Machinery and Systems

Abstract

To provide a machine that can efficiently automate agriculture. [Solution] A machine that recognizes crops and the environment around the machine based on information obtained from one or more sensors, and uses AI to determine whether or not agricultural work can be performed based on information about the crops and the environment.

Description

[Technical Field]

[0001] The present invention relates to a machine or system. [Background technology]

[0002] The statements in this section only provide background information regarding the present disclosure and are not intended to be limiting unless otherwise specified. Not necessarily composed.

[0003] Patent Document 1 discloses a temperature control system for an agricultural greenhouse. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Laid-Open No. 2016-18972 Summary of the Invention [Problem to be solved by the invention]

[0005] However, the inventor believes that at least the above embodiment has a configuration that can efficiently automate agriculture. I realized there was a drawback to not having it. [Means for solving the problem]

[0006] At least one aspect of the present disclosure provides a method for manufacturing a semiconductor device, comprising: Agricultural machinery, Recognizes crops based on information from one or more sensors, Recognizes the environment around the machine, AI will determine whether or not agricultural work can be performed based on crop and environmental information. machine to provide. [Effects of the Invention]

[0007] This configuration is useful at least for efficiently automating agriculture.

[0008] These and other aspects, features, and advantages of the present disclosure are best understood from the following detailed description taken in conjunction with the following drawings. As will become apparent from the following detailed written description of preferred embodiments and aspects, variations and modifications thereof are within the scope of the present invention. and modifications may be made without departing from the spirit and scope of the novel concepts of this disclosure. Aspects of one embodiment in this specification may be used in conjunction with other embodiments disclosed herein, unless inconsistent. This can be combined with or substituted for one or more of the aspects in the above. DETAILED DESCRIPTION OF THE INVENTION

[0009] In the following disclosure, many different implementations are presented for implementing different features of the presented subject matter. To simplify the disclosure, specific examples of components and arrangements are provided below. Of course, these are merely examples and are not intended to be limiting. For example, a first feature may be covered by or border on a second feature subsequently disclosed. The structure may include an embodiment in which the first feature and the second feature are formed in direct contact. Also, an additional feature may be formed between the first feature and the second feature to allow the first feature and the second feature to be combined. This disclosure may include embodiments where direct contact between features is avoided. Reference numbers and / or letters may be repeated in the various examples. The repetition of the above is for the sake of brevity and clarity and as such may be used in conjunction with various embodiments and / or No relationship is required between the components or structures described. When a first element is described as being "linked" or "coupled" to a second element Such a description does not necessarily mean that the first element and the second element are directly connected or coupled to each other. and the first and second elements may be connected to one or more intervening It also includes embodiments in which the elements are indirectly connected or coupled to each other by having other elements.

[0010] As used herein, the phrase "at least one of" "A" includes all the variations exemplified. For example, "A, B, and C" includes all the variations exemplified. es at least one of !, B, or C)" means "A, B, C, and combinations of these" (consisting of A, B, C and combinations thereof)" and A, B, This disclosure encompasses all possible variations of A, C, A+B, A+C, B+C, and A+B+C. Disclosure of an embodiment combining two or more features is included unless inconsistent or otherwise specified herein. Unless otherwise specified, any one or more components may be implemented as separate embodiments. For example, the statement "perform A, B, and C" should be interpreted as "perform A, B, or C, and any combination thereof." "comprising of A, B or C and combinations thereof" is synonymous with "A" , B, C, A+B, A+C, B+C, A+B+C.

[0011] In this disclosure, disclosures using a machine, electronic operator, or computer include methods, The present invention may include embodiments of a recording medium, a device, or a program. A statement such as "A is B" is true unless there is a contradiction or unless otherwise stated in the specification. As long as "A contains B" is used, it can be replaced with "A contains B."

[0012] Terms in this disclosure, including terms in the claims, are defined as described in the specification. The present disclosure is not intended to be limiting unless it is expressly stated otherwise. Anything that has been so named, designated, or understood by one or more citizens in the past, present, or future. It can be interpreted based on what has been done or is likely to be done. The operation method used in at least one embodiment may take the following embodiments. The following description will be made with reference to JP6456303, which clearly explains at least one embodiment. (Quote begins below.)

[0013] As used herein, the term "computer" refers to a computer system as known in the art. processor, memory, such as a hard drive, disk drive or flash drive, flash drive or memory stick, or other non-transitory computer-readable medium or at least one information storage / retrieval device, such as a non-transitory storage device, e.g., a keyboard; , mouse, point and touch device, touch screen, or microphone. At least one input device and a display such as a familiar computer screen. In addition, the computer may be connected to one or more As is known in the art, such computers may include a network connection. or a computer system may include more or less of the above listed items. other electronic media, such as but not limited to tablet computers and smart devices. This includes media and electronic devices.

[0014] As used herein, the terms "cloud" or "cloud computing" " is a centralized system where all computing resources are shared. Refers to virtualized and virtualized computing facilities. Application systems and subsystems, all in the "cloud" So it can no longer be pointed to a specific machine.

[0015] As used herein, the term "distributed internet service system" means: Modifying Internet applications to run in a variety of computing environments The DIS system is a distributed internet service platform that exchanges Component Distribution Server er) / Asset Distribution Server Internet applications, including content, data, and logic, To whatever extent appropriate, and distribute it to any number and type of devices along the network. We provide Internet applications with services based on the needs of each user. Host and manage it centrally, maintaining its integrity while delivering it to users on their devices or nearby It can be cached and executed locally at the location of the web-enabled computing device. The device will be upgraded with DIS software to provide distributed internet services. Decentralized Internet Services can be DIS-enabled, allowing users to enjoy and execute their services. The system is protected by U.S. Patent Nos. 7,136,857, 7,150,015, and 7,181,731. , No. 7209921, No. 7430610, No. 7685183, No. 7685577 , No. 7752214, No. 8326883, No. 8386525, No. 8443035 , No. 8458142, No. 8458222, No. 8473468, No. 8527545 , and No. 8650226, and U.S. Patent Publication No. 20120005205, and The present invention is fully described in any one of the patent family No. 20130091252, All of these, as well as the present invention, are owned by O.P. 40 Holdings, Inc. All rights reserved, all of which are incorporated by reference. (End of quote)

[0016] The method of operation used in at least one embodiment uses a distributed internet. Regarding the conventional Internet method, the following embodiments can be taken. The following description will be made with reference to the description of JP7113047, which clearly explains at least one embodiment (hereinafter, the reference start of use).

[0017] The embodiments including the matters specifically disclosed in this specification are based on artificial intelligence and are actually human-made. It is possible to provide an automatic response system that is realized in a form that allows you to have a conversation with someone, This allows for more natural conversations with users while also speeding up inquiries, reservations, delivery orders, etc. And it can be conveniently processed.

[0018] The electronic devices 110, 120, 130, and 140 are implemented by a computer system. The electronic devices 110, 120, 130, 140 may be fixed terminals or mobile terminals. Examples of 40 include AI speakers, smartphones, mobile phones, navigation systems, and PCs ( personal computer), notebook PC, digital broadcasting terminal, PDA ( Personal Digital Assistant), PMP (Portable Multimedia Player, Tablet, Game Console, Wearable devices, IoT (internet of things) devices, VR (virtual reality) Reality (AR) devices, Augmented Reality (AR) devices As an example, FIG. 1 shows an AI speaker as the electronic device 110. However, in an embodiment of the present invention, the electronic device 110 may be configured to communicate with the electronic device 110 via a substantially wireless or wired communication method. and / or other electronic devices 120, 130, 140 and / or or various physical computer systems that can communicate with the servers 150, 160. It may mean one of the following:

[0019] The communication method is not limited, and the network 170 can include any communication network (including Examples include mobile communication networks, wired internet, wireless internet, broadcasting networks, satellite networks, etc. ), but may also include short-range wireless communication between devices. For example, ,Network 170 is a personal area network (PAN), L AN(local area network), CAN(campus area network) network), MAN(metropolitan area network), W AN(wide area network), BBN(broadband network) ork), any one or more networks such as the Internet Furthermore, the network 170 may include a bus network, a star network, Ring network, mesh network, star-bus network, tree or tier It may include any one or more of the following network topologies, including layered networks, etc. However, it is not limited to these.

[0020] The servers 150 and 160 respectively connect to a plurality of electronic devices 110, 120, 130, and 140. and communicates with the network 170 to provide instructions, code, files, content, services, The present invention may be implemented by one or more computing devices, such as a server, The server 150 is a server that connects multiple electronic devices 110, 120, 130, and 140 via a network 170. 140, the server 160 may also be a system that provides a first service to the network. A second service is provided to a plurality of electronic devices 110, 120, 130, and 140 connected via a network 170. As a more specific example, the server 150 may be a system that provides a plurality of The computer that is installed and executed in the electronic devices 110, 120, 130, and 140 Through an application, which is a computer program, A service (for example, an automatic answering service) is provided as a first service to a plurality of electronic devices 1 As another example, the server 160 may provide the The file for installing and running the application is stored in a plurality of electronic devices 110. , 120, 130, 140 may be provided as a second service.

[0021] FIG. 2 is a diagram illustrating the internal configuration of an electronic device and a server according to an embodiment of the present invention. 2 is a block diagram showing the internal configuration of an electronic device 110 as an example of an electronic device. The internal configuration of the server 150 will be described. The electronic device 140 and the server 160 are the same as or similar to the electronic device 110 or the server 150 described above. The internal configuration may be as follows.

[0022] The electronic device 110 and the server 150 include memories 211 and 221, processors 212 and 223, and 22, communication modules 213, 223, and input / output interfaces 214, 224 The memories 211 and 221 may include non-transitory computer-readable storage media. It may be a RAM (random access memory), a ROM (read only memory), memory only), disk drives, SSDs (solid state Non-transitory storage devices such as hard disk drives, flash memory, etc. Here, ROM, SSD, flash memory, disk Non-transitory mass storage devices such as drives are separate from memories 211 and 221. The electronic device 110 or the server 150 may be included as a non-transitory recording device. The memory 211, 221 contains an operating system and at least one program code. A browser installed and executed on the electronic device 110, for example, Applications installed on the electronic device 110 for providing specific services Such software components may be stored in memory 21. 1, 221 may be loaded from a computer-readable recording medium other than the Other such computer readable storage media include floppy drives, diskette drives, and Computer-readable disks, tapes, DVD / CD-ROM drives, memory cards, etc. In other embodiments, the software components may include a removable storage medium. The memory is transmitted through the communication modules 213 and 223, which are not computer-readable recording media. For example, at least one program may be loaded into the development library 211, 221. A file distribution system (including a For example, the above-mentioned server 160 provides a file via the network 170. computer programs installed by the The data may be loaded into the memories 211 and 221 based on the program.

[0023] Processors 212, 222 perform basic arithmetic, logic, and input / output operations The processor may be configured to process instructions of a computer program by: The memory 211, 221 or the communication module 213, 223 allows the processor 212 , 222. For example, the processors 212, 222 may be provided to the memories 211, 222. Executes instructions received according to program code stored on a storage device such as The device may be configured to:

[0024] The communication modules 213 and 223 communicate with the electronic device 110 and the service via the network 170. The electronic device 110 and / or the server 150 may provide functionality for communication with each other. Alternatively, the server 150 may communicate with other electronic devices (for example, the electronic device 120) or other servers (for example, For example, the electronic device may provide functionality for communicating with a server 160. The processor 212 of the device 110 reads a program code stored in a storage device such as a memory 211. The request generated according to the code is transmitted to the network according to the control of the communication module 213. The information may be transmitted to the server 150 via the network 170. Conversely, the information may be transmitted to the server 150 via the network 170. Control signals, commands, contents, files, etc. provided under the control of the , the communication module 223 and the network 170 of the electronic device 110 13 to the electronic device 110. For example, through the communication module 213 The received control signals, instructions, content, files, etc. from the server 150 are sent to the processor 2 12 or memory 211, and the contents, files, etc. may be transferred to the electronic device 110. It may also be recorded on a recording medium (the non-transitory recording device mentioned above) that may include the recording medium.

[0025] Input / output interface 214 is for interfacing with input / output device 215. For example, the input device may be a keyboard, a mouse, a microphone, a camera, or the like. The output devices are devices such as displays, speakers, and haptic feedback devices. As another example, the input / output interface 214 may include any device. Interface with devices that integrate input and output functions into one, such as a touch screen. The input / output device 215 may be a means for connecting the electronic device 110 to one device. The input / output interface 224 of the server 150 may be configured as follows: 150 or may be included in the server 150 (as shown) A more specific example is an electronic device. The processor 212 of the 110 reads instructions of a computer program loaded into the memory 211. When processing the data, the server 150 and the electronic device 120 are used to provide the data. The service screen and contents to be displayed are displayed through the input / output interface 214. It may be displayed in (i).

[0026] In another embodiment, the electronic device 110 and the server 150 may be implemented using the components of FIG. However, most of the prior art components are clearly illustrated. For example, the electronic device 110 may include any of the input / output devices 215 described above. It may be realized to include at least some of the transceiver, camera, various sensors, It may further include other components such as a database. When the electronic device 110 is an AI speaker, various sensors that are generally included in an AI speaker are camera module, various physical buttons, buttons using touch panel, input / output Various components such as force ports, vibrators for vibration, etc. may be further included in the electronic device 110. It may be implemented so that it is included. (End of quote)

[0027] According to at least one embodiment, a user terminal includes a control unit. ,RAM, storage section, graphics processing section, communication interface, interface In at least one embodiment, ,User terminal includes terminals owned by users.,On the other hand, terminals owned by users The owners are not only users but also sellers and traders of goods and services, governments and local governments. For example, the provision or advertisement of goods or services (hereinafter, (hereinafter referred to as "the provision, etc.") (including those transferred or lent), terminals provided for the provision, etc., This includes the terminals of the recipient of the goods or services. This includes devices owned by others that the user is only permitted to use temporarily, This includes loaned devices.

[0028] According to at least one embodiment, the control unit is configured with a CPU and a ROM. The RA executes the programs stored in the storage unit and controls the user terminal. M is the work area of ​​the control unit. The storage unit is used to store programs and data. The control unit reads programs and data from RAM and performs processing. The control unit processes the programs and data loaded in RAM to execute drawing commands. The graphics processor outputs the command.

[0029] According to at least one embodiment, the graphics processing unit is coupled to a display unit. The display unit has a display screen. The control unit outputs a drawing command to the graphics processing unit. The graphics processing unit outputs a video signal for displaying an image on a display screen. Here, the display unit may be a touch panel equipped with a touch sensor. The touch panel functions as the input section.

[0030] According to at least one embodiment, the communication interface is a wireless or wired communication network. It is possible to connect to a network and send and receive data with a server device via the communication network. Data received via the communication interface is loaded into RAM. The interface unit is connected to an external memory (for example, An SD card or other device is connected.

[0031] According to at least one embodiment, the user terminal is a computer having a display screen and an input section. There is no particular limitation on the type of computer device. The user terminal may be, for example, a conventional mobile phone. Telephones, tablets, smartphones, desktop and laptop personal computers VR goggles, i.e., strapped or attached to the head A screen (or two displays) mounted in a frame (or headset) The user terminal may be configured with an audio output section. Has.

[0032] According to at least one embodiment, a user terminal communicates with a service via a communications network. It is possible to establish a communication connection with the server device via a communication network and transmit information. or receive information.

[0033] According to at least one embodiment, the server device includes a control unit, a RAM, a storage unit, and and a communication interface, which are connected to each other by an internal bus.

[0034] According to at least one embodiment, the control unit is configured with a CPU and a ROM, and The control unit executes the program stored in the memory unit and controls the server device. It has an internal timer that measures the time. RAM is the work area for the control unit. The memory area is a storage area for storing programs and data. and data from RAM, and executes the program based on the information received from the user terminal. Execute the execution process.

[0035] According to at least one embodiment, artificial intelligence (AI) is disclosed. This includes layer learning, generative AI, large-scale language models, LLM, foundational models, and generative AI. It uses a formifier and a number of mechanisms called attention. Self-supervised learning, Extract t Prediction is used. In this case, the AI ​​can guess the next word. Then, guess the next word from the sentence up to that point. This creates a large number of supervised learning problems. By doing so, we can create an AI that can guess the next word. Generative AI uses grammatical structure, topic connections, It is possible to predict what kind of sentences a person with a certain writing style will write. AI can learn the structure, causal relationships, and knowledge behind the next sentence simply by guessing it. The generation AI has a scalability, and the greater the number of parameters, the higher the accuracy. Statistics and machine learning tend to make the model parameters larger compared to the sample size of the data. If the number of parameters is too large, overfitting occurs. The accuracy of LLM decreases as the number of parameters increases. One generation AI has 175 billion parameters, and the generation AI can smoothly carry out dialogue. Supervised learning is applied so that the students do not say strange things. Writing and playing a call center operator.

[0036] According to at least one embodiment, Large Language Models (LLM) ) is a non-comprehensive approach built using large datasets and deep learning techniques. , a machine learning natural language processing model. Generally, it is trained on a specific task. Using a technique called "fine tuning," which involves automating the process, we are able to perform text classification and generation, sentiment analysis, and other tasks. It is adapted to various natural language processing (NLP) tasks such as analysis, text summarization, and question answering. According to at least one embodiment, self-supervised learning approximates the innate intelligence of humans. When acting, the robot is always predicting the next event and the next input. In the process, we learn the structure of the external world. Predicting the next word is an essential intelligence, and the cerebrum According to at least one embodiment, a large-scale language model It memorizes the information it receives, but generalizes enough to predict the next word. Do not try to generalize all information from the beginning. Large-scale language models use a lot of memory to remember information. Capacity is required. Parameters are also required for this. At least one embodiment According to the study, large-scale language models are models with 175 billion parameters and 220 billion parameters. It is equipped with eight of these.

[0037] According to at least one embodiment, the video or image is converted into small text tokens similar to the LLM. We represent the image as a set of visual patches, which are small data units. Effectively represent your model and train your generative model on a wide variety of videos and images. First, we convert the video into a low-dimensional latent space (LST). The video is then converted into patches by compressing it to, and then decomposing the representation into spatiotemporal patches.

[0038] According to at least one embodiment, a video compression network The network (work) is a network that reduces the dimensionality of visual data, taking raw video as input. The AI ​​outputs a latent representation that is compressed both temporally and spatially. It is trained between, and then generates videos in this compressed latent space.

[0039] According to at least one embodiment, Spacetime Latent Patches are Given a compressed input video, we generate a sequence of transformer tokens. The patch-based representation allows Sora to extract spatiotemporal patches of various resolutions and lengths. It can be trained on videos and images of any aspect ratio and uses randomly initialized patterns during inference. Control the size of the resulting video by placing the switches in a grid of the appropriate size. .

[0040] According to at least one embodiment, the AI ​​is a diffusion model and As more patches (and conditioning information such as text prompts) are entered, the original The AI ​​is trained to predict "clean" patches. It is a model that can be used in a variety of fields, including language modeling, computer vision, and image generation. The diffusion transformer exhibits remarkable scaling characteristics in the video production area. As the amount of training computation increases, the AI ​​becomes more efficient at sampling. The quality of the image is significantly improved.

[0041] According to at least one embodiment, AI applies caption regeneration technology to Train a descriptive caption model and then use it to generate a caption for the training set Generate text captions for all videos in your video library. The training improves the overall quality of the generated video as well as the fidelity of the text. It converts short user prompts into long, detailed captions and sends them to the model. This allows the AI ​​to generate high-quality videos that accurately follow user prompts. do.

[0042] According to at least one embodiment, the AI ​​performs natural language processing in the following manner: Vectorization can be performed. First, the given text is cleaned as a preprocessing step. The cleaning process removes JavaScript code and HTML contained in the text. Remove unnecessary words such as tags. These codes are used to display on the internet. This is a code used for the purpose of identifying the target language, and is therefore information that is not generally used in natural language processing. Next, the sentence is divided into words by morphological analysis. It is the classification of a written natural language sentence into the smallest meaningful linguistic unit. As a morphological analysis tool, "MeCab", "JUMAN" and "JANOME" can be used. In the process of merging, words with the same meaning, such as variations in spelling, are unified into a single word. These are words that are not processed because they cannot be used in natural language processing. Examples of words include particles and auxiliary verbs that do not have meaning on their own. When calculating the vector, these are removed and only meaningful words are considered. Sometimes vectorization is performed without removing these stop words. Vectorization is the process of converting a string of words into a vector. Convert data into numerical data. When converting words into vectors, use Bag of Words or This is done using a method called distributed representation. A bag of words is a set of words that appear in a given sentence. This is a method to vectorize a sentence by using the number of occurrences of words. The focus is on whether a word or sentence appears, so the order of words and sentences is not taken into consideration. This is a method of vectorizing words by focusing on their meaning. It can give vectors that are close to words with similar meanings and usages, and can also be used simply. Relationships between words can also be expressed as vectors. Vector representations allow us to distinguish between the meanings of words. The application process converts natural language into numerical data and uses it as input for machine learning. Specifically, vectorized natural language is fed into a classifier to classify sentences. The tools used here are "TensorFlow," "scikit-learn," and "PyTo Examples include "rch".

[0043] In at least one embodiment, the machine comprises one of the methods of the present disclosure. The above embodiments or functions may be combined.

[0044] The term "machine" includes a machine that is self-moving or a machine equipped with a power source that is self-moving. The machine may be equipped with an image capturing device. The machine includes a robot. The machine has a communication means, Other computers, computing facilities, cloud, internet, and apps In one example, the machine can communicate information with at least one Wheeled machinery, including self-propelled vehicles. Self-propelled vehicles are vehicles or machines that are unmanned and self-propelled. Furthermore, the term "machine" includes a walking or running machine with at least one moving part. It is possible for the robot to walk on two legs. An example is a humanoid robot. A humanoid robot is at least Each robot has two moving parts, which correspond to the legs. These machines include robots or humanoid robots that incorporate robotics. By utilizing technology and machine learning algorithms, the robot can recognize its surroundings and act autonomously. The machine is equipped with AI technology and sensors, which allow it to understand the surrounding environment in real time. For example, it can move around while avoiding obstacles and select appropriate actions. It is possible for the machine to communicate with people around it. An example of an aircraft is a machine that can be used for aviation. This includes airplanes, rotorcraft, gliders, airships, and other equipment specified by government ordinance. It may have one or more propellers. The unmanned aerial vehicle is equipped with an image capture device. The aircraft is capable of capturing video, including thermography.

[0045] A video capture device or means is disclosed. In at least one embodiment, the video capture The obtaining device may be implemented as any of the machines and devices described in this disclosure. The image capture device includes a device that captures an image and converts it into digital data. cameras (take still images and videos and save them on a recording medium), video cameras (mainly for taking videos) webcams (connected to a computer and used for video conferencing and live streaming); Surveillance cameras (installed for crime prevention purposes and constantly recording video), including smartphone cameras In this disclosure, the term "image" has a very broad meaning. It refers to the image created by the image, that is, the visual information that can be perceived by the eyes. includes thermography (a technology that visualizes invisible heat). It uses an infrared camera to detect infrared rays emitted from an object and displays the intensity of the rays in different colors. By doing so, the temperature distribution is captured as an image. This includes the device that acquires it.

[0046] A method for acquiring human motion data using motion capture is disclosed. In at least one embodiment, motion capture refers to the digital recording of a person's or object's movements. Optical technology involves the use of multiple infrared cameras attached to the target. The inertial sensor attached to the object converts the three-dimensional position and orientation of the marker into data. The markerless (video) system measures the target's silhouette using a video camera. The system reads the data and estimates the bone position. The human is selected depending on the purpose of the service provided. For example, a person who performs work. Work is any physical action, regardless of its type. It is acceptable. Light work is included. For example, carrying luggage to a designated location, loading and unloading, cooking, etc. They make things, do farm work, weld, do regular work on factory lines, and clean. This includes two or more people working together to carry out a certain task. This includes the task of transferring the resulting product to the next designated person.

[0047] A method for learning motion data using AI is disclosed. In at least one embodiment, Then, using reinforcement learning, we collect motion capture clips of real people. We train a control policy that mimics human behavior. The policy is trained in physics simulation to track the reference motion pose of Second, by using different reference motions in the reward function, we can improve the simulated Robots can be trained to imitate a variety of skills.

[0048] A method for generating adaptive behavior in response to changes in the environment around a machine is disclosed. In one embodiment, simulators generally provide only a rough approximation of the real world. Therefore, policies trained in simulation cannot be deployed to real robots. This can lead to poor performance. Therefore, we need to optimize the latent space with high sample efficiency. Using adaptive techniques to transfer policies trained in simulation to the real world First, to encourage the policy to learn behavior that is robust to changes in dynamics, In addition, we change the physical quantities of the robot, such as its mass and friction, to change the dynamics of the simulation. Randomize the values ​​of these parameters during training in simulation. Since the data has access to the image, it is also possible to use the learned encoder to map it to a lower-dimensional representation. This encoding is passed as an additional input to the policy during training. Since the physical parameters of the actual robot are unknown in advance, it is necessary to apply the policy to the actual robot. When it's time to deploy, you can remove the encoder and watch the robot successfully perform the desired skill in the real world. This technique searches directly in the latent space for a set of parameters that allow the Real-world data can be used to adapt the policy to the real world. In our approach, we train policies in simulation and then apply them to the real world. It is possible.

[0049] However, when a task involves complex and diverse physical phenomena, it is often necessary to learn directly from real-world experience. Multitask learning procedures, safety-constrained learners, and carefully designed Using several hardware and software components, Develop an automated learning system with hardware components. Multitask learning. generates a learning schedule that directs the robot to the center of the workspace. Prevent the bot from leaving the training area and design safety constraints to reduce the number of falls. This safety constraint is solved using double gradient descent. The controller selects tasks where the desired walking direction is centered, e.g., forward and backward walking. If the robot is at the back of the workspace, it performs the forward task. During the episode, the learner chooses the right task and vice versa for the backward task. A distributional descent procedure is performed to treat both the task objective and the safety constraints as a single goal. Instead of iterative optimization, if the robot falls over, the robot automatically gets up and goes back to normal. Summon Ra to proceed to the next episode.

[0050] Information on changes in the environment around the machine is acquired and training is carried out using those variables. This allows the robot to generate adaptive actions in response to changes in the environment. and based on that information, generate adaptive behavior in response to changes in the environment around the machine. For example, a factory line can handle products and parts that flow on a conveyor belt. It refers to a flow production method in which the workers process and assemble the parts. The movement data is acquired using motion capture. The machine uses AI to The machine that learns and performs the task may include any of the machines described herein. (hereinafter referred to as "work machine"). Work machines respond to changes in the environment around the machine. Industrial machines are given physical parameters about their environment. For example, the flow of The shape of the work object, the coefficient of friction, the places that should not be touched, and a combination of one or more of these The work machine uses physical parameters to perform operations that are adapted to the real world. The physical parameters may be given to the work machine or may be detected by the work machine itself using a sensor or the like. In this way, the work machine can use the human's actions as a reference action and This embodiment allows the robot to autonomously capture images of behaviors adapted to the actual environment while imitating the behaviors of the robot. At the very least, it will be possible to realize automation that can be applied based on human actions, which will provide convenience and productivity. There is potential for industrial use.

[0051] When generating the movement, the machine recognizes the environment around it based on information obtained from one or more sensors, A method for modifying operations based on the acquired information is disclosed. In this case, the sensor, whatever its type, measures variables about the environment around the machine or physical This includes devices that can acquire dynamic parameters as data. Examples include LiDAR, cameras, and force sensors. LiDAR includes one or more combinations of these. LiDAR emits laser light and reflects it. It measures the distance to an object and the shape of the object based on light information. It is possible to grasp the distance, shape, and positional relationship of preceding vehicles, pedestrians, buildings, etc. in three dimensions. The machine searches or acquires the physical parameters by itself, for example, by a camera, etc. Understand the speed at which the items to be worked on flow through the factory line, and determine how many seconds it will take to move at that speed. Calculate how much work needs to be done and set up the machine so that the work can be completed within that time. Increase the driving speed. Change the operation based on the information obtained as a result of the inventor's sincere consideration. In particular, in the case of LiDAR, the shape of the object, It is possible to obtain extremely accurate physical parameters related to distance. Noise in learning and calculation based on physical parameters is less likely to occur, and as a result, the results of the learning The results of this behavior are of high quality. This effect was not known in the prior art. This embodiment has novelty and remarkable effects. This has the advantage of being convenient and industrially applicable, since it allows for automation that can be applied to the original process.

[0052] Obtaining motion data of human movements using video, and learning the motion data through reinforcement learning. A method for generating adaptive behavior for similar or different behaviors is disclosed. In at least one embodiment, video is used to acquire motion data of human movements. For example, an image capturing device is placed above a person's head, and the image capturing device captures the person's movements. Any physical action, regardless of its form, is sufficient, and speech or gestures are also acceptable. This also includes a video capture device located at the side of a person to capture video of the person walking while carrying luggage. There is an image capture device on the ceiling of the large warehouse, which captures images of the path people take. The machine performs area estimation. Based on the video, the feature points of the person's body are identified as a designated area. For example, the wrist enters the area where the part is to be picked up. The machine performs posture estimation by learning the feature points of the person's body and detecting the posture of the person. For example, it can detect when a person is picking up a tool and assembling it. The system learns feature images in the background and classifies the background. For example, when working, Detect whether it is in use or not. The machine performs object detection. Detect images that match the learned image from the set area. For example, using a specific tool The machine learns from the motion data using reinforcement learning.

[0053] Reinforcement learning maximizes the score by feeding back the score to the output. In reinforcement learning, unlike supervised learning and unsupervised learning, pre-prepared learning data is not used. It uses two functions: Agent and Environment. An agent is an AI model that we want to develop. An agent receives "state" from the environment as input. A simulator is often used for the environment, and the agent receives input from the environment. The process of observing the environment is sometimes called "observation." The agent observes the environment and generates an output accordingly. , or return a random output (take action). What output, what operation, Whether or not it behaves depends on whether the development target is robot control. The values ​​and codes output by the function are determined by the AI ​​you want to develop. It depends on whether it is a breakthrough game or a Go or Shogi program. It receives the output of the other node and returns a new state as a response. A reward is also given to the agent. The reward is also called a score and is a numerical value. The agent observes the environment, takes some action, and chooses the action that maximizes the reward. The relationship between evaluation of behavior and reward (sometimes punishment) is something humans think about. It is an important tuning technique that determines the function and accuracy of reinforcement learning. In reinforcement learning, this repetition teaches the agent what kind of action to take. The robot learns that if it can stand up, it can effectively accomplish the task. Control of the robot motor, in a block-breaking game, to remove as many blocks as possible without dropping the ball Instead of humans programming the paddle operation, the machine (agent) learns through trial and error. For example, if you activate the motor that puts one foot forward to walk, the robot will lose balance and fall. If you do, you will get no reward or a negative reward. Next, you will move another motor. That is the axis. If the robot does not fall by shifting the center of gravity of its feet, the reward increases, so the operation is considered to be valid. The basis of reinforcement learning is to observe the environment and then It is an algorithm or method for selecting and deciding on actions. The amount of information provided by the environment varies. In the case of a game of Othello, there is little information available immediately after the start of the game. In this case, the position and arrangement of the pieces become more diverse. Evaluation is performed using analytical methods or neural networks. Reinforcement learning is also a "function." The agent receives values ​​(states) from the environment as inputs. The environment also receives the agent's actions as input. It can be said that it is a function that returns a new state. Unlike unsupervised learning, reinforcement learning adjusts the next process based on the output result (reward) of the function. This can be considered dynamic learning. The reward is "the reaction of the environment to the agent's random actions." Maximizing the reward means that the agent randomly selects the The state change when no action is taken (this is called "value") and the random action The change in state (value) when the action was taken is compared, and the one with the higher value is called the "Measure (successful)." The value is calculated using the state value function and the action value function. There are two functions: the state value function calculates the value of not taking random actions. The action value function calculates the value of taking a random action. If the number is higher, the random action is incorporated into the strategy as the correct action. As a result, the AI ​​can develop strategies to win at each stage of the game, as well as the motors that make the robot move. You can learn how to control it.

[0054] The machine learns from the motion data through reinforcement learning. The machine learns by trial and error. It learns optimal behavior. For example, in a logistics warehouse, the task of transporting specific items to specific locations Learn how to hold an object and how to move it. As a result, the robot learns to perform tasks similar to human actions or According to this embodiment, adaptive motions are generated for different motions. This has the advantage of enabling more efficient automation than conventional manufacturing, and has industrial applicability.

[0055] The machine may also use human gestures, voice, gaze, or other similar techniques to generate the action. Recognizing feedback behavior of one or more combinations of the above, and generating feedback behavior based on the feedback behavior The method for evaluating the behavior and modifying the behavior based on the evaluation is disclosed. In an embodiment, the machine captures images of people around the machine using an image capture device. The machine acquires sound by a sound acquisition device. The sound to be acquired may be sound from the surroundings of the machine or sound generated by the machine. Audio capture devices located physically separate from the machine can also capture audio from the area around the machine. Good. The machine recognizes the features of a person's face from the video and recognizes the person's face. The system recognizes the facial expression features of the person and estimates or recognizes the person's gaze from the image of the person's eyes. The person's gestures are estimated from the video. Gestures include body movements and hand gestures. However, it is not limited to hands, and any physical movement is acceptable regardless of its form. For each action, a reward value is set. For example, for making a circle with your hand, A high reward is set for making a "X" sign with the hand, and a low reward is set for making a "X" sign with the hand. Set a high reward for saying "OK" and a low reward for saying "No" In the case of gaze, a high reward is set for gazes directed towards the machine, and a low reward for gazes not directed towards the machine. Set a low reward. In this way, the machine can recognize feedback behavior. The machine evaluates the generated actions based on feedback actions and uses the rewards to perform reinforcement learning. For example, a machine on a factory line can perform a given task. The human sees this action and either says "OK" or makes a "X" gesture. The machine recognizes the feedback action and generates a new Evaluate the operation and change the operation based on the evaluation. After the inventor has thoroughly considered the matter, The results of learning or trial and error are not necessarily the best results that humans can think of. Learning results that differ from the diagram may be beneficial to humans, but they may also produce contradictory results. Therefore, humans can provide feedback on the machine's behavior and report that information. By incorporating it as a reward in reinforcement learning, it is possible to produce the best learning results in line with human intentions. This effect was not known in the prior art and is novel. It has a remarkable effect.

[0056] The machine then acquires video of the generated action being performed, and the AI ​​learns from the video. and modifying the behavior based on the acquired information. In this case, the image capture device may be installed in the machine or may exist independently of the machine. The image capture device captures an execution image of the machine performing the generated action. It includes still images and videos. In the case of videos, it is a series of images from before to after the motion is generated. This includes videos. Machines learn using AI based on the videos. For example, the videos are used to train machines to act like humanoid robots. When it comes to the work at hand of the robot, the image is captured by the image capture device of the machine. The images include capturing images of the robot moving around in a large warehouse, When a robot moves using its body or a part of its body, a separate image acquisition device may be used depending on the movement. This allows for the acquisition of data with a greater amount of information about the overall movement of the robot. The first learning method is reinforcement learning. In this case, the machine autonomously evaluates the actions that appear in the video, sets a reward value, and performs a task based on the reward value. Based on this, should you imitate that movement or not, and what part of the whole movement should you imitate? For example, if you fall, you learn to imitate the action. The reward value of is low. Secondly, there is supervised learning and unsupervised learning. Supervised learning is positive learning. The solution is prepared by processing the training data and outputting the correct answer. The machine extracts patterns and features from the image through computational processing. For example, users can select the best machine from among many machine images. The correct answer is given to the video that the child judges to be correct. Based on this information, the machine will give a wrong answer to the video of the person who acted in that way. In the case of unsupervised learning, cluster analysis of images etc. Then, the user can decide whether the cluster is correct or not. The machine learns based on this information.

[0057] We will disclose unsupervised learning. The word "cluster" in cluster analysis means "a bunch" or "a mass" in English. " and refers to a state in which things with similar characteristics are gathered together. The analysis involves grouping similar data (those with similar features) from a large amount of data into several groups. This method is used to create a group of data with similar characteristics, which is called a "cluster." In the case of clustering, classification is performed without any correct answer data. However, the meaning of each group may not be clear. The results are left to humans to interpret. There are types of cluster analysis that are used when there are only a few objects to classify. There are two types of clustering: hierarchical clustering, which is used when there are many objects to be classified, and non-hierarchical clustering, which is used when there are many objects to be classified. Hierarchical clustering is a method for clustering data with similar characteristics. Repeat this process hierarchically until you have one large cluster. The progress is visualized in a diagram like a tournament table (tree diagram), so you can easily see the data. It is an analytical method that makes it easy to grasp the characteristics. It is not necessary to determine in advance how many clusters you should divide the data into, and then Divide the data according to the number. There is also a method where the machine will automatically divide the data without specifying the number. .

[0058] Learning is not limited to the machine that performed the action. The machine also performs the action it generated. The image is acquired, and the image is transmitted to a machine other than the machine that performed the action, and the machine that performed the action Machines other than robots can learn using AI based on the execution video and change their behavior based on the acquired information. As a result of careful consideration by the inventor, it has been found that in the conventional method, each robot is able to function independently. Since it uses machine learning, other robots can learn from the data that the machine has used through trial and error. As a result, it was not possible to improve the collective intelligence of robots in the shortest possible time. According to this method, when multiple machines work, the learning material of one machine is This allows other machines to learn, which synergistically improves the intelligence of the entire machine. This is something that was not known in the prior art, and has novelty and significant effects.

[0059] It may also be implemented as a program that operates any of the machines described herein.

[0060] It has a communication means to send and receive information to and from a server or other machine, and receives requests or This paper describes a method for modifying machine behavior based on training data. In an embodiment, the machine includes a communication means for transmitting and receiving information to and from a server or other machine. As mentioned above, the machine acquires images and the like from other machines or servers, and As a result of that learning, the machine's behavior can be changed. For example, two or more machines can work together to perform a task. In a logistics facility, machine A transports goods to a designated location and then machine B In this case, Machine A requests Machine B to receive the goods. The method of requesting includes sending a request or signal by electromagnetic means. Machine B receives the request. Receive and receive the goods. After the inventor has considered the request or learning data received in good faith, The operation of two or more machines working together can also be studied by changing the operation of the machines based on the It has been found that the behavior of the subject can be improved by learning the language. This effect is not known in the prior art. This is something that has never been done before, and it has novelty and remarkable effects.

[0061] This document discloses a method for providing a means for avoiding collisions with humans, obstacles, or other machines. In at least one embodiment, the machine uses the image captured by the image capture device. to detect obstacles or other machines (hereinafter referred to as "obstacles, etc.") in the machine's direction of travel. If an obstacle is present, the machine will change course or continue moving until the obstacle is removed. In another embodiment, the machine has a device that transmits its location information. This device includes a GPS. The machine transmits its location information to other machines or a server. The machine receives location information of other machines or location information of obstacles, etc. from other machines or servers. The location information of obstacles, etc. is transmitted or used by the machine that detects its presence. This includes a method for users to register their location. The machine will detect obstacles based on the location information it receives from other machines. Change course to avoid collision with obstacles, or temporarily stop moving until the obstacle is gone. According to this embodiment, it is possible to automate a task in which at least two or more machines cooperate with each other. This has the advantages of convenience and industrial applicability.

[0062] A method for providing control means for cooperation with other machines or humans is disclosed. In at least one embodiment, the method described in any of the preceding claims is employed. The image capture device captures images of people around the machine. When there are people around the machine, To avoid collisions with humans, the robot will change course or temporarily halt its movement until the obstacle is clear. It analyzes the video and takes over the work that humans are doing based on the video. For example, According to this embodiment, at least the machine and It has the convenience of enabling humans to work together and has industrial applicability. This technology has the advantage of being convenient and has industrial applicability, as it can automate tasks that require cooperation between robots and humans.

[0063] Equipped with energy efficiency measures to improve the efficiency of one's own energy consumption, It plans its own operations based on calculations, or optimizes motor output or operating speed according to the work content. In at least one embodiment, a method for controlling or charging the battery is disclosed. In this method, the machine sets rewards related to its own energy consumption during reinforcement learning. The machine performs a certain operation and calculates the energy consumption related to that operation. Regardless of the measuring means, the power consumed can be measured. The power measurement means includes a method installed in the machine itself and a method installed in the machine itself. This includes methods that are not installed on the machine itself, but are installed on a supply device that provides power to the machine. The calculation is performed by determining the amount of power consumed during the machine's operation. The difference in power consumption after the supply is calculated. For each operation, the amount of power consumed by that operation is calculated. Reinforcement learning is performed with a reward. The lower the power consumption, the higher the reward. Through reinforcement learning, the machine This will result in less energy-intensive behavior. According to this embodiment, at least the machine work can be performed. The convenience and industrial benefits of reducing power consumption or achieving efficient power consumption are realized. There is a possibility of using.

[0064] A method for transmitting one's own learning data to a machine other than one's own is disclosed. In an embodiment, the method described anywhere in this specification is used. The machine transmits data to a machine or server other than its own. The machine receives learning data related to a machine other than itself. Based on the received learning data, the machine It changes its behavior or uses its training data to learn. In this way, the machine can The ability to inherit machine learning data increases learning efficiency, making it convenient and suitable for industrial use. There is a possibility.

[0065] The contents disclosed in this specification can also be configured as a system.

[0066] In at least one embodiment, agriculture is disclosed. Regardless of whether they are farmers or not, they use the land to grow crops and raise livestock to produce the materials necessary for food, clothing, and shelter. We use the power of the land to cultivate useful plants and raise useful animals, and we engage in organic production. , agricultural processing, forestry, and combinations of one or more of these.

[0067] A method for recognizing crops based on information obtained from one or more sensors is disclosed. In one embodiment, the sensor, whatever its form, collects information about the surroundings of the device. The sensor includes an image capture device. The machine includes a humanoid The machine may have its own sensors, but the sensors may not be able to detect the movement of the machine. It can be separate from the machine. For example, the sensor can be installed on the farm, separate from the machine that is performing the work. The other machine may be equipped with a sensor and the data obtained from the sensor may be The information may be transmitted and received by a machine that does not have a sensor. Analyzing the video: The machine can recognize crops through background estimation or image analysis. For example, the machine can use an image of a plant without tomatoes as a reference and calculate the difference between the acquired image and the image. The machine detects the red image as a difference. The machine stores or retrieves the data. Then, search for what the red image is identical to or similar to. If it resembles a tomato, If it looks like a strawberry, it will be recognized as a strawberry.

[0068] A method for recognizing an environment around a machine is disclosed. In at least one embodiment, the method comprises: The method described anywhere in this specification is incorporated herein by reference. This includes data that can be obtained from variables or physical parameters about the environment surrounding the machine. Examples include LiDAR, cameras, force sensors, and combinations of one or more of these. The sensors include hygrometers and soil moisture meters. Soil moisture meters are devices that measure the amount of moisture contained in soil. As a result of careful consideration by the inventor, it was found that the surroundings of the machine change depending on the season and weather on the farm. Therefore, if the machine moves around the farm under similar conditions, it will be affected by the different conditions. For example, a humanoid robot cannot adapt to muddy ground without falling over. In farms, the strength of the footing is measured by measuring moisture such as a hygrometer or soil moisture meter. The machine can predict the amount of soil moisture underfoot based on the information from the soil moisture meter. This effect was not known in the prior art. , has novelty and remarkable effects.

[0069] We will discuss how to use AI to determine whether or not agricultural work can be performed based on crop and environmental information. In at least one embodiment, the machine provides the crop information to the AI. The information about the product may include an image of the crop. Further, the machine may be equipped with a saccharometer and a hardness meter. The machines recognize the crops around them, use these machines on those crops, and measure their sugar content, hardness, The sugar content meter acquires information on these combinations. It is good if possible. It includes refractometer and non-destructive sugar content. Refractometer measures the refractive index of light that travels straight through water or air. The sugar content is calculated by measuring the refractive index. The non-destructive sugar content meter measures sugar content using a sensor that uses near-infrared light. It takes advantage of the property of sugars that they easily absorb light of specific wavelengths, allowing measurements to be taken without damaging the crops. The hardness measuring instrument may be any type as long as it can measure hardness. The needle is pressed against the sample and the reading is read. As a result of careful consideration, it was found that the use of a sugar content meter, a hardness meter, or a combination of these would be effective. However, this is sufficient to determine whether the crop is ripe. Although they could see that the color had changed, they were unable to determine the texture, taste, or softness of the food. If the machine is equipped with a sugar content meter and a hardness measuring device, it can automatically determine the maturity of the crop. The effect is something that was not known in the prior art and has novelty and significant effects. The environmental information may be any information about the weather, temperature, humidity, images of the environment, or any combination of these. This includes combinations of the above. For example, the temperature is above or below a certain level, or the humidity is above or below a certain level. Or, the following events, a typhoon is approaching, it is raining, or images of plants (e.g. For example, the plant is brown and dying), and combinations of these. Such information may be relevant to or supersede information about crop maturity and may be used to This information can be used to determine whether or not to harvest crops. For example, harvesting crops, not harvesting, fertilizing, watering, etc. This includes feeding, weeding, thinning, or a combination of one or more of these. AI trained using supervised learning, reinforcement learning, unsupervised learning, or a combination of these For example, learning whether to harvest for a combination of the information disclosed herein. For example, the sugar content must be above a certain value and the strength must be below a certain value to be considered a harvest condition. Also, the sugar content may be 10% below the specified value and the strength may be 10% above the specified value, but the plant According to the video, if the plants are dying, it may be time to harvest. The machine can set the conditions for harvesting and perform supervised learning. Even if the harvest is not possible, the data sent from the machine every day is used to harvest the fruit under the conditions that give it the highest sugar content. The inventor has carefully considered the following: By combining machines that collect data and AI that learns, we can perform agricultural work correctly and efficiently. This effect is not known in the prior art and is therefore novel. and has a significant effect.

[0070] When AI determines that an action is possible, it will disclose how to generate that action. In the embodiment, the machine generates an action based on the judgment of the AI. For example, If the machine determines that the crop should be harvested, it will harvest the crop.

[0071] The machine includes a humanoid robot. After careful consideration by the inventor, it has been found that in agriculture, humanoid robots are particularly Robots can be useful. If the inventor has made a good faith decision that the machine is a humanoid robot, The farm was designed to accommodate humans walking on two legs. Because it is designed, there is a possibility that self-driving cars and drones will not be able to operate satisfactorily. The terraced fields are difficult to reach by self-driving car, and drones fly into areas with tangled branches and close to the ground. It is difficult to make a bipedal robot move in such places and in any other places. We found that this is the optimal configuration for machinery to enter farms and harvest crops. That is, humanoid robots have the same or similar range of motion and size as humans, so they can perform the same tasks as humans harvest. It can almost certainly reach plants that can grow, making it the most versatile. Therefore, one of the most desirable forms for providing services was a humanoid robot. The effect is something that was not known in the prior art and has novelty and significant effects. do.

[0072] The motion data related to the past generated motions is learned by reinforcement learning, and the corresponding motions are generated. The crop or the environment that generated the past behavior may be similar or different. A method for generating adaptive behavior in response to an environment is disclosed. The machine learns from the action data of past generated actions by reinforcement learning. For example, if the ground is muddy, the strength and friction of the ground will be different. In the case of such different variables, if you lose your balance and fall, you will not get a reward. Next, another motor is activated, which shifts the center of gravity of the supporting leg, preventing the robot from falling over. In this case, the reward increases, so the operation is considered valid and is incorporated into the control. , and adaptive behavior to an environment similar or different from the environment that generated the past behavior. Suppose the machine is trying to harvest a mandarin orange and accidentally crushes it. In this case, you will not get any reward or it will be negative. Next, we will change the motor strength and get oranges. If the harvest can be made without crushing the plants, the reward will increase, so this operation will be considered effective and incorporated into the control. Next year, when I harvest the mandarins, if I still crush them the way I did in the past, By using reinforcement learning, the machine can harvest the soft oranges of that year without crushing them. According to this embodiment, it is possible to automate at least agricultural work with high quality. This has the advantage of convenience and industrial applicability.

[0073] There are two or more of the above machines in the same field, and the operation or position of one machine in the field is communicated. By doing so, we disclose a method for two or more machines to cooperatively generate actions. In one embodiment, the method described anywhere in this specification is used. When other machines are harvesting crops in one row, the machine will harvest crops in another row. "Two or more machines generate operations in a coordinated manner" means "two or more machines exist in the same field." Based on the information obtained by communicating the operation or position of the machine in the field, This can be replaced with "change the behavior of the machine." In other words, it can be used to automatically change the behavior of the machine based on information obtained from other machines. If the behavior of each group changes, they can naturally coordinate their actions. When harvesting crops, if a harvesting machine comes from the front of the same row, In order to avoid collisions, the robot changes its behavior to move to another row. Convenience and industrial applicability of efficient agricultural work using at least two machines There is a possibility.

[0074] Based on information about the environment around the machine, it determines its path within the field and moves while avoiding obstacles. In at least one embodiment, a method is disclosed, comprising: The method is used. Obstacles include other machines. Information about the environment around the machine includes fallen trees and puddles. This information may be collected from the machine's image capture device, the farm's image capture device, or other The information is generated from the image capture device of the machine. The image capture device of the other machine is, for example, This includes a situation where the camera captures images of the farm. The machine recognizes obstacles through background estimation and image analysis. The machine determines the route it should take in the field based on the information about the obstacles. Identify a route that avoids the obstacle in order to visit all the areas that need to be walked within the venue. After careful consideration, we decided to combine a humanoid robot that performs the work with a drone that captures the images. This system is particularly effective. Robots cannot obtain sufficient information about the field environment. By combining the two, it is possible to adapt to the environment. This effect is not known in the prior art. This is something that has never been done before, and it has novelty and remarkable effects.

[0075] A method is disclosed in which the machine is a walking robot. Use the law.

[0076] The embodiments described herein may also be implemented as a system.

[0077] The embodiments described in this specification may also be implemented as a program.

[0078] The growth status of crops and the presence or absence of disease are determined using image recognition, and harvesting and fertilization are carried out according to the determination. In at least one embodiment, the machine determines the growth status of a crop. For example, the image capture device can be used to capture the pigment variables of crops. If the crop is healthy, the pigment changes. If the crop is diseased, This can be determined by comparing the image with that of a normal plant. If scale insects are visible, it means there are pests. The machine uses this information to make a decision. The AI ​​is able to Supervised learning is performed by labeling the actions taken in the past as correct or incorrect. The AI ​​can learn to harvest, fertilize, control pests, or a combination of these. The AI ​​will decide whether to take the action of combining two or more of these. This includes AI that has undergone supervised learning or reinforcement learning. A unique aspect of this is that the combination of harvesting, fertilization, and pest control can affect the yield and quality of the crop. Therefore, learning based on two or more variables is effective. This effect is not seen in conventional techniques. It is something that has never been known in the art and has novelty and remarkable effects.

[0079] Obtain weather data and generate it based on the weather information, environmental information around the machine, and crop information. In at least one embodiment, a method for identifying an action to be taken using AI is disclosed. Weather data includes data that represent the state of the Earth, regardless of its form. It includes data representing wind, waves, and other phenomena found in the atmosphere and oceans. The current state of the Earth is being observed from space using various sensors and machines. Using current observation data as a starting point, a supercomputer is used to calculate the changes in the Earth over time according to the laws of physics. By calculating according to the above, we can predict the state of the Earth in the future. Weather data includes weather data covering two or more days. The weather data includes weather forecast data for the following days and beyond. The machine itself can Alternatively, weather data transmitted from an external source may be received. As a result of the investigation, it was found that, as a specific situation in agriculture, weather information, environmental information around the machine, and crop information interact with each other. Even if it is judged that fertilization is necessary based on the environmental information around the machine and the crop information, There may be cases where it is judged that fertilization is not necessary based on meteorological information. Even if it is judged based on weather information that harvesting is necessary, it is not judged based on environmental information around the machine. In other words, there may be cases where it is judged that harvesting is not necessary. Even if machine learning or reinforcement learning is performed using only one variable, the machine's behavior may fail due to the presence of other variables. Therefore, learning efficiency is poor. This allows the machine to automatically perform appropriate agricultural operations. This effect is not possible with conventional technology. It is something that was not known before and has novelty and significant effects.

[0080] This document will explain how to change the operation of machinery based on field information acquired by drones. In at least one embodiment, the method described anywhere in this specification is used. For example, drones can collect weather information, environmental information around the machine, crop information, or a combination of these. The drone is equipped with a hygrometer to measure the humidity of the air and a video capture device to capture the The machine detects the presence of fallen trees in the field, approaches the crops and measures the sugar content using a saccharometer. According to this embodiment, the robot changes its behavior based on these variables. By emphasizing machines with different attributes, agricultural automation can be optimized for convenience and industrial There is a possibility of using.

[0081] The machine communicates with a terminal other than the machine and responds to requests from the terminal and decisions made by AI. In at least one embodiment, a method for modifying the operation of the machine is disclosed. In this regard, the methods described elsewhere herein are incorporated by reference.

[0082] The machine is equipped with a lifting mechanism, recognizes the environment around the machine, and controls the movement of the crop based on information about the crop and the environment. In at least one embodiment, the lifting means is This includes any mechanical mechanism that raises one's coordinates to a higher position, regardless of the form. This includes machines and platform climbers. For example, this includes elevators. The machines may The machine recognizes that the crop to be harvested is above it. In that case, the machine uses the lifting means to For example, a humanoid robot can use its knowledge of the environment to determine if the land is muddy or not. If there is no lift, harvest by yourself. If the land is muddy, harvest by other means. The robot communicates with a drone, a machine that harvests the crops. The robot is equipped with a machine that can grab crops with its arms and cut the stalks with a cutting machine to harvest the crops. Since the image capturing means is provided, it is possible to distinguish and recognize the crop and the stem. This would at least provide convenience and industrial benefits such as efficient harvesting of crops at higher elevations. There is a possibility of use.

[0083] When the sensor detects the approach of a human, the audio output unit issues a warning that the device is in operation. A method for issuing an alert is disclosed. In at least one embodiment, the machine The machine detects the approach of a person by using a sensor or an image capture device. When the device is close to the sound output unit, an audio alert will be emitted to warn that the device is in operation. Any sound, regardless of its form, is sufficient. This is because a memorized voice message such as "Working!" can be judged to be a warning. As a result of careful consideration by the inventor, it was found that, due to the unique circumstances of the field, it is difficult to see the plants. Since the field is blocked, there is a risk of collision when humans and machines work together. Furthermore, the machines use blades for harvesting and weeding, and spray toxic chemicals for pest control. In this case, the machine itself would be alerted. In other embodiments, the machine itself that performs the task may be equipped with Even if the audio output unit is not the same as the one that can be used, an alert can be issued from the audio output unit of another machine. The audio output unit may be the earphones worn by the user, or may be an audio output unit provided in the field. This effect was not known in the prior art and is therefore novel and has a significant effect. Has.

[0084] The following will disclose an outline of the above-described embodiment.

[0085] Agricultural machinery, Recognizes crops based on information from one or more sensors, Recognizes the environment around the machine, AI will determine whether or not agricultural work can be performed based on crop and environmental information. machine

[0086] The machine as described above, When the AI ​​determines that an action is possible, it generates that action. machine

[0087] The machine as described above, The motion data related to past generated motions is learned through reinforcement learning, Similar or different to the crop that was the subject of the past action or the environment that generated the past action and generating adaptive behavior for the given behavior or environment. machine.

[0088] A system relating to agriculture, comprising: There are two or more of the above machines in the same field, By communicating the operation or position of one machine in the field, two or more machines can coordinate their operations. Generate, system

[0089] The machine as described above, Based on information about the environment around the machine, the machine determines its route within the field, Move around obstacles, machine

[0090] The machine as described above, The machine is a walking robot, machine

[0091] A system relating to agriculture, comprising: Based on the information obtained by the machine, AI will determine whether or not it can operate. The work machine that performs each operation receives communication from the machine and executes the operation that it determines is possible. system.

[0092] The program that runs the above machine.

[0093] Agricultural machinery, The growth status of crops or the presence or absence of disease is determined using image recognition. Depending on the result, the action of harvesting, fertilizing, pest control, or a combination of one or more of these may be generated. Use AI to determine whether or not machine.

[0094] Agricultural machinery, Obtaining weather data, Based on weather information, environmental information around the machine, and crop information, AI determines the action to be taken. Determine, machine.

[0095] A system relating to agriculture, comprising: Based on the field information acquired by the drone, Alter the operation of any of the above machines; system.

[0096] In the above system, The machine is connected to a device other than the machine, Change the operation of the machine based on the request from the terminal and the judgment by AI. machine.

[0097] Agricultural machinery, Elevating means, Recognizes the environment around the machine, Approaching crops based on crop and environmental information, machine.

[0098] The machine as described above, When the sensor detects the approach of a human, The audio output unit issues an audio alert to warn that the device is in operation. machine.

[0099] The invention according to the present disclosure may achieve at least one of the above-mentioned effects. stomach.

[0100] Terms in this disclosure, including terms in the claims, are defined as described in the specification. The present disclosure is not intended to be limiting unless it is expressly stated otherwise. Anything that has been so named, designated, or understood by one or more citizens in the past, present, or future. At least one embodiment can be interpreted based on the fact that the invention is or can be performed or possibly performed. In this situation, when observing the object or method in question, one or more citizens are reasonably understood or recognized to be within the meaning of the terms set forth herein. Wherever possible, unless inconsistent with the teachings in this disclosure, the terms used herein are It can be judged that the meaning is included in the present invention. or by letting others use the services of others, or by letting general consumers use the services of others, By combining one or more of these actions, all the actions taken together constitute the entire present embodiment. In this specification, when the present invention is implemented in a manner that is not limited to the above, it is considered that the present invention is implemented. The terms in this section include cases where they are used as verb stems. In one embodiment, technical terms, symbols and signs are generally those commonly used in the art. This includes those that have been adopted.

[0101] The following embodiments are, as already explained, examples of the present disclosure. An aspect of the present disclosure may be combined with one or more aspects of another embodiment of the present disclosure, unless inconsistent. They can be combined or substituted.

[0102] Embodiment 01. Robot Tractor In at least one embodiment, the robotic tractor is equipped with a Global Navigation Satellite System (GNSS) and an inertial sensor. The vehicle is equipped with a measuring device and autonomously drives around the field to perform tasks such as plowing, sowing, and harvesting. Equipped with multiple environmental sensors (e.g., LiDAR and cameras), it detects obstacles and grasps the road surface condition. The control unit is based on highly accurate map information and a pre-defined work plan. The system generates a route based on the vehicle's position and posture, and corrects the vehicle's position in real time as the work progresses. The communication module allows remote monitoring and instructions, and automatically stops and notifies the operator when an abnormality is detected. Such autonomous agricultural machinery can significantly reduce the amount of manual driving work required, The work efficiency is improved by more than 20% compared to the conventional method. Furthermore, stable work is possible day and night. This contributes to shortening working hours and reducing the labor required for agricultural work.

[0103] Embodiment 02. Precision weeding robot In at least one embodiment, the precision weeding robot uses a high-resolution camera and AI image recognition algorithms. The robot uses algorithms to distinguish between weeds and crops in real time and performs the minimum necessary weeding. The platform uses GPS and RTK positioning to navigate autonomously between the furrows, with positioning accuracy of just a few centimeters. It moves to a specific weed location in meters. It uses a small mechanical arm as a weeding method. It is equipped with a structure that physically pulls out weeds and a nozzle system that sprays chemicals with pinpoint accuracy. Each work head can adjust to an accuracy of a few centimetres to each individual plant. This robot can access the soil and selectively treat only the weeds while minimizing the impact on the crops. This allows continuous weeding over a wider area of ​​the field, significantly reducing the amount of manual weeding work required. In addition, precise detection and localized treatment using AI can reduce the amount of herbicide used by nearly 90% compared to conventional methods. This reduces the environmental impact and contributes to suppressing the effects of chemicals on agricultural crops.

[0104] Embodiment 03. Fruit and vegetable harvesting robot In at least one embodiment, the fruit and vegetable harvesting robot is a manipulator-type robot. By combining an arm with a multi-wavelength camera, the system automatically determines the ripeness and position of fruits and vegetables for harvesting. Image processing is used to evaluate the color, shape, and size of the fruit in real time, and to determine the degree of ripeness and the optimum time for harvesting. The robot arm is equipped with a special picking hand at the end. It approaches the target fruit at the optimal angle and with the optimal amount of force, cutting or The robot picks the fruit by grasping it. The robot moves on a moving platform or rails. However, it can harvest hundreds of fruits per hour and can be safely used in high places that are difficult for humans to reach or at night. By having multiple robots work together in the field or facility, harvesting efficiency can be improved. This will further improve productivity and reduce quality loss and waste caused by delayed harvesting. The automatic harvesting technology reduces the number of people required for harvesting by approximately 30% and improves the quality of the harvested products. Uniformity and improved yield (about 10% increase) can be expected.

[0105] Embodiment 04: Greenhouse working robot In at least one embodiment, the greenhouse work robot is attached to a rail or ceiling-mounted mechanism within the greenhouse. The robot moves along the path, automatically monitoring and managing the cultivation beds and assisting with harvesting. It is equipped with an environmental sensing module that measures humidity, CO2 concentration, etc., and a high-resolution camera. The control system constantly monitors the plant's growth status and signs of disease. Based on the data, it directs irrigation and ventilation as needed, and works in conjunction with the climate control system to create the optimal growing environment. It also has a robot arm and special nozzles for point spraying of detected pests. The robot can also carry out tasks such as weaving cloth or assisting in picking crops that are ready for harvest. It repeatedly patrols pre-programmed routes, making it difficult for humans to patrol frequently or at night. This will significantly reduce the manpower required for greenhouse management and allow early detection of disease. This contributes to improving the discovery rate (20% or more improvement compared to conventional methods) and stabilizing yields.

[0106] Embodiment 5. Multi-robot cooperative work In at least one embodiment, a multi-robot collaborative work system includes a plurality of autonomous robots. The bots work together via a wireless communication network to carry out tasks in parallel across a vast field. The teams adjust their work areas and roles in real time through a central management server or through mutual communication. Even if some robots stop due to fault detection or battery replacement, other robots will automatically resume operation. It has a fail-safe mechanism that dynamically covers the This includes having another robot do the work that the robot is doing instead. For example, if one machine plows, While a robot is planting, another robot is sowing seeds at a later stage. Robots share sensing data and optimize their movements. The robot behind adjusts the amount of water it sprays based on the soil condition information acquired by the preceding robot. This cooperative operation allows for adaptive processing by utilizing information from each other. Compared to working sequentially, the work time can be reduced by more than half, dramatically improving the work efficiency of the entire field. In addition, flexible scheduling of robots improves machine utilization rates, It is also expected that idle time of equipment will be reduced and costs will be dispersed by using a large number of small robots.

[0107] Embodiment 6: Pesticide spraying by drone In at least one embodiment, the agricultural drone-based pesticide spraying system is an autonomous drone. Drones are used to precisely spray pesticides and fertilizers from above the fields. The drones use GPS navigation and topography. Equipped with a tracking sensor, it follows a pre-set flight path, maintaining a constant altitude and tracking the entire crop. The spray nozzle is controlled by a pressure sensor and a flow control mechanism, and the flight speed is By automatically adjusting the spray amount according to the temperature and altitude, overlapping spraying and scattering of chemicals are minimized. A single drone can cover an area of ​​several tens of hectares per hour, which is much faster than conventional ground spraying. It is possible to treat a wide area in a short time. It is also possible to treat wet rice paddies and steep slopes that are inaccessible to humans. This system allows spraying from inside the building, contributing to improved worker safety and reduced work time. The system optimizes the amount of pesticide used, and spraying according to the target reduces the amount of pesticide used by approximately This can reduce chemical use by 30%, reducing the chemical burden on the environment and reducing costs.

[0108] Embodiment 07. Drone remote sensing In at least one embodiment, the drone remote sensing system is a multispectral Drones equipped with torch and thermal infrared cameras are used to observe the condition of crops and fields from above. The drone flies autonomously over the field in a grid pattern, capturing not only visible light but also near-field light. By acquiring data including infrared and thermal information, the stress state and moisture status of crops can be grasped. Images are transmitted in real time to a ground base station, where AI analysis is used to detect disease spots early. The system performs tasks such as diagnosing the nutritional status and predicting yields. A single flight mission covers tens of hectares. It has higher spatial and temporal resolution than satellite data, and can monitor individual It can detect even the smallest abnormalities in the field without missing anything. The drone's patrol frequency can be adjusted according to the weather. For example, by taking pictures several times a week, the progress of pest damage can be continuously tracked and appropriate control measures can be taken. This kind of aerial sensing technology makes it possible to judge the timing of the situation. It is now possible to detect diseases in their early stages, which were previously difficult to detect with visual inspection, and to This is expected to reduce yield loss by more than 10%.

[0109] Embodiment 8: Cooperative operation of multiple drones In at least one embodiment, the multiple drone cooperative operation system includes a plurality of drones. They will fly simultaneously, dividing up roles and coordinating communications to efficiently monitor and operate large-scale farmland. Each drone is connected via a common communication protocol, and their position and planned route are communicated in real time. By sharing the information, collision avoidance and area sharing can be autonomously adjusted. For example, a leading aircraft can The camera detects the diseased area, and a separate machine following behind sprays pesticides at the exact coordinates. The ground station that controls the entire drone fleet can It monitors the remaining battery power and tank capacity, and sends out replacement units as needed. This system optimizes the operation of each drone, resulting in faster operation than when they are operated independently. The coverage area per unit time has doubled, enabling uniform monitoring and spraying in a short time even in large fields. Furthermore, even if some of the drones break down or fall off due to strong winds, the remaining drones will continue to fly. automatically redistributes the areas covered, providing high reliability and redundancy to the entire system. .

[0110] Embodiment 09. Drone pollination In at least one embodiment, the drone pollination system uses small drones in orchards and greenhouses. This method uses artificial pollination to promote stable fruiting even in environments where natural pollination is difficult. The drones are equipped with fine brushes and air blowers, and fly close to flowers during the flowering period, collecting pollen. Pollination is carried out by spraying pollen directly onto the flower or by contacting the style with a brush containing pollen. The flight path is planned taking into account the position of the flowers, and the entire field is covered in multiple flights. This prevents missed pollination. The sensor and image recognition detect the flowering state and position of the flower, and the opened flower It takes an efficient flight pattern targeting only the pollinators, so it can pollinate effectively while reducing battery consumption. This system can be applied to greenhouse-grown melons and other crops that require artificial pollination, This reduces the need for meticulous manual pollination work and ensures stable pollination even when natural pollinators such as honeybees are in short supply. The introduction of drone pollination has made it possible to secure high yields, particularly in areas that are easily affected by natural conditions. Improved fruit set and timely and reliable pollination increase yields by 15% It is expected that the degree of improvement will be achieved.

[0111] Embodiment 10. Federated learning of agricultural data In at least one embodiment, federated learning is applied to agricultural data analysis. and perform highly accurate machine learning across multiple farms and equipment without sharing data collected from multiple farms and equipment. Each farm uses soil sensor data, weather data, and images of crop growth to build a model. The data is stored in a central database, and only the gradients and parameters required for training are sent to the central model on the cloud. The central server updates the global model using the aggregated parameters. By distributing the model to each farm, each site can access the data of other farms without directly accessing it. For example, this system allows for the development of disease detection models. When constructing the system, it is necessary to create a model equivalent to training data on the scale of tens of thousands of images without disclosing the image data of each field to the outside. This achieves model accuracy, improving detection accuracy by approximately 10% compared to training each farm individually. This method also complies with local privacy and data regulations, and is ideal for breeding and harvesting. It can also be applied to other analytical fields such as volume forecasting. By utilizing learning, data-distributed collaborative learning is realized, and security and privacy are improved. This will enable advanced analysis and decision support using large-scale data while ensuring data security.

[0112] Embodiment 11: Production history management using blockchain In at least one embodiment, a production history management system using blockchain technology It contains various data from agricultural production to distribution (e.g., pesticide spray history during the cultivation period and harvest date and time). Each data entry is time-stamped and digitally signed. Once recorded, the information is added to the blockchain in a way that makes it difficult to tamper with. The farmer automatically uploads the cultivation records from the farm management system. This allows distributors and retailers to access real-time temperature control information and inventory status during shipping and transportation. This creates a complete history for each produce, which is then made publicly available to consumers and inspection agencies. It is possible to access historical information on the blockchain within a certain range, making it easy to verify the origin of the product and its organic certification history. This system dramatically improves traceability in food safety. In the unlikely event of a quality problem, the affected lot can be identified in seconds and recall measures can be implemented. Furthermore, increased transparency of production history will increase consumer trust and boost brand value. Economic effects such as increased value and promotion of transactions at fair prices can also be expected.

[0113] Embodiment 12: Data sharing and verification using blockchain In at least one embodiment, the blockchain-based data sharing platform is used in the agricultural sector. Various sensor data and transaction information can be securely shared among parties, and the integrity of the data can be verified. The sensor node (e.g., soil moisture sensor or meteorological observation device) digitally converts the measurement data. Signatures are added and periodically written to the blockchain network, and agreement is reached between nodes. By forming a consensus, the risk of data tampering or loss is reduced. Farmers and advisory service providers, who are the stakeholders, can refer to the records on the blockchain. The data provided is verified to be reliable, and growth analysis and appropriate agricultural practices are carried out. It can also be used for business guidance. , automatically granting data access permissions under certain conditions and automatically notifying relevant parties when abnormal values ​​are detected. It is now possible to share data between organizations in real time, which previously required time. This allows for consistent data reliability to be ensured even when the sensor installation entity is different. This will result in data-driven collaboration across the agricultural IoT ecosystem, increasing convenience. A sustainable information infrastructure that is both reliable and secure will be built.

[0114] Embodiment 13. Smart Contract Automated Trading In at least one embodiment, automated trading employs smart contracts in agricultural trading. The system will enable the buying, selling and securing of agricultural products based on predefined contract terms on the blockchain. For example, the quality inspection process for fresh vegetable wholesale contracts can be automated. When the data and shipping quantity meet the specified conditions, the smart contract automatically executes the payment. In addition, agricultural insurance linked to weather sensors and satellite data will be implemented. In this case, parameters such as a lack of rainfall for a certain period or a continuation of high temperatures are set as contract catalysts, and conditions are If the conditions are met, the insurance payment will be made immediately without any human intervention. Because it operates on a chain network, all transaction history and contract execution logs are tamper-proof. This will serve as objective evidence in the event of a dispute between the parties. The introduction of the Intrac system has significantly reduced the time and cost required to execute contracts, Payments and insurance claims that previously took days can now be completed in seconds or minutes. This will reduce the risk of delays and payments, improving the reliability and efficiency of the entire agricultural supply chain. do.

[0115] Embodiment 14: AI farming support platform In at least one embodiment, the AI ​​farming support platform uses a variety of data collected from the field. It integrates various data (drone images, soil sensor values, weather data, market price information, etc.) and Machine learning is used to improve decision-making in cultivation management. The cloud-based analysis module uses past cultivation data. A crop growth prediction model is generated based on cultivation history and real-time environmental data, and harvest times and The platform's user interface suggests optimal timing for fertilization and watering. The dashboard visualizes field maps, weather forecasts, and alert information, allowing farm managers to You can intuitively grasp the field conditions from your PC or smartphone. In addition, you can use AI chatbots and voice They are equipped with assistants who provide appropriate agricultural technical advice and troubleshooting in response to questions from farmers. By introducing this platform, we can immediately present the routing procedure. Cultivation decisions that previously relied on experience and intuition are now based on data, improving productivity. This leads to risk reduction. In fact, in some test cases, yields have improved by an average of 10%, and fertilizer and water consumption have been reduced. It has also been reported that the input resources have been reduced (by about 20%), and the effects of precision agriculture based on AI have been quantified. It is shown in a concrete manner.

[0116] Embodiment 15. Disease sign detection system In at least one embodiment, the disease sign detection system is a sensor network deployed in a field. By utilizing network and AI analysis, it automatically detects early signs of crop disease and pest damage. Cameras placed in the field, pest sensors installed in traps, and abnormal chemical changes in the soil By combining multiple detection methods, such as sensors that capture the pattern of disease occurrence (for example, The collected data is then used for edge data analysis. AI models analyze the virus on the device or in the cloud to identify the disease at an early stage before it spreads. Identify risk areas and issue alerts to relevant parties. The alerts include the suspected disease name and recommended control measures. These measures are delivered as smartphone notifications and alerts on the farm management system. This system can prevent the spread of infection in the early stages, which would have been easily overlooked by visual inspections in the past, and By implementing early control, it is possible to halve the area of ​​damage. This makes it possible to model the timing and conditions for disease occurrence, which will also contribute to the development of preventive measures.

[0117] Embodiment 16. Precision irrigation control through environmental sensing In at least one embodiment, a precision irrigation control system utilizing environmental sensing technology The system adjusts irrigation timing based on real-time data from soil moisture and weather sensors in the field. The soil moisture sensors are buried in multiple locations in the field to measure the moisture content of the root zone. It measures the water content and volumetric moisture content and transmits them wirelessly. It also has weather sensors (temperature, humidity, solar radiation, wind speed, etc.) The control unit is placed around the field and is used to estimate evapotranspiration and link with rainfall forecasts. This data is aggregated to maintain the optimum soil moisture range set for each stage of crop growth. For example, it can automatically water only the areas where dryness is detected. On the other hand, if rain is expected, decisions such as stopping irrigation in advance are made based on rules. This precision irrigation significantly reduces water waste. This reduces water consumption by 20 to 30% compared to conventional methods, preventing nutrient loss due to excessive irrigation and reducing crop stress. This is expected to improve yields (by several percent to 10%).

[0118] Embodiment 17. Field Weather Monitoring and Forecasting In at least one embodiment, the field weather monitoring system provides high-precision weather data to the farmland. Observation equipment will be installed to collect micrometeorological data to help with local weather forecasts and risk warnings. The sensor node measures parameters such as temperature, humidity, air pressure, wind direction and speed, solar radiation, and rainfall. This real-time data is then transmitted to the cloud via a wireless network. The data is input into a weather model, and then machine learning techniques using AI are used to analyze the data in a highly localized manner (e.g., 1k The system calculates rainfall forecasts for a grid of 1000m mesh or less and predicts the risk of frost damage. Forecast information is sent to the farm management dashboard and mobile app, so that, for example, when a frost warning is issued, In such cases, an alert is automatically generated to prompt countermeasures such as pre-operating anti-frost fans or installing covering materials. In addition, when heavy rain is predicted, drainage equipment inspections are called for, and when strong winds are predicted, instructions are given in advance to reinforce greenhouses. This system provides specific guidelines for action according to various risks, such as The time to prepare for localized abnormal weather events has been significantly extended (warnings can be issued 1 to 2 days earlier than before). ), contributing to reducing damage to crops caused by weather disasters and ensuring stable production.

[0119] Embodiment 18. Smart Greenhouse Environmental Control In at least one embodiment, the smart greenhouse environmental control system controls the temperature and humidity in the greenhouse. - 24 / 7 monitoring of environmental parameters such as CO2 concentration and light levels, and automatic control of air conditioning and irrigation The equipment is controlled to maintain the optimum cultivation environment. Environmental sensors are installed in various places in the greenhouse. Data from the sensors is collected in a central control unit, which then sets preset target environmental conditions and crop Real-time feedback control is performed based on a control algorithm that corresponds to the growth stage. Specifically, if the temperature rises above the target, the ventilation fan or blackout curtains will be activated. If the humidity is too high, a dehumidifier will be activated. In addition, by taking into account the amount of sunlight and weather forecast information, the system automatically turns on supplementary LED lighting when there is insufficient sunlight, Advanced control strategies have been implemented, including energy-efficient cooling using outdoor air at night. This integrated environmental control provides ideal growing conditions for crops throughout the year. For example, in greenhouse cultivation of tomatoes, yields have increased by more than 10% and energy consumption has decreased by 15% compared to conventional methods. It has been reported that the cost has been reduced. Furthermore, the system's remote monitoring and operation functions have improved management efficiency. This allows for quick response when an abnormality occurs, contributing to stable, high-quality production.

[0120] Embodiment 19. Precision fertilization system In at least one embodiment, the precision fertilization system generates a soil nutrient map of the field and a crop nutrient map. This is a variable fertilization technology that spreads only the necessary amount of fertilizer in the necessary places based on growth information. Before the experiment, soil samples were analyzed and remote sensing was used to measure nitrogen, phosphorus, potassium, etc. in the field. The distribution of fertilizer content is grasped, and a fertilizer prescription for each plot is created in the management software. The fertilizer application machine (a tractor-mounted spraying device or a robot vehicle) uses GPS guidance to move and distribute the fertilizer. The amount of sprayed water is controlled in real time according to the flow rate data installed on the machine. The control valve opens and closes in conjunction with position information, reducing input in high fertility areas and controlling input in deficiency areas. As a result, uneven growth caused by excessive or insufficient fertilization can be eliminated. This will reduce the total amount of fertilizer used and also reduce the burden on the environment, such as nitrogen runoff. In experiments, the introduction of precision fertilization reduced the amount of chemical fertilizer used by approximately 15% while maintaining the same or higher yield. Cases where this has been achieved have been reported, and it is expected to be a technology that achieves both cost reduction and environmental conservation. .

[0121] Embodiment 20. Precision seeding system In at least one embodiment, the precision seeding system uses an automated seed planter to adapt to field conditions. This technology optimizes the seed sowing density and depth depending on the soil fertility map and The field is divided into several management zones based on drainage characteristics, past yield data, etc., and each zone The automatic seed sowing machine calculates the appropriate seed amount (number of seeds per square meter) and the distance between rows and plants. Seeds are sown in precise positions under PS control, and the sowing device mechanism adjusts the speed to match the running speed. The seed discharge rate is variably controlled. Also, a depth control mechanism is provided, allowing for adjustment depending on the hardness and softness of the soil. The seeding depth is automatically adjusted using pressure sensor feedback according to the water content. Precision seeding ensures optimal planting density in each zone, improving germination and initial growth. This leads to a reduction in competition during the growing season. As a result, production efficiency per material increases and the amount of seeds used decreases ( The expected effects include a reduction in crop yield (approximately 10%) and an increase in yield (average of about 5%).

[0122] Embodiment 21. Utilization of satellite remote sensing In at least one embodiment, the satellite remote sensing utilization system is an Earth observation satellite. The wide-area image data acquired by these systems will be incorporated into agricultural production, allowing for understanding of crop growth conditions and regional-scale It is used for environmental monitoring. NDVI (normalized The vegetation index map visualizes the density of vegetation in each field, allowing for early detection of poorly growing areas. This allows for quantitative evaluation of growth levels. Satellite data is also updated relatively frequently, such as weekly. Therefore, it is necessary to regularly monitor the situation across the entire vast agricultural region and assess the impact of abnormal weather on crop yields. The system can evaluate the macroscopic situation of water supply and demand for irrigation water. The system automatically analyzes the data and issues warnings to managers of individual fields as necessary (e.g., when vegetation declines in a specific field). It provides decision support at the regional aggregation level, such as notifying the trend of satellite remote monitoring. Combining ground-based drone sensor data with ground-based sensing provides high-resolution data. It is a multi-layered agricultural system that utilizes both wide-area background information that cannot be covered by data and detailed local information. This method allows for industrial monitoring over a scale of several hundred square kilometers, which was previously difficult. This will enable the collective monitoring of the entire area, allowing for the rapid understanding of the extent of damage in the event of a disaster and the estimation of agricultural production in the entire region. It is expected that the accuracy will improve (approximately 15% improvement).

[0123] Embodiment 22. AR / VR remote farm support In at least one embodiment, a remote viewing experience utilizing AR (Augmented Reality) and / or VR (Virtual Reality) technologies is provided. The farm support system allows experts and managers who are not on-site to remotely check the field conditions and give instructions. On-site workers will be equipped with AR-enabled smart glasses and smartphones. By using this function, field data and work procedure guidance are superimposed on the camera image in real time. Agricultural consultants and engineers in remote locations can view 3D models and live images of fields in a VR environment. Visualize the image and observe and analyze the condition of the crops and soil as if you were there. Two-way communication allows experts to present points and annotations in the field worker's field of view in AR. It can also provide real-time advice by viewing the image of the worker's hands. The introduction of this system will enable advanced agricultural knowledge to be shared even from geographically distant locations. This makes it possible to provide insights to the field, significantly shortening the time it takes to resolve problems. In the event of a disease outbreak, experts can immediately remotely diagnose and provide guidance on countermeasures, which would previously have taken several days. The response can be completed within a few hours, helping to prevent damage from spreading and maintaining productivity.

[0124] Embodiment 23. Machine predictive maintenance In at least one embodiment, the agricultural machinery predictive maintenance system includes a tractor or combine. By attaching various sensors (vibration sensors, temperature sensors, oil pressure sensors, etc.) to agricultural machinery, By constantly monitoring the condition data of the engine and hydraulic system, signs of failure can be detected. The AI ​​model analyzes abnormal vibrations and temperature rise patterns, and detects the need for maintenance before a breakdown occurs. Telemetry data sent from the machine is stored in the cloud and A statistical anomaly detection algorithm evaluates deviations from the normal baseline in real time. If an abnormality is detected, the system sends a warning to the maintenance personnel and This will help prevent work interruptions due to sudden machine breakdowns by suggesting specific inspection items and recommended replacement parts. This can prevent problems before they occur, reducing repair costs and downtime. On farms where this has been done, the average availability of machinery has improved and the number of breakdowns has decreased by more than 30% compared to before. The effects have been reported.

[0125] Embodiment 24. Real-time processing using edge computing In at least one embodiment, edge computing technology is introduced into agricultural IoT systems. By inputting data and processing it in real time near the point of collection, response speed and reliability are improved. The gateway devices and high-performance sensors installed in the fields are locally equipped with CPUs and GPUs. The camera then loads the data onto the computer and performs pre-processing and AI inference on the data before sending it to the cloud. The edge device detects abnormal behavior of cows from the video feed and sends an alert only when an abnormality is detected. Sending data to the cloud helps save network bandwidth and protect privacy. In addition, computers installed on drones and agricultural machinery instantly analyze image data while in flight and analyze the miscellaneous data in the field. It recognizes weed locations on the spot and issues weeding instructions immediately, without worrying about communication delays to the cloud. A hybrid configuration of edge and cloud allows edge processing to be used during normal times. Flexible system operation that ensures autonomy while linking to the cloud for large-scale analysis and model updates. As a result, the overall efficiency of data processing and the response time are improved (control within a few seconds). This achieves robustness by ensuring that important functions continue even in the event of a communication failure.

[0126] Embodiment 25: Agricultural communication network In at least one embodiment, the agricultural communication network system comprises a widely dispersed In order to stably connect sensor nodes and mobile machines, LPWA (Low Power Wide Area) technology is being used. It uses a combination of 5G and low-power technology, and is suitable for large-area farmland with communication distances of several kilometers. We will develop LPWA networks such as LoRaWAN and NB-IoT that can connect soil sensors and weather sensors. These devices can transmit data at low cost and for long periods of time. On the other hand, automated agricultural machinery and real-time video transmission, which require high-speed, low-latency communications, require 5G base stations and platforms. Installing a private LTE network within the farm ensures uninterrupted transmission and reception of large volumes of data and stable remote control. The sensor data is first collected by the gateway device, and then the appropriate network is selected according to the communication situation. This will automatically assign the traffic to the network route (either via LPWA or via the high-speed network). This allows for high-density deployment of IoT devices while maintaining high communication speeds even in environments with limited infrastructure, such as farms. It reduces costs and ensures real-time data transmission. In the field trial of smart agriculture, the loss of sensor data was almost zero, and the remote monitoring response delay of automatic machines was reduced. Operational reliability has improved dramatically, with the number of errors reduced to less than half of the previous level.

[0127] Embodiment 26. Farm Digital Twin In at least one embodiment, the farm digital twin system simulates the actual field environment and crops. The state of affairs is reproduced in virtual space, and the effects of agricultural policies are evaluated through simulation and data analysis. The data collected from each field, such as soil, moisture, weather, and crop growth stage, are also used to evaluate the In addition, a virtual field model is constructed and dynamic simulations of crop growth and water and nutrient circulation are carried out. The model allows for the adjustment of fertilizer and irrigation amounts, changing varieties, shifting planting dates, etc. Various scenarios can be tested, and the results can be used to predict growth, estimate yields, and improve environmental performance. The simulation results are visualized and the farm manager can By comparing and examining these policy proposals, the optimal cultivation plan can be formulated. Improvement measures that would take time and cost to test experimentally on the entire field can be quickly verified in virtual space, This will enable evidence-based decision-making while minimizing risk. As a result of optimizing measures for some crops through simulations using digital twins, It has been reported that water consumption was reduced by 15% while maintaining yields, demonstrating the effectiveness of this technology. is shown.

[0128] Embodiment 27. AI crop planning and market analysis In at least one embodiment, the AI-powered crop planning and market analysis system We comprehensively analyze production data, weather forecasts, market supply and demand trends, etc., and select the crops and The machine learning model helps optimize planting area and shipping time. They are learning how trends and weather affect yield and quality, for example, when demand is likely to increase the following season. It can quantitatively suggest the best crops to grow and the best time to grow them with low risk. Consider multiple scenarios presented by the program (e.g., a plan to increase crop A and decrease crop B), You can choose based on your business goals. The system will calculate the expected income and expenditure based on the proposed plan. The system automatically calculates the amount of materials and labor required, and provides a function to compare the economic efficiency and labor load of each scenario. This kind of AI-based planning support will help reduce the need for planting plans that rely on experience and intuition. This allows you to move away from the conventional thinking and make management decisions backed by data. Benefits include reduced risk of excess inventory and price stagnation, and improved profitability (average profit increase of 5 to 10%). Results can be expected.

[0129] Embodiment 28. Smart livestock management In at least one embodiment, the smart livestock management system utilizes sensors and automation technology to It will be installed in livestock farms to monitor the health of livestock, optimize the breeding environment, and reduce labor. The amount of food and water intake and activity level can be measured using wearable sensors (such as acceleration sensors and heart rate sensors). It acquires biological information such as body temperature in real time and detects abnormal signs (such as estrus or signs of illness). The detection is done using AI. Temperature, humidity, and harmful gas (ammonia, etc.) sensors and cameras are installed inside the barn. The system will be deployed to monitor the animal's comfort and hygiene through environmental sensing and video analysis. The control system automatically controls the ventilation fans and mist spraying devices based on this data. Maintain a comfortable environmental range and provide appropriate food and water in cooperation with feeding robots and automatic watering equipment. Furthermore, by introducing automatic milking robots and robot vacuum cleaners, milking and cleaning work can be simplified. This system realizes labor-saving and unmanned operation. Improved accuracy in health management and early detection of diseases will not only lead to reduced mortality rates and maintained productivity, This will significantly reduce the labor hours required for animal care and management.

[0130] Embodiment 29. AI diagnostic support platform In at least one embodiment, the AI ​​diagnostic support platform uses images of crops and sensors. The system analyzes the data to automatically diagnose pests and growth abnormalities, and provides farmers with specific countermeasures. It allows users to take and send photos of crop leaves and fruits via a smartphone app, and the photos are stored in the cloud. Image recognition AI determines the name of the disease and the extent of the damage. The AI ​​model uses a dataset of more than dozens of pests and diseases. It has been trained on a computer and can detect major diseases (such as powdery mildew and downy mildew) and pest damage with an accuracy of over 90%. Identify the symptoms of the pest and disease, and provide the appropriate pesticides, organic control methods, and countermeasures for the pest and disease. The level of urgency is displayed, and if necessary, contact information for nearby agricultural guidance institutions and related technical documents are provided. The platform also provides a link to the diagnostic data collected from multiple farmers. It also has a function to analyze data, detect regional disease trends and new disease cases, and issue warnings. This AI diagnostic support makes it possible to make time-consuming disease diagnosis decisions that previously required consulting an expert. This allows for immediate on-site identification, and rapid response contributes to preventing damage from spreading and ensuring yields. .

[0131] Embodiment 30. Vertical farming automation system In at least one embodiment, the vertical farming automation system is a multi-tiered plant factory (vertical farming In the field, integrated control of the cultivation environment and robotic technology enable high-density cultivation without human intervention. Each cultivation tray installed on the shelf rack is equipped with a system that monitors temperature, humidity, light intensity, and liquid fertilizer concentration. The system incorporates sensors that allow a central controller to control pumps, fans, LED lighting intensity, and carbon dioxide supply. By automatically adjusting the amount of feed, uniform growth conditions are maintained throughout the entire layer. Robot arms and automated guided vehicles (AGVs) have been introduced to handle everything from sowing to harvesting, sorting, and packaging. For example, a harvesting robot uses AI image processing to determine which leafy vegetables are best harvested. The harvest is then picked by an AGV, which then collects the produce from each shelf and transports it to the packing area. This vertical farm allows for year-round production that is not affected by the outside air or weather. This makes it possible to produce peanuts, with yields per unit area reaching several dozen times that of open-field cultivation, and water usage reduced to 90%. Furthermore, automation of production will lead to significant reductions in labor costs, ensuring stable supply and high productivity. It is expected to be a next-generation agricultural management model that combines high productivity and profitability.

[0132] Embodiment 31. In at least one embodiment, an AI-based smart irrigation control system. This is an irrigation control system that combines AI and a sensor network. Data is collected in real time, and machine learning models calculate the optimal water supply timing and amount. Based on the calculation results, the valves and sprinklers in the field are autonomously controlled, leading to efficient use of water resources and Aim to reduce crop stress.

[0133] Embodiment 32. In at least one embodiment, a group of AI-equipped automatic seeding robots. This is an automatic seeding system in which a group of small seeding robots equipped with AI work together. The robot autonomously navigates within the field using high-precision GPS and sensor-based self-location estimation, ensuring uniform seed distribution. Deep learning image recognition detects missed seeds and overlaps, and mutually By adjusting the work, efficient sowing and improved germination rates can be achieved even in irregular or small fields. .

[0134] Embodiment 33. In at least one embodiment, an AI image recognition type autonomous weeding robot. It is an autonomous weeding robot equipped with a high-resolution camera and AI image recognition technology. The learning model distinguishes between crops and weeds in real time, and the laser is directed only at the weeds. This will significantly reduce the use of herbicides while also reducing the need for manual intervention. This allows for highly accurate weed control in fields with reduced labor.

[0135] Embodiment 34. In at least one embodiment, a pest outbreak prediction AI-linked drone control system Tem. Analyzing sensor data and weather information with machine learning AI to predict the risk of crop pests and diseases in advance Only when it is determined that there is a high risk of an incident, the autonomous drone will fly to the scene. They will be dispatched to the rear and will carry out pinpoint spraying of pesticides and release of natural enemy insects as necessary. By preventing damage through precision pest control, we can reduce the amount of pesticides used, thereby reducing the environmental impact and improving pest control efficiency. We aim to make it more efficient.

[0136] Embodiment 35. In at least one embodiment, an autonomous flying artificial pollination drone. This is a system that uses AI-controlled drones to artificially pollinate plants. The camera and image recognition AI detect the location of the flower, and the drone autonomously flies around the field or greenhouse. By spraying powder or blowing air, pollen is accurately attached to each flower, and natural pollination by bees and other insects is reduced. This ensures stable pollination rates and yields even in harsh environments.

[0137] Embodiment 36. In at least one embodiment, an AI variety selection and crop rotation planning support system. Analyzing local soil conditions, weather data, past production results, and other big data, we develop optimal This is an AI system that proposes crop variety selection and crop rotation plans. Machine learning models predict the growth of each crop. By simulating the risk of disease occurrence and measuring the yield, we aim to maximize yield and maintain soil fertility for each field. This allows for a cultivation plan based on scientific evidence, rather than relying on experience. This will enable the establishment of efficient farm management and contribute to improving sustainability.

[0138] Embodiment 37. In at least one embodiment, an AI greenhouse environment automatic control system. This is a system that uses AI to optimally control the greenhouse environment, including temperature, humidity, light intensity, and carbon dioxide concentration. The system is equipped with sensors and uses a growth model based on machine learning to create a target environment suitable for crops in real time. Depending on the calculation results, ventilation fans, heaters, mist sprayers, blackout curtains, etc. may be installed. The system is automatically controlled to constantly optimize the greenhouse environment in response to changes in day and night and weather. This maximizes yield and improves energy efficiency while reducing manual fine-tuning.

[0139] Embodiment 38. In at least one embodiment, an AI plant factory automatic management system. This is a system that fully automates the cultivation process in a plant factory that uses only artificial light. The environmental sensor and AI control unit monitor the light intensity and quality of the lighting, the nutrient concentration of the culture solution, and the amount of carbon dioxide. Optimize carbon supply in real time. Meanwhile, automated transport robots and robot arms are becoming more widespread. All tasks from seeding to planting and harvesting are carried out unmanned, ensuring stable production throughout the year and significant labor savings. This will enable highly efficient crop production that is not affected by weather conditions.

[0140] Embodiment 39. In at least one embodiment, an AI crop sorting and pre-processing system. This is a system that automatically sorts and processes agricultural products after harvest. Images are taken with a camera and analyzed using deep learning to determine size, color, shape, and degree of damage. After further sorting, the product is washed and peeled by a robot arm. This allows for highly accurate sorting and uniformity without relying on human labor. This allows for one-time processing, contributing to improved product quality and labor savings.

[0141] Embodiment 40. In at least one embodiment, an automated produce conveying system. This is a transportation system that automates the transportation of harvested products on and off the farm. Autonomous tractors and AGVs (automated guided vehicles) use onboard sensors to detect obstacles and transport goods to the warehouse. The vehicle runs autonomously along a designated route. AI controls the optimal route and vehicle formation, ensuring safe and efficient transportation. This will reduce the labor required for transportation after harvest and improve the speed.

[0142] Embodiment 41. In at least one embodiment, a distribution optimization system using AI supply and demand forecasting . It is a platform-based system that optimizes distribution based on supply and demand forecasts for agricultural products. AI integrates and analyzes demand data, growth sensor information, and inventory and price trends to determine harvest times, shipping volumes, and distribution. Dynamically adjusts transportation routes. Harvest timing and logistics are adjusted to eliminate demand surpluses and shortages and maintain freshness. This will contribute to reducing food waste and maximizing profits. Achieve efficiency and profit improvement across the entire value chain, from producers and distributors to consumers.

[0143] Embodiment 42. In at least one embodiment, a weather-linked agricultural work planning AI system. It is an AI system that automatically creates farm work schedules in conjunction with weather data. Based on this, AI calculates the appropriate dates and times for each task, such as sowing, fertilizing, pest control, and harvesting. If extreme weather such as strong winds, heavy rain, or extremely hot days is predicted, measures such as pest control and early harvesting should be taken in advance. Propose work and mitigate damage. Plans that incorporate weather risks improve work efficiency and safety. This will increase the efficiency of the system and contribute to reducing weather-related losses.

[0144] Embodiment 43. In at least one embodiment, an extreme weather response automated protection system. This is an automatic protection system to protect farmland and facilities from extreme weather. When the data detects the occurrence of heavy rain, strong winds, hail, late frost, etc., the AI ​​will quickly activate the protective devices. Specifically, windbreak shelters are deployed when strong winds occur, and automatic covering devices are deployed when hail is forecast. In the event of excessive rainfall, the system will operate drainage pumps. This will minimize crop damage caused by extreme weather and contribute to stable production.

[0145] Embodiment 44. In at least one embodiment, a plant bioinformation feedback control system Hmm. This is a system that monitors crop bioinformation in real time and provides feedback to the cultivation environment. The physiological responses of plants are measured using stem diameter sensors, sap flow meters, and leaf surface potential sensors, and AI collects the data. The system analyzes data to detect signs of stress or poor growth. Based on the detection results, it adjusts the amount of irrigation, fertilizer application, temperature, etc. The system automatically controls temperature and other factors, making immediate adjustments to ensure that individual plants grow under optimal environmental conditions. This will allow for the correction of growth disorders at an early stage, improving yield and quality. .

[0146] Embodiment 45. In at least one embodiment, the AI-controlled agricultural work assistance exoskeleton suit includes a waist - Brushless motors with a peak torque of 30 Nm are built into the knees and shoulders, and IMU (9-axis inertial measurement unit) The robot uses a flexible strain sensor and a robotic arm to monitor the worker's posture and the load on each joint at 10 ms intervals. The AI ​​controller uses a long short-term memory (LSTM) model to learn movement patterns. In repetitive tasks such as transporting seedling trays, the system predicts the joint trajectory and predicts the required auxiliary torque 0.1 seconds in advance. The power source is a removable 48 V lithium-ion pack (capacity 600 Wh), and in normal mode it outputs continuously. It ensures continuous operation for 4 hours, and can be quickly replaced and charged within 15 minutes at the battery station in the field. The suit weighs 9 kg, but it is equipped with a Peltier-type forced air-cooling back plate, so it can withstand the summer heat. Even indoors, the wearer's body surface temperature rise is kept within +1.5 °C. As a safety mechanism, when an overload is detected, The torque is released within 0.05 seconds, and the ratchet brake for fall prevention activates a warning vibration. In addition to issuing a warning, the system also sends an error log to the management terminal via BLE communication. The psoas muscle activity during repeated lifting was reduced by 48% and the subjective fatigue score (Borg S) per work hour was The average improvement in the field size was 3 points, and the working time of elderly workers increased by 1.6 times. By linking with the IoT platform, work data is stored in the cloud, allowing for optimal labor allocation during busy farming seasons. It can be used for optimization and updating personalized assistance parameters of exoskeletons.

[0147] Agricultural machinery, The machine is a humanoid robot, Acquire human agricultural behavior data, AI learns movement data, The machine generates operations based on the learned data, information about the crop, and information about the environment around the machine or the field. accomplish, machine

[0148] Agricultural machinery, The machine is a humanoid robot, Acquire human agricultural behavior data, AI learns movement data, Based on the learned data, weather information, environmental information around the machine or the field, and information on the crop, Generates behavior, machine

[0149] Agricultural machinery, The machine is a humanoid robot, Obtain at least one of the following information: crop information, environmental information, and weather information. Require AI to learn based on the information it obtains, Generate actions based on the learned data and acquired information. machine

[0150] Agricultural machinery that uses AI to determine whether or not agricultural work can be performed based on crop information or environmental information. Judging by The machine is a humanoid robot, and its movements are generated by AI. machine

[0151] The above embodiments may be described in detail with respect to any part or all of the embodiments in this specification. Using Ming as a reference.

[0152] Agricultural machinery, The machine is movable while in contact with land or part of a building, Acquire human agricultural behavior data, AI learns movement data, The machine generates operations based on the learned data, information about the crop, and information about the environment around the machine or the field. accomplish, machine

[0153] Agricultural machinery, The machine is movable while in contact with land or part of a building, Acquire human agricultural behavior data, AI learns movement data, Based on the learned data, weather information, environmental information around the machine or the field, and information on the crop, Generates behavior, machine

[0154] Agricultural machinery, The machine is movable while in contact with land or part of a building, Obtain at least one of the following information: crop information, environmental information, and weather information. Require AI to learn based on the information it obtains, Generate actions based on the learned data and acquired information. machine

[0155] Agricultural machinery that uses AI to determine whether or not agricultural work can be performed based on crop information or environmental information. Judging by The machine can move on land or while attached to a part of a building, and its movements are generated by AI. machine

[0156] The above embodiments may be described in detail with respect to any part or all of the embodiments in this specification. In at least one embodiment, the machine, regardless of its form, Land includes any structure that can be moved on land, land, buildings, etc. This includes parts of objects or structures, and floors of buildings (referred to as "land, etc." in this specification). At least some of the components are movable on land or the like. Includes carts and humanoid robots. The robot moves by contacting the ground with wheels, or the ground with the robot's hands or feet. As already explained, the first feature is covered by the second feature to be subsequently disclosed. The structure in contact with or adjacent to the first feature and the second feature are in direct contact. and an embodiment in which an additional feature is added between the first and second features. and forming the first feature and the second feature such that they do not come into direct contact. That is, the present invention is not necessarily limited to an embodiment in which the soil is in contact with natural soil on land such as a farm field. The condition is that a plastic sheet (corresponding to an additional feature) is laid on top of the field. Even if the building is fixed to the land, it is considered to be on land. Regarding structures adjacent to the facility (mobile cultivation facilities, tents, temporary buildings, etc.), plant factories, etc. The phrase "movable on land" shall be used in conjunction with "" unless a contradiction arises. "Can move while in contact with other objects," "Can move while in contact with objects in the direction of gravity," "On land "Land-based mobility module" and "Land-based mobility module" are interchangeable. It is exchangeable.

[0157] Agricultural machinery, Acquire human agricultural behavior data, AI learns movement data, The machine generates operations based on the learned data, information about the crop, and information about the environment around the machine or the field. accomplish, machine

[0158] Agricultural machinery, Acquire human agricultural behavior data, AI learns movement data, Based on the learned data, weather information, environmental information around the machine or the field, and information on the crop, Generates behavior, machine

[0159] Agricultural machinery, Obtain at least one of the following information: crop information, environmental information, and weather information. Require AI to learn based on the information it obtains, Generate agricultural actions based on the learned data and the acquired information; machine

[0160] Agricultural machinery that uses AI to determine whether or not agricultural work can be performed based on crop information or environmental information. Judging by Generate actions using AI machine

[0161] The above embodiments may be described in detail with respect to any part or all of the embodiments in this specification. Using Ming as a reference.

[0162] The machine as described above, The apparatus is characterized by comprising an image sensor for acquiring data on the human's agricultural activities. A machine.

[0163] The machine as described above, The AI ​​learning of the motion data is performed using a neural network. A machine that does this.

[0164] The machine as described above, The machine, wherein the crop information includes the type of crop and its growth stage.

[0165] The machine as described above, The information on the environment around the machine or the field includes topographical information and meteorological information of the field. A machine that does this.

[0166] The machine as described above, A machine characterized in that it is equipped with a movement mechanism for autonomously moving within a farm field.

[0167] Agricultural machinery, the machine is a land mobility module; Obtaining behavioral data related to agricultural work, Generate actions using artificial intelligence, machine

[0168] The above embodiments may be described in detail with respect to any part or all of the embodiments in this specification. In at least one embodiment, the "behavioral data related to agricultural work" is Regardless of the form, human operation data, machine operation data, or a combination of one or more of these "Behavioral data" can be interchangeably referred to as "operation data" unless a contradiction arises. It is Noh.

[0169] As already explained, it is sufficient for the invention according to the present disclosure to achieve at least one of the effects described above.

Claims

1. This is an agricultural machine that is a humanoid robot equipped with sensors for monitoring the health of crops, analyzes sensor data using AI, determines nutrient deficiencies and disease risks, and automatically performs irrigation or fertilization work based on the results of the determination.

2. This is an agricultural machine that is a bipedal humanoid robot equipped with sensors for determining the quality of crops, and analyzes time-series sensor data using a machine learning algorithm while continuously learning, analyzes it using AI that comprehensively evaluates environmental changes and individual differences in crops, determines the optimum time to harvest, and automatically executes the process.

3. An agricultural method, characterized in that a humanoid robot acquires image data of crops, an AI analyzes the images to determine the presence of disease or pests, and automatically performs control work based on the determination results.

4. An agricultural system characterized by the fact that a humanoid robot collects data on the health status of crops, a central server analyzes the data using AI, generates a nutrient management plan, and the robot carries out tasks based on the plan.

5. 10. The machine of claim 1, wherein the sensor includes a multispectral camera and analyzes the spectral signature of the crop to determine its health.

6. The machine according to claim 2, characterized in that the AI ​​determines the quality grade of the harvested product through deep learning, and simultaneously with the determination, a humanoid robot automatically performs the quality sorting work.

7. 4. The method according to claim 3, wherein the control operation comprises pinpoint application of pesticides or release of natural enemy insects.

8. 5. The system of claim 4, wherein a central server integrates data from multiple fields to generate a region-wide health management strategy.

9. 10. The machine of claim 1, wherein the AI ​​quantifies the stress level of the crop based on sensor data and determines work priorities.

10. 3. The machine according to claim 2, characterized in that a humanoid robot transports the harvested product and records quality data by integrating the crop condition at the time of harvest, quality data, work history, and environmental conditions to ensure traceability.

11. The machine according to claim 1, characterized in that the sensors include at least a near-infrared sensor, a multispectral camera, an RGB camera, and a thermal infrared camera, and by integrating these sensor information with the motion control of a humanoid robot, the machine simultaneously performs real-time environment recognition, object detection, crop condition determination, and autonomous movement control by bipedal walking to generate optimal motion.