is the speed error, and k is a constant size appropriate [°s/m]. This means that if the car slows down, the butterfly valve will open, and the machine will accelerate by reducing the error. of course also vice versa, for example if you start a descent.
It seems to work, but there is a problem. with purely proportional control, the error cannot be brought to zero. In fact the more the error approaches zero the exit of the controller is reduced. practically, when the system exit approaches a lot to the reference, the control stops working, reducing its output more and more. if the exit of the controller is u(t) = kp * and(t), for a constant reference, to regime we have that the error is e = u / kp. this means that the higher the kp the smaller the error to regime, but without being able to cancel. If it were e = 0, it would also be u = 0, or the process would be uncontrolled.
However it is not possible to increase to dismisura kp. First of all, proportional control involves fluctuations. returning to the example of the car means that if the car is too slow the system will accelerate until it exceeds a tot the reference speed, at this point it will slow down until it descends (of a little less than that tot) under the reference value, and so on. increasing kp means increasing the extent of these oscillations to the risk of achieving a real instability.
proportional-integralto avoid these problems you can add an integral action. This action is like a memory. its entity is not only based on the present error, but also on the past one. as "the story teaches", the addition of this action in principle improves control action. the first important effect is that even for very small ki, of botto the error to regime goes to zero. Maybe for too small ki it will take a while, but we are sure it will go to zero. demonstrating it is very simple. the exit of control is

u = kp and + integral ki(e dt)

if you derive from the time you get:

♪

but for the concept of "regime" all that is derived from time goes to zero. then you get:

ki and t = 0 -> e = 0 (at regime)

the effect of an integral action also behaves as a "smotor" reducing the entity, or completely erasing overelongations and oscillations.
Of course, there are counters in the use of integral action, in particular a slowdown in the controlled system, and the triggering of rather insidious instability situations.
proportional-integral-derivederivative action in theory improves the system even more. is to generate a rate of control action that is proportional to the resulting error. It is a kind of "predictor" (controllers pass me the term) in fact in a way moves in advance previewing the evolutions of the system. practically if the error begins to grow very quickly (for example the car meets a steep climb), the proportional rate does not notice anything, because the error is still small; the integral rate itself history, because the error is not only small, but has been present for a short time; the derivative rate, however, immediately realizes the impennata of the error and runs away for time, increasing the control action.
derivative action, from the mathematical point of view, has a stabilizing action on the system. However there is a big problem that practically does not recommend use in most applications. The error is calculated as a difference between the setpoint and the system output, but the system output is read through transducers, whose output is affected by noise. the derivative of a noisy entrance is not beautiful to manage, it has large completely incorrect excursions of the system under control. However to avoid this problem you can use a stratagem. a control system that is respected brings the system output rather close to the reference setpoint. therefore the derivative of the reference, instead that of the error is therefore wrong from the theoretical point of view, but rather functional from the engineering one.

 

[Fulvio Romano]

robot control

Let's finally talk about robot control. the purpose of control of a robot is to generate electrical currents with which to feed the engines, so that the tool performs a predetermined path, preciding from what can happen in the world around the robot, if of course this is physically possible
.
As we have already said, control cannot ignore the knowledge of the mechanical structure of the robot. For control of a Cartesian manipulator, for example, the problems to be faced are considerably different, and in many ways simpler, than those to be faced for an anthropomorphic manipulator. the choice of engines also has an influence on the type of control. high-relation gear motors will have a linearizing effect on the dynamics of the manipulator, however introducing games, elasticity and accelerations of coriolis (the fast shaft is... really fast!), vice versa the use of torque engines simplifies the processing of the single joint, but it requires the modeling of the superior dynamics of the manipulator, because they are immediately sensitive on the electric side.

There are several more or less complex and more or less performing control schemes. But we can make a first big distinction, that is, control in the operating space and that in the space of the joints. the two control philosophies arise from the fact that typically the characteristics of the desired motion are described in the operating space, while the real control must be expressed in the space of the joints.

the control schemes in the space of the joints simplifies a bit the treatment as it clearly separates the two problems. first the desired motion, through the cinematic inversion, is transported into the space of the joints, and only then the control strategy operates in this last space. but on the one hand the separation of the two problems leads to undeniable advantages of both conceptual and computational, there is to say that the control action is carried out on quantities other than those you really want to control. the transformation between the operating space and the space of the joints is strongly non-linear, thus operating a closed ring control that maintains small the error exclusively in the space of the joints, does not guarantee anything, in general, on what happens in the operating space, where conceptually the control remains in open ring through the mechanical structure of the manipulator.

in the control schemes in the operating space instead you go directly to control the variables of interest, that is position, speed and acceleration in the Cartesian space. However, the fact that the cinematic inversion takes place directly in the control ring, i.e. the innermost point, more intimate, faster and more delicate than the whole system, makes it easy to guess how much algorithmic and computational complexity increases. Although the control comes in the operating space, and therefore the error is kept small in such space, in reality it is a “virtual” error. the position in the space assumed by the manipulator is not measured directly, for example by means of cameras, but reconstructed by direct cinematics starting from the data of the resolvers, which ahimè measure greatness in the space of the joints.
a robot modeling error, for example a longer or shorter arm than its denavit-hartenberg parameter passed to control, will result in a systematic error completely ignored by control.

 

[Fulvio Romano]

but without entering into the dynamics of a manipulator, whose treatment is rather complicated and also a little boring, it is easy to guess the following:

the law of newton on the bike tells us that a force is equal to a mass for an acceleration:

♪

We try to describe in a similar way the law of motion of a manipulator. we called “tau” the generalized forces to the joints, that is the forces or couples, depending on the type of joint, and “q” the joint variable. we call then:

b(q) = is the inertia matrix of the manipulator. is naturally dependent on the current configuration, and consists of elements that represent the inertia of the entire manipulator compared to the various joints*.

c(q, q’) = is a matrix that describes the mortgage terms of mating. we imagine a two-arm planar manipulator in a given configuration. b(q1) represents the inertia of the manipulator felt by the first joint and b(q2) that felt by the second joint (only the second arm). If, however, the q2 joint moves, this movement will be felt in some way even by q1. This rate is not present in b, because this matrix, row by row, considers the joints different from the current one blocked. the matrix is concerned with describing this mortgage not considered in b.

fv(q’) = viscous friction coefficients of joints. should be considered because usually the gearboxes are in the oil bath, and the viscous friction is felt. the static one instead we neglect it because of poor entity and because it is more difficult to treat, being not linear.

g(q) = is a function that represents the gravity force “feel” from the various joints according to the current configuration of the manipulator. therefore represents potential energy

defined these quantities we can write the equation of the motion of a manipulator who will be:

tau = b(q) * q’’ + c(q, q’)*q’ + fv*q’ + g(q)

the manipulator is quite large and heavy compared to the transported loads, so inertia is remarkable. we can therefore think of breaking the matrix b in two. an independent part of the configuration, or an average inertia, and another part as a difference:

b = bcost + deltab

control therefore can be divided into two blocks. one will be linear and decoupled, where each joint can be controlled regardless of others, the other will be non-linear and coupled.

the easiest way to control a manipulator with a control system "independent on the joints", is to treat every single joint as a system to be controlled by itself, and consider the whole non-linear part and coupled as a disorder. It is not much as a solution, because the trouble is great, however it will be at least partially rejected by a control system that respects itself.

More advanced methods of control include modelling and online calculation of increasing nonlinear and coupled contributions. The simplest of these, for example, is the calculation of the contribution to the gravity force joint. the calculation is carried out with a rather simple model and is already of considerable effect, given the strong weight of the manipulators.


(*)
or, the kinetic energy of the manipulator can be written as
t = 1/2 q’(transposed) * b(q) * q’
in the case of a single joint is reduced to the known formula:
== sync, corrected by elderman ==

 

[Fulvio Romano]

After so long I would like to add a chapter to this thread, and I would like to talk to you about artificial intelligence.
putroppo, accomplices many science fiction directors, there is much confusion about this concept. many associate with the term "artificial intelligence" something magical and at the same time terrifying, intimately connected to a remote future where machines will take over us humans.

nothing of this, at least from the scientific/engineering point of view.

the term "artificial intelligence" was coined by John mccarthy in 1955 during an important seminary, where with this term it was meant something, made by a machine, that if it were made by a man would require intelligence. It is quite evident that so defined, artificial intelligence, it means everything and nothing. and in fact the biggest problem to solve is to describe in a rigorous way what is meant by intelligence. for simplicity we see before the human one. Although it is quite common to label people with "stupid", or "intelligent", we actually have no idea what intelligence is. often confuses a brilliant person with a smart person; often even a "sympathetic" person, in the extended sense of the term, that is a person who thinks like us, with whom we are in tune.
einstein was certainly intelligent, in fact he developed many innovative theories both in theoretical and experimental physics. so we could define this as intelligence, with the result of considering mozart a fool, since of mathematics he understood very little. vice versa einstein could be considered stupid, since he was unable to play any instrument.
According to some theories there are even nine different types of intelligence, distributed among the population in a more or less varied manner. for example an alighieri dante was certainly rich in intelligence linguistics, where einstein was richer than that logic-matematic. if a canoe had a strong intelligence space space carla fracci certainly has corporeo-cinestesica. chopin certainly had excellent intelligence musical, but excelled also in that intrapersonal, which, on the other hand, often lacks to those who are rich in the logical-matematic one. other people with a strong intrapersonal intelligence could be gandhi, mariateresa of calcutta, etc. the last two types of intelligence are those nature ed existential. if some types of intelligence, such as the logical-mathematic one, are very simple to identify, and perhaps even to measure, there are others such as the body-cinestesic one that are practically impossible to catalog. For example if you can compare an einstein with a stop, and you can maybe even catch on who is smarter, compare a kubrick with spielberg would certainly be much more embarrassing.
We have understood that the very concept of intelligence is very complex to define, and without being able to define it becomes impossible to study rigorously and systematically.

Good. What is a dog smarter than a turtle? It's an extremely simple question, which doesn't call into question all that dirt of different types of intelligence. anyone would claim that a dog is smarter than a turtle, and yet in fact the only cognitive processes that the dog manage clearly and the turtle does not, are those related to the carry of a stick, or to drool at the time of the pappa. the dog therefore is not " smarter", but only closer to the image a man wants to find in a pet. as to say, it looks more like us, so we consider it smarter. even in this therefore, it is very complex to abstract a concept of intelligence that has some formal meaning. .

 

[MBT]

therefore, a very normal modern pc, on which a software is run, is an "artificial intelligence", as a "deficent" person would not be able to do just as... .

Am I right?

 

[PierArg]

No mbt.
a common software only executes commands.

a pc with sw with a lot of neural networks for training is a "intelligent" pc.

 

[MBT]

No mbt.

a pc with sw with a lot of neural networks is a "intelligent" pc

hum. Why do you have neural networks??

 

[Fulvio Romano]

therefore, a very normal modern pc, on which a software is run, is an "artificial intelligence", as a "deficent" person would not be able to do just as... .
Am I right?

depends on what the software does. I just wanted to pay attention to the fact that the first important step to solve a problem is to define it carefully. and the problem of intelligence has not yet been well defined. . .

No mbt.
a common software only executes commands.
a pc with sw with a lot of neural networks is a "intelligent" pc.

uhuuu... how do you run... Give me time, don't you? :finger:

 

[PierArg]

neural networks are computer algorithms that aim to make something "apprehensive" to the PC.

for example think of a pc and its various uses. a neural network, properly trained, would be able to track a "profile" of your pc's use.
Suppose you use it for calculation: your neural network makes your pc have an increasingly performing processor.
How?
Perhaps you will have a machine language code such as to "evolve yourself" (depending on the information collected from the neural network) and make the most of every single clock of the cpu.
or has drivers that follow the same philosophy.

neural networks work just like neural networks of the brain.
the more trained and the more links manage to establish (the famous synapses).
more connections have more performance.

 

[PierArg]

Capers, I'm sorry.
I just read your post now.