Military Review

The first flight of a "computer brain" on a UTAP-22 drone took place in the USA

38

On April 29, the first flight of the so-called "computer brain" Skyborg, which was on board the Kratos UTAP-22 Mako drone, took place at the Tyndall base of the US Air Force in Florida (USA). This device is an early version of the intelligent autonomous piloting system.


This is reported by the American portal The Drive.

This "computer brain" was developed by Kratos Defense & Security for the US Air Force. The Skyborg program provides for the simultaneous development of UAVs and software for the operators who will operate them. There is also a possibility that this project will be used in manned aircraft (aircrafts) in order to reduce the workload on pilots.


It is assumed that Skyborg will be able to autonomously solve various tasks related to the control of aircraft. Among them are not only takeoff, landing and flight at a given course, but also much more complex functions. For example, in real combat conditions, he is able to assess the situation, on the basis of which to make and implement decisions.

In the future, it is planned to test the system on drones created by Boeing, General Atomics and Kratos.

Photos used:
https://twitter.com/MIL_STD
38 comments
Information
Dear reader, to leave comments on the publication, you must sign in.
  1. UPS
    UPS 7 May 2021 06: 19
    -4
    After landing, was the BRAIN carried out?
    1. Victor_B
      Victor_B 7 May 2021 06: 57
      -5
      Quote: USV
      After landing, was the BRAIN carried out?

      Did you get it on your brains?
      1. For example
        For example 7 May 2021 07: 07
        -6
        We poured water into a sieve.

        It can take off, fly, land and much more ...

        With a supplement, - along a predetermined route.

        Why is this autopilot called the brain?

        Marketing is demanding.
        For all the good versus all the bad.
        1. NIKN
          NIKN 7 May 2021 09: 05
          -5
          Quote: For example
          Why is this autopilot called the brain?

          Marketing is demanding.

          Well yes. :) If you listen to it so it turns out that you have created a brain, well, then you need to educate correctly, you never know what he wants, maybe become a drug addict and gay, it's already scary for the pilot. :))
    2. RealPilot
      RealPilot 7 May 2021 10: 09
      +1
      Algorithms are good, but very predictable. I mean that after several meetings with real pilots or air defense (and at first the AI ​​may have an advantage, I do not exclude), a methodology for countering these drones will be developed ...
      Especially if a flight of planes is attacking.

      For example, approaching this drone from the optimal angle. Simultaneous attack, but how will he react? Or laser illumination from a distance: it can begin to maneuver and consume fuel / charge, and to the point of being unable to return to base - and this is tantamount to shooting down. All sorts of false targets and portrayed threats are quite suitable, because there will not be a person on board who can appreciate the obvious, but incredible for a computer.
      And there are many such examples ...
      1. voyaka uh
        voyaka uh 7 May 2021 10: 56
        +5
        "that after several meetings with real pilots or air defense (and at first the AI ​​may have an advantage, I do not exclude), a methodology for countering these drones will be developed." ///
        ----
        Quite the opposite. This AI will lose the first battles. But then he will endure everyone in a row.
        These AI software are self-learning. From every real encounter he brings experience
        for the future. And if we consider that the speed of reaction and decision-making at a computer is thousands of times
        faster than human, then real pilots do not shine.
        1. koramax81
          koramax81 7 May 2021 15: 11
          0
          It feels like your knowledge is that you designed it. Or have the terminator been revised?
        2. Nikolaevich I
          Nikolaevich I 7 May 2021 15: 22
          -2
          Quote: voyaka uh
          Quite the opposite. This AI will lose the first battles. But then he will endure everyone in a row.
          These AI software are self-learning. From every real encounter he brings experience

          "Corpses" are not "self-learning"! And you can't put experience in burned-out "brains"!
          1. voyaka uh
            voyaka uh 7 May 2021 17: 10
            +3
            good
            That's right, otherwise fighters with AI would have already fought. It is in the transfer of information to the survivors - a hitch. The battle should be monitored by "observer" drones and receive packets of information from the belligerents in real time.
            It is not very easy, but doable.
            1. OgnennyiKotik
              OgnennyiKotik 7 May 2021 17: 50
              +2
              Quote: voyaka uh
              It is in the transfer of information to the survivors - a hitch.

              This is one of the reasons why there is a combination of fighter + Loyal Wingman. Full autonomy is still far away.
        3. Victorio
          Victorio 7 May 2021 21: 52
          0
          Quote: voyaka uh
          "that after several meetings with real pilots or air defense (and at first the AI ​​may have an advantage, I do not exclude), a methodology for countering these drones will be developed." ///
          ----
          Quite the opposite. This AI will lose the first battles. But further will be endure everyone.
          These AI software are self-learning. From every real encounter he brings experience
          for the future. And if we consider that the speed of reaction and decision-making at a computer is thousands of times
          faster than human, then real pilots do not shine.

          ===
          okay, if this AI begins to endure the Americans themselves, only after all, the whole world will have to pay for them and their papers
      2. alexmach
        alexmach 7 May 2021 13: 24
        +10
        Algorithms are a good thing, but very predictable.

        Yes of course. Play chess with a computer program, see who re-predicts whom.
  2. Bare
    Bare 7 May 2021 06: 25
    -3
    Departure of the brain, left without brains. F - 35 can not be brought to the point. What can we talk about. ... ...
    1. UPS
      UPS 7 May 2021 06: 30
      -5
      About the fact that the brains on the drone flew out, but whether to return ...
      1. Egoza
        Egoza 7 May 2021 06: 36
        -4
        in real combat conditions, he is able to assess the situation, on the basis of which to make and implement decisions.

        This means that the pilots will not need brains at all. They already do not rely on them.
    2. vadim dok
      vadim dok 7 May 2021 15: 28
      -6
      What can they do! SU-57 is a thing, there are no analogues!
  3. Thrifty
    Thrifty 7 May 2021 07: 13
    -4
    As always, in typical conditions, with multiple safety net of flight on the ground and in the air, no matter what happens! This is bullshit, not a test! They would have sent it in bad weather, in conditions of limited visibility, and with an imitation of an air attack on this drone!
  4. rocket757
    rocket757 7 May 2021 07: 47
    +1
    Air Force in Florida (USA) on April 29, the first flight of the so-called "computer brain" Skyborg took place, which was on board the Kratos UTAP-22 Mako drone.
    It is clear that they are rushing ahead ahead of the whole planet ...
  5. Revolver
    Revolver 7 May 2021 07: 52
    +4
    Skyborg is already done. The next number in the program is Skynet.
  6. APASUS
    APASUS 7 May 2021 08: 26
    -3
    UTAP-22 lands, the foreman comes up and says what he thinks about the recruit's flight laughing laughing laughing

    And that the problem of data transmission has already been solved by the Americans?
    1. alexmach
      alexmach 7 May 2021 13: 27
      +2
      And that the problem of data transmission has already been solved by the Americans?

      You will not believe...
      1. APASUS
        APASUS 11 May 2021 13: 45
        0
        Quote: alexmach
        You will not believe...

        I won’t believe it. Because you need a huge amount of information to process and transmit in a split second, otherwise it’s just another expensive rattle.
        1. alexmach
          alexmach 11 May 2021 15: 01
          0
          1.the same Link-16 (sixteen?) They have had a tactical data transmission system for a long time
          2. And why would he transfer a huge amount of data? What for?
          1. APASUS
            APASUS 11 May 2021 15: 15
            0
            Quote: alexmach
            and why would he transfer a huge amount of data? What for?

            In my opinion, here in the material indicated
            in real combat conditions, he is able to assess the situation, on the basis of which to make and implement decisions.

            Quote: alexmach
            That same Link-16 (sixteen?) They have had a tactical data transmission system a long time ago

            What are the constant materials about the lack of speed when transferring data, for use in a pair of UAVs - a fighter plane.
            1. alexmach
              alexmach 11 May 2021 16: 10
              0
              In my opinion, here in the material indicated
              in real combat conditions, he is able to assess the situation, on the basis of which to make and implement decisions

              AND? Why transfer data somewhere for this? Do you personally assess the situation around you? Are you making decisions? Doing this without data transfer?

              What are the constant materials about the lack of speed when transferring data, for use in a pair of UAVs - a fighter plane.

              This is the first time I hear about such materials. Who has the Americans?
              I generally have no idea what problems there may be when transferring data during a pair flight, with a line of sight between the vehicles, and in polygon conditions. If in a combat situation they manage to control drones from the ground and at the same time do not experience any problems with data transmission channels, why should they experience these problems in flight, and with an autonomous drone that does not require such detailed control or transmission of a video signal.
              1. APASUS
                APASUS 11 May 2021 16: 48
                0
                Quote: alexmach
                I have no idea what problems there may be when transferring data when flying in pairs,

                That's how it turns out to be simple, but the Americans are interested in the course?
                1. alexmach
                  alexmach 11 May 2021 17: 16
                  0
                  That's how it turns out to be simple, but the Americans are interested in the course?

                  In the course of what? The fact that they have problems? I don’t know, it seems to me vryatli.
  7. donavi49
    donavi49 7 May 2021 08: 27
    +11
    By the way, who wrote there that the Korean K21 is a layout at the presentation? Jogging for July-August. Now ground tests are underway at the LII. Flight in winter.
  8. Antidote
    Antidote 7 May 2021 08: 41
    +12
    How many malice and jokes for 300. However, c70 and, scary to say, a blizzard, for some reason cause a continuous "hurray"
  9. val43
    val43 7 May 2021 09: 11
    +12
    Quote: Antidote
    How many malice and jokes for 300. However, c70 and, scary to say, a blizzard, for some reason cause a continuous "hurray"

    And what do you want - it's a topvar. There is no other audience here, with rare exceptions. request
  10. Ural resident
    Ural resident 7 May 2021 09: 33
    -7
    What kind of brain is this? Judging by the video, he simply performs the tasks set by the operator. There is no need to talk about some kind of independent choice of actions.
  11. Synoid
    Synoid 7 May 2021 12: 23
    -1
    Quote: voyaka uh
    "that after several meetings with real pilots or air defense (and at first the AI ​​may have an advantage, I do not exclude), a methodology for countering these drones will be developed." ///
    ----
    Quite the opposite. This AI will lose the first battles. But then he will endure everyone in a row.
    These AI software are self-learning. From every real encounter he brings experience
    for the future. And if we consider that the speed of reaction and decision-making at a computer is thousands of times
    faster than human, then real pilots do not shine.

    Well, if he loses, where is the chance to win back?
    If you mean training battles, how much do they correlate with reality by 30%? They have already invented missiles once, abandoned cannons, as if the planes would not engage in close combat. Hundreds of pilots paid. As help in a non-combat situation - yes, well, or, in extreme cases, bomb defenseless countries. Well, about AI, it can't always win in chess, but how many options are there?
    1. voyaka uh
      voyaka uh 8 May 2021 13: 19
      +1
      "Well, about AI, he can't always win in chess, but how many options are there?" ///
      ----
      Always playing chess. 100%.
      Even in Go - the "intuitive game of mathematicians" where it is fundamentally impossible
      enumeration tactics
      - AI smashes world champions dry.
      And in training battles on the F-16 simulator, the AI ​​defeated the experienced ace pilot dry.
  12. RealPilot
    RealPilot 7 May 2021 13: 47
    +4
    Quote: voyaka uh
    "that after several meetings with real pilots or air defense (and at first the AI ​​may have an advantage, I do not exclude), a methodology for countering these drones will be developed." ///
    ----
    Quite the opposite. This AI will lose the first battles. But then he will endure everyone in a row.
    These AI software are self-learning. From every real encounter he brings experience
    for the future. And if we consider that the speed of reaction and decision-making at a computer is thousands of times
    faster than human, then real pilots do not shine.

    Alex, good afternoon!
    You are right, but in part.
    Self-learning through artificial neural networks, adaptive filtering and other algorithms is currently very imperfect. And it won't drastically improve in the foreseeable future. Only by speed and a large number of processing threads.
    These drones will behave well in some scenarios, and in some others, they will be vulnerable.
    That is, the heights, color of camouflage and the speed of opponents, the intensity of electronic warfare, distance from allied platforms and the presence / absence of AWACS, false targets, the density of air defense - these and many other factors will directly affect the effectiveness of the application.
    There is also such an effect as "retraining" of the neural network, when there is so much information that serious errors begin to creep into even the initially programmed tasks. Personally, while writing a diploma in Baumank on a related topic (neural networks in near-range radar, I will not write the exact name - it even stands as a "special topic" in the diploma), although for a long time, however, I simulated neural networks on a computer. And I saw how the realizations of signals from real radars (namely, they were given to me by my scientific advisor) in certain configurations were determined exactly against the background of the control "nonsense", and as the complexity of the network increased, the results began to blur. Then they recovered again, and then lost again ...
    All this is not easy! It is quite possible to get auxiliary platforms, but to make atomic superhunters that will react accurately is still unrealistic. I emphasize that this requires technologies in at least a generation or two from the best current ...
    1. Intruder
      Intruder 7 May 2021 16: 16
      +1
      Self-learning through artificial neural networks, adaptive filtering and other algorithms is currently very imperfect. And it won't drastically improve in the foreseeable future. Only with speed and a large number of processing threads
      Probably, according to the past and ancient data about NN, you have such a judgment !? Now convolutional networks, even localized "on one pass", with machine learning, in the field of nonlinear analysis of optical signatures of images, outrun their counterparts another 2-3 years ago, and these are biometric access technologies even in Chinese sensors are used everywhere and massively, not to mention About visual orientation with mapping, a drone on Mars uses a similar autonomous system with one video sensor in total! And this is already a cardinal improvement, but about complex systems - heteromorphic types of NN, with peripheral accelerators, there, in general, a year is considered a human epoch ... what
  13. RealPilot
    RealPilot 7 May 2021 13: 59
    +2
    Quote: alexmach
    Algorithms are a good thing, but very predictable.

    Yes of course. Play chess with a computer program, see who re-predicts whom.

    In chess 32 pieces, 64 cells, millions of recorded parts. This combinatorics can be solved easily, just by substitution of known moves. The openings and endgames have been worked out, the methods of "corralling" the king with 2 or more pieces are standard. I had the 2nd category, we were taught typical techniques.
    Chess players, backgammon players improvise, while the computer searches the tables with moves from many games of the strongest players. Such is the system of linear equations ... Numerical mathematical methods.

    And here, in the case of drones, the algorithms work "under conditions of a priori uncertainty." This is a term that was used in Baumank, when I was studying near-range radar. You can't say it better, just the definition of the topic.
    1. Intruder
      Intruder 7 May 2021 16: 07
      0
      And here, in the case of drones, the algorithms work "under conditions of a priori uncertainty"
      nonlinear algorithms, if we enter the area of ​​uncertainty of one kind or another, already in the first / second year of specialized faculties pass in the field of control theory:

      functional nonlinear algorithms;

      logical nonlinear algorithms;

      optimizing nonlinear algorithms;

      parametric nonlinear algorithms.

      An important difference between nonlinear algorithms and linear ones is that they give the system fundamentally new properties. If a linear algorithm always generates a signal proportional to the input variable or its derivative, etc., then with a nonlinear algorithm the very nature of the action of the control system on the object can change significantly depending on the magnitude of the input. In other words, if for linear systems the change in the size of the deviation is only a change in the scale, not in the form of processes, then in a nonlinear system the form of processes can also change significantly, up to fundamental qualitative changes in the picture of processes. These special properties of nonlinear algorithms can be advantageously used in automatic control engineering. yes
  14. Intruder
    Intruder 7 May 2021 16: 02
    -1
    It is assumed that Skyborg will be able to autonomously solve various tasks related to the control of aircraft. Among them are not only takeoff, landing and flight at a given course, but also much more complex functions. For example, in real combat conditions, he is able to assess the situation, on the basis of which to make and implement decisions.
    That is, the pilot in the cockpit will already be a redundant link, for making a single decision, pressing with one finger on the unlock button for the onboard armament, and to talk to the combat control center of the operation, namely live communication, and not the telemetry data stream from the board, boldly and a little quickly ... decided to introduce such a narrow solution for unloading the crew of a manned bird.