Network-based distributed agent system for cooperative problem-solving

Abstract

The proposed invention provides a network-based distributed agent system for cooperative problem-solving, wherein individual agents, which can be present as physical hardware components, cooperatively perform a common task. The agents can be, for example, drones, robots and / or autonomous vehicles. They are managed in a decentralised and autonomous manner and, by means of an interface unit, are given a task which they carry out cooperatively in individual steps. The proposed method is iterative and, after sub-steps have been assessed, the task is adapted such that a final result is obtained after a plurality of iterations. In this case, the individual agents are capable of autonomously interacting and communicating with one another. In this case, a central server unit is not required, but instead the present invention makes use of so-called swarm intelligence. Therefore, for example, some of the agents can be equipped with imaging sensors, which then monitor the work steps and identify potentially cooperating agents and make contact with them. The present invention is likewise directed to an accordingly configured system assembly for carrying out the method and to a computer program product comprising control commands that implement the proposed method or operate the proposed system assembly.

Description

[0001] Network-based distributed agent system for cooperative problem-solving

[0002] The proposed invention creates a network-distributed agent system for cooperative problem-solving, wherein individual agents, which may be physical hardware components, cooperatively accomplish a common task. The agents can be, for example, drones, robots, and / or autonomous vehicles. These manage themselves in a decentralized and autonomous manner and, via an interface unit, are assigned a task which they complete cooperatively in individual steps. The proposed method is iterative; after assessing sub-steps, the task is adapted such that a final result is achieved after several iterations. The individual agents are able to interact and communicate with each other autonomously. No central server unit is necessary here; instead, the present invention utilizes so-called swarm intelligence.For example, some of the agents can be equipped with imaging sensors, which then monitor the work steps and identify and contact potentially collaborating agents. The present invention is also directed to a correspondingly configured system arrangement for carrying out the method, as well as to a computer program product with control commands that implement the proposed method or operate the proposed system arrangement.

[0003] US 2014 / 289733 A1 discloses a system and method for scheduling jobs in a cluster of compute nodes. A job with an unknown resource requirement profile is received. The job comprises a plurality of tasks. The execution of some of the plurality of tasks is scheduled on compute nodes of the cluster with different capability profiles. Timing information regarding the execution time of the scheduled tasks is received. A resource requirement profile for the job is derived based on the received timing information and the different capability profiles. The execution of the remaining tasks of the job is scheduled on the compute nodes of the cluster using the resource requirement profile.

[0004] Known methods work by providing a central server or multiple central servers, which then communicate with the individual agents. However, this solution is error-prone and, in particular, not robust against attacks. Unauthorized third parties can gain access to the server and then disrupt the entire system. Furthermore, sensitive data can be stolen, and in particularly disadvantageous cases, tasks can even be modified in such a way that they contradict the originally intended tasks.

[0005] So-called software agents are also known from the state of the art. These negotiate communication protocols over a network and also perform tasks together. However, these use different technologies and are typically not capable of autonomously identifying potential cooperation partners. For example, image processing or imaging sensors play no role at all here. This refers to the identification of potential cooperation partners as well as the assessment of the final result. Software, therefore, involves cooperation in the purely virtual realm, as distinct from the real-world realm.

[0006] Multi-agent systems (MAS) are well-known collections of autonomous, interactive agents that can collaboratively solve tasks beyond their individual capabilities. In the current state of the art, agents are typically software entities capable of collecting, processing, and acting on environmental information.

[0007] State-of-the-art surveillance systems are known that monitor large events, for example, by directing cameras at specific areas, such as a building. Cameras are mounted on the ceiling so that they can view the area to be monitored. Thus, with the state of the art, a specific area is monitored regardless of the number of agents present there. This unrestricted range of agents delivers larger amounts of data, assuming sufficient computing power is available. This computing power is typically unavailable, especially with mobile devices.

[0008] The state of the art also stipulates that surveillance cameras typically do not carry out any data processing themselves, but rather transmit the captured image data to a remote computing unit. This computing unit typically has extensive hardware capacity. Data is transmitted to at least one server, which then handles the data processing. In this respect, surveillance cameras are typically passive imagers that deliver image data based on their viewing angle. Active control units are located away from these cameras, allowing surveillance personnel, for example, to manually enter a viewing angle and zoom factor for the camera, which is then implemented remotely. The state of the art therefore does not involve the capture and processing of data on a single mobile device. The group of people is typically not restricted here, but rather many people are captured. The people or groups that are always captured are those who are captured.Detects objects that are currently in the camera's field of view.

[0009] Furthermore, it is known from the prior art to identify individual people or objects from a multitude of objects in an image. For example, people can be stored in a database. A photo is taken of these agents, and this photo is then compared with the recorded data. If a match is found, information can then be output indicating that this person is located at a specific location.

[0010] Furthermore, it is known from the prior art for autonomous driving to point a camera at the road ahead and then perform object detection. This makes it possible to read traffic signs and interpret them in such a way that, for example, a speed limit can be read. Typically, there are only a few traffic signs in the field of view of a car's camera, so this does not pose any problems with the underlying hardware. This image information is evaluated within the car, and the technical problem of excessive data volumes thus does not arise. This scenario also typically proposes a passive camera that only records the traffic signs, which are then interpreted.

[0011] The current state of the art faces the problem that a single image source can potentially contain a large number of agents or objects to be recognized. Especially with mobile devices, hardware capacity is limited, and bandwidth is often also limited. Thus, the current state of the art has the disadvantage that large amounts of image data often cannot be transmitted to a remote server. However, if the data is processed within the mobile device itself, this requires extensive hardware capacity. Hardware capacity refers in particular to processor performance and the size and speed of the memory.

[0012] The aforementioned problems are exacerbated by the fact that image processing typically requires real-time processing, meaning that, from the user's subjective perspective, processing must take place immediately. Moving a camera, in particular, generates a large amount of image data, which then needs to be processed immediately.

[0013] It is therefore an object to provide a self-organizing method for distributed and cooperative problem-solving in real-world space that is particularly error-robust. Furthermore, it is an object of the present invention to propose a correspondingly configured system arrangement. Furthermore, it is an object to provide a computer program product with control commands that implement the method or operate the proposed device and arrangement.

[0014] The object is achieved by a method having the features according to patent claim 1. Further advantageous embodiments are specified in the subclaims.

[0015] Accordingly, a method for distributed and cooperative problem solving in real-world space by autonomous agents with automated establishment of communication channels is proposed, comprising providing a real-world task profile; identifying potentially executing agents based on their respective registration information and / or by means of a continuous readout of imaging sensors by known executing agents; comparing the identified, potentially executing agents with the provided task profile, wherein the task profile is broken down into subtasks with selection criteria and the agents provide selection information regarding the selection criteria via a communication channel; selecting agents to be executed based on the comparison of the selection criteria and the provided selection information;Assigning the respective subtask to an agent to be executed via the communication channel; visually recording a result after the respective subtask has been executed by at least one agent; and adjusting the real-world task profile depending on the recorded result.

[0016] The proposed invention advances the state of the art and creates added value by creating a system that integrates concepts into physical agents capable of operating autonomously and cooperatively in the real world. Such systems could have a wide range of applications, from autonomous monitoring and maintenance to search and rescue operations and beyond. Innovations in this area include technologies such as machine learning, distributed computing, and advanced sensor technology. By integrating these technologies, the agents can learn to better understand their environment and respond more effectively to changes. The iterative approach of the proposed invention, in which the agents' actions are adjusted after assessing partial results, is a key element for flexible and adaptive systems.This approach allows agents to learn from experience and refine their approach to solving future tasks.

[0017] Furthermore, the absence of a central server unit in such systems can help improve resilience and increase scalability, as new agents can be added without extensive configuration changes.

[0018] The proposed method is a distributed method because the individual agents are connected to one another via a network. This can be done by having the individual agents have a communication module and thus communicating with one another, for example via mobile communications. Several communication options are conceivable here, such as Bluetooth, WLAN and / or mobile communications. The agents cooperate with one another and are therefore distributed in a network that creates so-called point-to-point network nodes through the individual agents. In general, it is possible for the proposed invention to communicate via central communication units, such as the packet-switched Internet. However, it is particularly advantageous if the individual agents establish point-to-point connections so that a first communication link can be implemented using a different channel, such as a second communication link.For example, a first agent can communicate with a second agent via Wi-Fi, and the second agent can communicate with a third agent via Bluetooth. This creates a fail-safe network that is not based on a single technology. Rather, the individual agents can check which communication methods are available and select a suitable one. If a first communication method fails for any reason, another communication method can simply be used. This way, local disruptions can be bypassed, creating a heterogeneous network that potentially uses multiple communication methods. An agent could, for example, be a pair of VR headsets coupled to a mobile device.The VR glasses can then perform a subtask by providing images of the environment, while the mobile device receives these via Bluetooth and performs the subtask of remote information transmission.

[0019] Problem-free operation generally refers to a real-world space and a real-world task. Thus, the individual agents interact physically, not in virtual space. For example, the task might be to secure an accident scene. For this purpose, a drone could contact an autonomous clearing vehicle. Both the drone and the autonomous vehicle could be equipped with imaging sensors, and it could be recognized that the drone has particular advantage in capturing the geographical space using imaging, allowing the drone to send communication instructions directly to an autonomous vehicle. The autonomous vehicle could also be equipped with a camera system, a radar system, or a lidar system, allowing both systems to visually detect the accident scene in a first subtask.No central control unit is necessary here; rather, in this example, the drone and the autonomous vehicle operate autonomously and share tasks. Optical detection is by no means limited to the visible range. The fact that the problem-solving generally takes place in real-world space does not preclude the possibility of subtasks being performed virtually between agents. It is advantageous for at least two agents to interact with each other in real space.

[0020] The individual agents act autonomously, but are integrated into the network of other agents. Thus, these agents take on autonomous subtasks, and the system as a whole accomplishes the assigned task autonomously. This happens in such a way that the individual agents take on subtasks, and thus the entire system of multiple agents solves the task or problem as a whole.

[0021] For this purpose, potentially executing agents can be identified by a first agent visually scanning the environment and, for example, detecting that a potentially helpful second agent is nearby. This can also be detected by a camera system. This agent can then be contacted and instructed on a subtask.

[0022] In a preparatory procedural step, a task profile is provided, which specifies what is to be implemented and which action steps are necessary for this. The task profile can be created by the agents themselves or transmitted with human intervention. Furthermore, a database can be used to determine which task is necessary in which situation, and thus, from past and stored task solutions, it can be identified what needs to be done in a given situation. The task profile specifies a complex, i.e., compound, task, which is then to be accomplished cooperatively by the agents.

[0023] Potentially executing agents are then identified based on the registration information and / or by means of continuous reading of imaging sensors by known executing agents. The potentially executing agents can therefore either be stored in a database or detected, for example, by a camera system. If, for example, a drone is dispatched over an accident scene, it can use a GPS sensor to read out which other agents are in the immediate vicinity. However, the drone can also use a camera system to detect which potentially executing agents are nearby and contact them. For example, an agent can be provided with a code, for example a barcode or a QR code, and the drone is thus able to identify the corresponding agent.Corresponding contact options can be stored in registration information, or a broadcast can simply be performed, i.e., a widespread, undirected communication, and then a response is waited for. Thus, the underlying process knows all available agents and can begin analyzing the task and contacting appropriate agents who could potentially be helpful.

[0024] To initially start the proposed process, agents, or at least one agent, are tasked with executing the process. This at least one agent can then locate other agents and involve them in solving the task. For example, if a specific location is selected, a drone, an unmanned aerial vehicle, or an autonomous vehicle can be dispatched to this region, and this agent can then request additional agents. The process is therefore initialized by at least one agent, and the cooperating agents are then identified, selected, and tasked. The system thus operates autonomously overall, with each agent autonomously processing its assigned subtask. In general, the proposed process can be initiated by a user, and this task is then delegated to an agent, who then delegates the subtasks further, or additional agents can task each other.This allows all available, potentially executing agents to distribute the subtasks among themselves, and a suitable agent is found for each subtask. If a task or subtask cannot be executed, an attempt is made to break it down further or an alternative, equivalent task is executed. The agents therefore work autonomously and independently find solutions to the task or subtask assigned to them. The task completion processes can be saved and shared among all agents. This creates a knowledge base that can be used in subsequent procedural steps. The process is typically executed iteratively, so that with each new use, the system learns from the result created, and instructions are saved that lead to the most efficient solution to the task.This knowledge base can be accessed by all agents and they can store advantageous solutions.

[0025] In order to find suitable agents, the identified, potentially executing agents are compared with the provided task profile. This means that a check is carried out to determine which of the potentially executing agents would be suitable for implementing the task profile, fulfilling the task, or contributing to it. A number of agents are therefore identified, although not every agent has the appropriate skills to contribute to solving the task. From these potentially executing agents, a number of suitable agents must be selected who will then actually take on subtasks. For this purpose, the task profile is broken down into subtasks, which in turn imply selection criteria for the agents. This means that it is possible to determine which subtasks must necessarily be carried out in order to solve the task using a stored knowledge base, a neural network, and / or a provided metric.to fulfill the task profile. Corresponding criteria are split into parameters. For example, a know-how-related selection can be made, i.e. a selection based on the action knowledge that the respective agent must have. Generally, the ability must also be present. Physically, an agent must be able to perform a specific task. For example, if an accident scene is to be cleared, agents must be available who have sensors with sufficient resolution to visually record the accident scene. Other agents, in turn, must have suitable clearing resources ready. For example, a clearing vehicle can be requested for the task of clearing the accident scene. In addition, agents who are generally suitable for a certain task can also be prioritized. In general, each task also has its own parameters that specify the result, such as a time limit within which the task must be completed.Based on these selection criteria for the subtasks, the suitable agents are identified. For this purpose, the agents provide selection information, and a comparison can be made as to which agents meet the selection criteria and therefore provide the desired selection information. In this process step, the requirements for the respective subtask are compared and the properties of the respective agent are provided, after which a decision can be made as to whether the agent is suitable for this subtask. In a subsequent process step, the agents to be executed are selected from the set of potentially executing agents based on the comparison of the selection criteria and the provided selection information. The agent selected for the corresponding subtask is the one that can best execute it.If the subtask is not performed by an agent, an attempt is made to iteratively break down the subtask and then search for suitable agents. If no suitable agent is found overall, the agent that is most suitable, i.e., the one that comes closest to the desired result or selection criterion, can be selected.

[0026] The respective subtask is then assigned to an agent to be executed via the communication channel. Since subtasks can be specified with any degree of granularity, it is possible for a subtask to be further subdivided into further subtasks, which can then be executed by multiple agents. Ultimately, each agent only needs to be assigned one task to complete, and the overall result must be the fulfillment of the higher-level task.

[0027] The agent(s) then perform their respective subtasks, and at least one agent then performs a visual capture. At least one agent then points its camera system at the relevant location and checks whether the task has been completed. It is also possible to involve multiple agents in the visual capture process, creating a comprehensive picture of the results of the subtasks.

[0028] Based on this recorded result, the real-world task profile is adjusted to determine whether the task has been completed or whether there is remaining work to be done. This remaining work is then saved in a new task profile, and the process begins another iteration, processing this task profile.

[0029] According to the invention, it is advantageous that the control commands implementing the method can be executed distributed among the agents. This allows the commands to be at least partially redundant, and each agent stores and executes only those commands it needs to run the method. A central storage unit is therefore unnecessary. Distributing the code prevents it from being easily tampered with, and the method is more reliable overall, since no central server is operated.

[0030] According to some further aspects of the present invention, the method or the intelligent platform is characterized in that

[0031] • it enables real-time communication between different devices using artificial intelligence;

[0032] • it includes a camera unit that records, evaluates and analyses the situation.

[0033] • the client identifies relevant communication partners and provides them with specific information, instructions or solutions;

[0034] • it enables the networking and interaction of different devices such as drones, autonomous vehicles, robots and electronic glasses;

[0035] • it is designed for use in military strategy, industrial automation, robotics and consumer applications;

[0036] • it enables secure and lossless data transmission;

[0037] • it can be implemented and equipped variably, with regard to specific application areas and / or as a cloud-based solution;

[0038] • geodesics, facial recognition, visual recognition and registration information can be used to identify agents, including human agents and / or machine agents;

[0039] • MR and / or AR glasses can act as agents;

[0040] • a specially designed camera unit is used to meet the specific requirements;

[0041] • it is designed for the areas of tourism, consumer and / or social apps and it can transmit information to smartphones or vehicle media centers;

[0042] • stationary and mobile camera systems transmit information to autonomous vehicles, or the vehicles can exchange maps with each other;

[0043] • it enables communication and interaction between all types of robots, drones, unmanned aerial vehicles, unmanned vehicles and / or electronic glasses;

[0044] • a camera system optionally equipped with a connected tool arm detects repair needs and carries out the repair autonomously based on the analysis and decision-making;

[0045] • the camera recognizes and identifies the user, and the payment transfer is processed directly via the camera system and the user's smartphone; • it enables drones, unmanned aerial vehicles (UAVs), and robots, i.e. agents in general, to adapt to each other's respective technical level. This includes the ability to exchange and synchronize software updates, knowledge databases, and operational algorithms with each other in order to achieve a uniform level of performance and competence within the group. Through this alignment, all participating devices or agents can communicate, coordinate, and collaborate effectively, regardless of individual differences in their original configuration or development stage; and / or

[0046] • it is carried out semi-automatically or fully automatically.

[0047] The present invention relates to an intelligent communications platform that uses artificial intelligence to enable real-time communication between different devices. The platform comprises a camera unit that captures, assesses, and analyzes situations, as well as a powerful AI capable of identifying relevant communication partners and transmitting specific information, instructions, or solutions to them. The communications platform is designed for use in the fields of military strategy, industrial automation, robotics, and consumer applications. It enables networking and interaction between different devices, such as drones, autonomous vehicles, robots, and stationary camera systems, in order to increase the efficiency and accuracy of workflows. A camera unit is used to collect and assess information and situations.Through its sensors, it can perceive various aspects of its environment – ​​such as movement, changes in brightness, or even specific shapes and patterns. This information is then analyzed by the AI ​​and passed on in the form of recommended actions or specific instructions. The platform also enables real-time connection between different devices – they can all be connected and coordinated via the platform. By using encrypted communication protocols, the platform also ensures that data is transmitted securely and without loss. The actual implementation and features of the platform can vary depending on the application. It can contain specific hardware components, be tailored to certain application areas, or be implemented as a cloud-based solution that guarantees maximum scalability and flexibility.

[0048] Some aspects of the present invention are described below: Recognition and identification: When a registered user is recognized by one of the cameras, the AI ​​platform identifies him based on the stored data.

[0049] Information provision: Based on the user's location, preferences and possibly the current context (e.g. weather, time of day), the system sends personalized information directly to the user's smartphone.

[0050] Interaction types: Information can include references to nearby tourist attractions, restaurant or event recommendations, driving directions, current offers at nearby stores, or even alerts about emergencies in the area.

[0051] Real-time communication. The platform enables fast and precise communication between devices in real time, leading to more efficient collaboration.

[0052] Automated decision-making: The platform's AI is able to select relevant communication partners for specific tasks and provide them with targeted information or instructions without human intervention.

[0053] Optimization of workflows. Intelligent communication between devices allows workflows to be optimized, comma errors minimized, and overall performance increased. One technical effect of the invention lies in the integration of AI-supported communication mechanisms into the interaction between different devices. This is achieved by integrating a camera unit for capturing and evaluating situations, as well as a powerful AI for identifying and selecting relevant communication partners.

[0054] Automated information transmission: the AI ​​is able to automatically generate relevant information, instructions or solutions and transmit them to the appropriate devices or users without human intervention.

[0055] Real-time decision-making: through rapid evaluation of situations and immediate communication with the relevant devices in real time, decisions can be made and implemented quickly.

[0056] Adaptive response: Based on ongoing analysis and learning processes, the AI ​​can adapt and continuously improve adaptive responses to meet user requirements and needs. Overall, the technical effect of the invention enables intelligent and effective communication between devices in different application areas, significantly improving the efficiency, accuracy, and performance of workflows. Various systems and protocols can be used to implement the intelligent communication platform to ensure seamless and efficient interaction between the devices. The communication system can be based on a reliable and fast network protocol such as TCP / IP to ensure stable and secure data transmission between the devices.Alternatively, a special lot communication protocol such as MQTT and data transmission in lot environment can be used.

[0057] Data analysis and processing: Systems such as Tensor Flow or PyTorch can be used to process and analyze data and implement machine learning algorithms, providing powerful and efficient AI support. Image processing and recognition: For image processing and recognition to capture situations, an open source library such as OpenCV can be used to equip the camera unit with the necessary functions.

[0058] Cloud Computing Platform: To improve computing power for complex data analysis and small calculations, the platform can be integrated with a cloud computing platform such as AWS, Azure or Google Cloud to ensure scalability and flexibility.

[0059] Security Protocols: To ensure the security of communication and data, appropriate security protocols such as SSL / TLS for encryption and authentication should be implemented to prevent unauthorized access and all that.

[0060] The present invention also relates to a method for the automated establishment of a communication channel using a camera, integrated into a system that utilizes artificial intelligence. The system identifies communication partners based on stored registration information and localizes them using geodesics and visual recognition. The system selects relevant communication partners based on defined criteria such as tasks, technical know-how, and priority. The system then instructs the selected partners by providing information, instructions, and control commands. According to the present invention, the terms machine, robot, humanoid robot, drone, vehicle, glasses, camera, smartphone, arm, smartwatches, electronic wristbands (...) are referred to as agents. The platform describes the claimed method.The method can use the described AI (artificial intelligence), where AI can be understood as a synonym for neural network. Communication between the agents can be language-based, particularly using natural language.

[0061] According to one aspect of the present invention, agents are provided in a preparatory method step, which execute control commands to carry out the method. This has the advantage that agents are initially provided with the necessary information, which then execute the method and subsequently call in additional agents as needed. In this way, swarm intelligence can be utilized, and potentially different angles of an object to be optically detected have been recorded.

[0062] According to a further aspect of the present invention, control commands for implementing the method are executed in a distributed manner across multiple agents. This has the advantage that no central control unit is required and the commands can be stored and executed in a distributed manner. Thus, data storage can also be decentralized and optionally redundant.

[0063] According to a further aspect of the present invention, the agents operate entirely autonomously to implement the method, without a central control unit. This has the advantage of providing fault-resilience because there is no central unit vulnerable to attack, and the distribution also ensures fail-safe operation.

[0064] According to a further aspect of the present invention, the potentially executing agents are stored and / or recognized by at least one other agent using optical information. This has the advantage that contact options, such as a network address or the like, are available and information about the agents' capabilities can already be stored. Based on this information, subtasks can be assigned to agents. The other agents can also be optically recognized in the field and coordinated based on the optical information, i.e., assigned further subtasks. According to a further aspect of the present invention, at least one communication channel is stored for each agent, which is used for communication and if a communication attempt fails, the agent is no longer considered. This has the advantage that alternative contact options or communication media can be stored.This allows agents to establish secure connections, preferably point-to-point. If no response is received from a requested agent after a specified period of time, an alternative agent can be assigned.

[0065] According to a further aspect of the present invention, the method is implemented using at least one neural network. This has the advantage that the selection of agents and the distribution of subtasks can be performed using AI, and completed problem solutions can be used to train the data. This creates a self-learning and self-improving system.

[0066] According to a further aspect of the present invention, the task profile is broken down into subtasks with selection criteria, using a predefined metric and / or a neural network. This has the advantage that the tasks can be broken down into subtasks until a suitable agent is found. Past solutions that have already solved comparable problems can be consulted, or the problems can be broken down using language analysis. This can be achieved, for example, using a taxonomy that breaks a term down into sub-terms.

[0067] According to a further aspect of the present invention, the selection criteria and / or the selection information are read optically, measured, read via an interface, and / or empirically determined. This has the advantage that agents and their capabilities can be automatically recognized. This allows the type of agent to be determined and its capabilities derived from this. The information can also be read via an interface, preferably an air interface.

[0068] According to a further aspect of the present invention, potentially executing agents are identified using image recognition, an optical code, pattern recognition, and / or an image processing method. This has the advantage that agents can be identified using a predefined code, and the information can be used to determine the agent's capabilities. Images of the agents can also be stored, which are compared at runtime, and based on the degree of similarity, the agent can be assigned to a known agent.

[0069] According to a further aspect of the present invention, the optical detection is carried out by at least one optical sensor of at least one agent. This has the advantage that multiple agents can work together and detect from different perspectives.

[0070] According to a further aspect of the present invention, the agents to be executed are selected based on a metric that determines which agent can perform the subtask particularly well. This has the advantage of providing a mapping that indicates which capability is required for each task. Weighting can also be applied here so that the most suitable agent is always selected.

[0071] According to a further aspect of the present invention, registration information comprises at least one image, at least one pattern, license plate information, a facial feature, biometric data, and / or at least one visual user recognition. This has the advantage that the existing communication partners can be compared with a stored image. If there is a sufficient match, it is then assumed that the agent in the stored image is actually located in the captured real-time image. If a communication partner is not a human, a pattern can also be stored, which can be detected particularly easily using image recognition. Parts or an entire license plate can also be stored visually, so that a car can be selected as the communication partner.For example, if you want to send a warning, a driver can enter his license plate information as textual registration information and then other drivers can select exactly this car and the driver of this car will receive a warning on his display.

[0072] According to a further aspect of the present invention, the plurality of communication channels comprises an audio channel, a video channel and / or a textual communication channel. This has the advantage that the agent can select from several formats and thus set the appropriate communication channel. A communication channel can be a textual conversation, i.e. a chat, or an audiovisual message exchange. According to a further aspect of the present invention, the terminal device identifier or agent identifier comprises an address, a telephone number, contact details and / or an alphanumeric character string. This has the advantage that the terminal device can be addressed in different ways and, moreover, different terminal device identifiers can also be present. For example, in addition to a telephone number, an address for video communication can also be stored.The agent can then select how the communication channel should be established.

[0073] According to a further aspect of the present invention, user input to the mobile device occurs by touching a touch-sensitive display unit and / or by voice input. This has the advantage that the user can either select their communication partner on their display or, in a car, make a voice input and then indicate which communication partner should be selected.

[0074] According to a further aspect of the present invention, the user's mobile device is in the form of a telephone, electronic glasses with a user interface, and / or a head-up display. This has the advantage that different hardware can be used. For example, a user can wear electronic glasses that can overlay information onto the lens. This provides the user with different information regarding the communication partners surrounding them. Further information can be used to show which communication partners are registered. Furthermore, it is often possible to overlay a camera image onto a display within the glasses, which then zooms in or out on the communication partners by the desired number.

[0075] According to a further aspect of the present invention, the communication channel is established using Bluetooth, WLAN, an air interface of a mobile network, a point-to-point connection, and / or a telephone connection. This has the advantage that different communication technologies can be used, and if a first communication technology fails, a second communication technology is used. These technologies can also be prioritized, and if a high-priority technology fails, a lower-priority technology is automatically switched on.

[0076] The method describes an advanced system for distributed and cooperative problem-solving in real-world scenarios, combining machine agents with optical sensors, a sophisticated communication infrastructure, and artificial intelligence. The agents operate autonomously and coordinate their tasks using a decentralized architecture that meets real-time requirements while ensuring high fault tolerance. At the heart of the method are optical sensors, whose technical features and interaction with the environment enable precise analysis and efficient problem-solving.

[0077] Optical sensors operate multispectrally and can capture data across multiple regions of the electromagnetic spectrum. These ranges include visible light, infrared, ultraviolet, and terahertz waves. In the visible spectrum, which covers the range from 400 to 700 nanometers, colors, textures, and patterns can be detected, enabling the visual identification and differentiation of objects. Infrared sensors, which cover wavelengths from 700 nanometers to 1 millimeter, detect heat sources and temperature differences. This makes them ideal for use in poor visibility conditions, such as darkness, fog, or smoke, as well as for applications where thermal anomalies need to be analyzed. Ultraviolet sensors, which operate in the range from 10 to 400 nanometers, enable the detection of specific chemical or biological markers that may be relevant, for example, in industrial processes or environmental monitoring.Terahertz waves, which range between 100 gigahertz and 10 terahertz, offer the ability to penetrate materials such as plastic, paper or clothing and to identify hidden objects, such as damage or contaminated areas.

[0078] The sensors feature high resolution, which can be adapted to the requirements of the respective task using adaptive mechanisms. Dynamic resolution allows for precise analysis of both large scenes with low levels of detail and very fine structures. A dynamic range of over 120 decibels ensures that even scenes with extreme brightness differences are reliably captured. The sensors are capable of acquiring data at refresh rates of up to 1,000 frames per second, enabling the analysis of dynamic processes such as object tracking or rapidly changing scenarios. Built-in filters allow for the selective capture of specific wavelengths, while polarization sensors provide additional material information by reducing reflections and analyzing surface properties.

[0079] These sensors are robust against external influences such as shock, vibration, extreme temperatures, and humidity. They can operate in a temperature range of -40°C to +85°C and feature a built-in self-calibration function that ensures their precision is maintained even under challenging conditions. These properties make them ideal for use in extreme environments such as the Arctic, desert regions, or underwater.

[0080] The data collected by the sensors is directly integrated into the agents' decision-making processes. A AI analyzes the data in real time, recognizes patterns, and derives appropriate actions. Machine learning enables the AI ​​to learn from past scenarios and continuously optimize decision-making. The sensors do not operate passively but actively interact with the environment, for example, by analyzing movement patterns of objects or people. This interaction makes it possible to detect and respond to dynamic changes in the environment. The information provided by the sensors is used for navigation, task coordination, and quality assurance.

[0081] The agents communicate via a state-of-the-art telecommunications infrastructure that uses dynamic protocols. This infrastructure supports a variety of communication channels, including cellular (4G, 5G), Wi-Fi, Bluetooth, and satellite-based connections. Switching between these channels is seamless and optimized by AI-driven algorithms that consider factors such as signal strength, latency, and bandwidth. Cellular networks are used for wide-area communication, while Wi-Fi and Bluetooth are deployed in localized networks with low power consumption and high transmission rates. Satellite connections enable reliable communication in remote or infrastructure-poor areas.

[0082] Communication between agents is protected by end-to-end encryption. AI-based security mechanisms monitor communication channels and detect potential threats such as attacks or tampering in real time. These mechanisms ensure the security and integrity of the system, even under critical conditions.

[0083] The system meets real-time requirements through the use of edge computing. Sensor data is processed locally on the agents, minimizing latency. Decisions can be made directly on-site, which is particularly advantageous in time-critical applications such as search and rescue operations. The agents' computing resources are scalable and can be dynamically distributed. For particularly computationally intensive tasks, the agents share their resources and collaborate in a distributed system to efficiently handle the demands. The multi-agent architecture enables scalable collaboration. Each agent operates autonomously but continuously communicates with others to exchange information and coordinate tasks. The optical sensors continuously provide precise data, which is analyzed by the AI ​​and integrated into decision-making processes.This data is crucial not only for the individual navigation and interaction of the agents, but also for the global coordination and adaptation of the system.

[0084] The combination of sophisticated sensor technology, advanced communications, and powerful AI creates a system that is adaptable, robust, and scalable. It is suitable for a wide range of applications, from industrial inspections and environmental monitoring to search and rescue missions and critical infrastructure maintenance. The ability to capture and respond to precise data in real time makes the system a pioneering solution for complex, real-world challenges.

[0085] According to a further aspect of the present invention, the method comprises real-time AI mechanisms that analyze data from distributed sources and make adaptive decisions. This has the advantage of efficiently meeting strict time constraints.

[0086] According to a further aspect of the present invention, the agents use dynamic telecommunications protocols to enable seamless switching between different networks, such as cellular and Wi-Fi. This has the advantage of ensuring that communication channels remain stable under varying network conditions.

[0087] According to another aspect of the present invention, an edge computing model is used in which computing tasks are processed locally by agents. This has the advantage of reducing dependence on central servers and increasing fault tolerance.

[0088] According to a further aspect of the present invention, agents can be equipped with multimodal sensors such as infrared and lidar. This has the advantage of improving decision quality through more precise environmental analyses.

[0089] According to a further aspect of the present invention, the agents utilize neural networks to continuously learn from interactions. This has the advantage of allowing them to dynamically adapt to new requirements. According to a further aspect of the present invention, communication is secured by end-to-end encryption and AI-based anomaly detection. This has the advantage of allowing potential attacks to be detected and repelled.

[0090] According to a further aspect of the present invention, the agents are interoperable with devices of different architectures and software versions. This has the advantage of allowing them to be deployed flexibly in diverse environments.

[0091] According to a further aspect of the present invention, a hierarchical role distribution between agents is introduced, with strategic decisions being made by leading agents. This has the advantage that specialized tasks can be performed more efficiently.

[0092] According to a further aspect of the present invention, a method is proposed that integrates pre-trained neural networks such as transfer learning to adapt the decision-making of autonomous agents. The agents utilize models trained on large datasets and refine them for specific on-site tasks. This has the advantage of reducing training time and minimizing the need for extensive local datasets, increasing system efficiency, especially in resource-limited environments.

[0093] Another aspect of the invention concerns the use of hierarchical reinforcement learning models, in which high-level agents make strategic decisions and low-level agents perform operational subtasks. This has the advantage that complex tasks can be systematically divided into smaller units and processed in an optimized manner, significantly increasing the overall efficiency of the multi-agent system.

[0094] The present invention further envisions the use of Explainable AI to make the agents' decision-making processes transparent. The agents can present their decisions in a comprehensible manner to users and control authorities, which is particularly important in safety-critical applications such as medical diagnostics or transportation. This has the advantage of creating trust in the technology and meeting regulatory requirements for traceability.

[0095] Another technical advancement of the invention is the implementation of post-quantum cryptography, which uses quantum-safe encryption algorithms to fend off future threats from quantum computers. This has the advantage that the agents' communication remains protected over the long term and can be used securely even in highly sensitive areas such as healthcare and transportation systems.

[0096] According to a further aspect of the present invention, blockchain technology is integrated into the system to enable tamper-proof data recording and authentication of agent interactions. This has the advantage of ensuring the integrity of communication while simultaneously enabling the implementation of a decentralized system without central vulnerabilities.

[0097] The present invention also proposes a hybrid cloud architecture that combines edge computing capabilities with global cloud services. Large amounts of data are first preprocessed and optimized locally before being transferred to the cloud. This has the advantage of reducing latency while enabling resource-efficient and scalable data processing.

[0098] Another innovative aspect of the invention is the introduction of swarm intelligence-based autonomous systems, in which multiple agents collaborate in dynamic environments such as rescue operations or industrial maintenance processes. This has the advantage that parallel tasks can be solved more efficiently and the overall system remains robust even under unpredictable conditions.

[0099] According to a further aspect of the present invention, self-healing mechanisms are implemented that use artificial intelligence to detect and automatically correct system errors. This has the advantage of increasing system reliability and significantly reducing maintenance costs through autonomous repair processes.

[0100] The invention also integrates an edge-to-cloud data pipeline, where data is first processed and optimized locally on the agents before being sent to the cloud. This architecture enables real-time local decision-making while simultaneously ensuring higher-level coordination through central models in the cloud. This has the advantage of efficiently combining local and global decision-making.

[0101] Another aspect of the present invention relates to application in medical diagnostics. Here, agents use AI-based image processing to analyze diagnostic images such as X-rays or MRIs in real time and provide results. This has the advantage of supporting physicians, particularly in remote regions, enabling fast and reliable diagnoses. Finally, according to another aspect of the invention, a redundant network architecture based on mesh networks is proposed. Agents can use multiple communication paths and automatically switch to alternative connections in the event of a failure. This has the advantage of maximizing uptime and maintaining system stability even in error-prone environments. These new features significantly expand the original invention and enable an even wider range of robust, efficient, and innovative applications.

[0102] According to a further aspect of the present invention, the function of cameras is only activated when environmental sensors detect and report relevant events or communication partners. This sensor-based preprocessing ensures that image and data processing only takes place when actually necessary, thereby significantly optimizing energy and resource utilization. This has the advantage of avoiding unnecessary camera activations and the associated strain on the data infrastructure, which is particularly crucial in mobile or energy-limited systems such as autonomous vehicles or surveillance systems.

[0103] Another aspect of the invention concerns the assessment of event relevance using an AI-based system. Environmental sensors such as lidar, radar, or acoustic sensors continuously detect physical changes in the environment and send signals to activate the camera only when predefined criteria such as movement, pattern, or intensity are exceeded. This has the advantage of reducing false alarms and ensuring that only critical events are recorded and processed, significantly increasing the efficiency of the entire system.

[0104] Furthermore, according to a further aspect of the invention, multi-sensor decision fusion is introduced, combining different sensor types such as acoustic sensors, lidar, and infrared sensors to increase the precision of event detection. By fusing this sensor data, irrelevant events can be even more accurately distinguished from critical ones. This has the advantage of significantly improving the system's detection accuracy, especially in complex or rapidly changing environments such as search and rescue operations or industrial monitoring.

[0105] Another innovative mechanism is AI-based pattern recognition. Here, neural networks such as convolutional neural networks (CNNs) analyze movement patterns or environmental data in real time to assess the relevance of events. This has the advantage that the system uses contextual information to better prioritize decisions, allowing for even more targeted and precise processing.

[0106] According to a further aspect of the present invention, all preprocessing is realized through edge computing, so that the analyses take place on local devices. This has the advantage of minimizing latency and allowing the system to react to detected events in real time without first having to send data to central data centers. This architecture not only enables faster response times but also reduces network load and the associated infrastructure costs.

[0107] Furthermore, according to a further aspect of the invention, a feedback mechanism is introduced in which feedback from communication partners or from camera analysis is used to dynamically adjust sensor thresholds and Kl parameters. This has the advantage that the system is continuously optimized and can independently adapt to new environmental conditions or application scenarios, which significantly increases the flexibility and performance of the system.

[0108] Parallel to camera activation, the sensors analyze potential communication partners in the environment and identify relevant actors, who are then provided with precise data or instructions. This has the advantage of ensuring targeted communication and allowing the relevant partners to respond immediately to events, thus increasing the effectiveness of the entire system.

[0109] To ensure that the agents are detected, at least one imaging sensor is continuously read to detect a plurality of potential communication partners. The camera of the mobile device is activated and continuously captures an image. It is also possible to use multiple cameras, whose images are combined. Using image recognition, it can then be determined how many agents or objects are in each image. The user aligns the camera and thus captures different numbers of communication partners. These are still potential communication partners in this step. The camera image is therefore a continuous sequence of individual images that is analyzed.

[0110] The user has previously stored a number of potential communication partners, and this exact number is then displayed. To do this, the proposed method automatically adjusts the magnification factor, focal length, or zoom of the imaging sensor. Specifically, it is possible for the user to have pre-specified five potential communication partners, with the agent merely specifying the number. If the agent now points their camera at a group of ten agents, the image is adjusted so that only five agents are in the image. Due to the dimensions of a smartphone screen, it is typically not possible to set the number of potential communication partners higher than ten.

[0111] According to the invention, it has been shown that a number of potential communication partners from 3 to 100 is suitable for use in the present method. Thus, when controlling the camera or the one or more sensors, the screen dimensions are preferably taken into account.

[0112] If the number of potential communication partners is displayed and the agent moves their camera, the process returns to the continuous readout step, where the image section, zoom factor, magnification factor, or focal length is adjusted again. For example, if only three potential communication partners are displayed instead of the desired five, the system zooms out until five potential communication partners are displayed.

[0113] The registration information can also contain characteristics about the potential communication partner. For example, data about the communication partner can be entered using the registration information. Possible parts of the registration information include a location identifier for a potential communication partner. This allows someone to specify that they want to be contacted, but only by agents who have a license plate that corresponds to their own city. The registration information can therefore contain information that the agent only wants to communicate with agents from a specific city. The corresponding service can be expanded as required, and agents can, for example, also configure it so that only potential communication partners with a certain relationship status are displayed. This makes it technically easy to get in touch with other agents, even if only limited screen space is available.By adjusting the imaging sensors, it is ensured that only potential communication partners up to a certain number are displayed.

[0114] A particularly advantageous feature here is that potential communication partners are automatically recognized, and the camera can then also be automatically adjusted. Thus, potential communication partners are displayed without any input from the user. The user simply needs to point their mobile device accordingly, and a suitable image with the communication partners is displayed. Ultimately, the user is responsible for choosing which potential communication partner they want to contact.

[0115] A selection of communication partners is then activated using the provided registration information. If, for example, five people are displayed as potential communication partners, but only three agents have registered for the underlying service, then only these three agents will be activated. Activation here means activation to create a communication channel. A communication channel can therefore be established using these activated agents or objects. It is also possible to limit the number of potential communication partners so that only activated, i.e. registered, agents or objects are displayed. If, for example, five communication partners are displayed, but only three agents have registered, the zoom factor is set so that at least five activated or registered communication partners are in the image to be processed.

[0116] Registration information can be provided by storing user data on a remote server, specifying which agent or object is to be contacted. Contact information can also be stored for this purpose. For example, an agent can register with a photo and then store an address for setting up a video channel.

[0117] The potential communication partner can thus specify which communication channel they would like to use. For example, a telephone number can be stored, or alternatively, a text service that is used. The user who executes the proposed procedure is then shown that a telephone call, text conversation, or chat is possible with a potential communication partner. The potential communication partner can be an agent, who is then contacted using their stored device.

[0118] According to a further aspect of the present invention, the agent systems are modular in design, with communication modules, sensor modules, and computing units being designed as independent, flexibly interchangeable units. These modules can be adapted and expanded depending on the application scenario without making structural changes to the overall system. For example, a communication module in urban smart city applications can switch between 5G for high bandwidth requirements and Wi-Fi for cost-effective local communication, while an infrared sensor module is used for personal location in disaster relief scenarios. This has the advantage that the system is scalable and can dynamically adapt to varying environments and requirements, significantly increasing efficiency and flexibility.

[0119] Another aspect of the present invention concerns the synergistic interaction of the modules. Data processing, sensor analysis, and decision-making are dynamically distributed among the modules, so that agents with higher computing capacity handle complex data analyses, while sensor agents capture environmental conditions in real time. This has the advantage of efficiently distributing the workload, allowing the system to not only respond more quickly to changing conditions but also optimizing energy and computing resources.

[0120] According to a further aspect of the present invention, modular communication modules are configured to support different network protocols such as cellular, Wi-Fi, and point-to-point connections, the selection of which is adaptively adapted to the respective application environment. For example, in an agricultural application, a point-to-point network can be used for drones and ground robots to ensure reliable and energy-efficient communication. This has the advantage of ensuring interference-free communication even in remote areas where no central network infrastructure is available.

[0121] Another technical aspect includes modular sensor units that contain flexibly combinable sensors such as multispectral cameras, lidar modules, and infrared sensors. These enable the agent systems to be optimized for specific applications such as autonomous vehicle navigation, smart city surveillance, or temperature monitoring in smart farms. This has the advantage of ensuring adaptation to a wide variety of scenarios, thus maximizing the functionality and precision of the systems in their respective areas of application.

[0122] According to a further aspect of the present invention, the method is supported by modular computing units that can switch between edge computing modules for real-time analytics and cloud data centers for large-scale data processing. This is particularly useful in smart cities, where traffic analytics are performed in real time using edge computing and long-term planning data is stored in the cloud. This has the advantage of optimally distributing the computing load and increasing the efficiency of the system both locally and globally. According to a further aspect of the present invention, the interaction of the communication, sensor, and computing units creates a synergistic system that enables dynamic task allocation in real time. For example, lidar drones and ground robots in disaster relief can perform real-time area analyses, which are then processed by computing units for coordination and decision-making.This has the advantage that the overall system performance is optimized through the close integration of the modules and even complex challenges can be mastered efficiently.

[0123] Another innovative aspect concerns the combination of virtual and physical modules. In an industrial environment, virtual simulation modules can be used to optimize manufacturing processes before physical robots become active in the real environment. This has the advantage that potential errors and inefficiencies can be identified and corrected during the planning phase, thereby increasing the overall efficiency of the system.

[0124] According to a further aspect of the present invention, high fault tolerance is achieved through redundant modularity. For example, if a communication module fails, alternative network protocols can be activated, or other agents can automatically take over the tasks of failed units. This has the advantage of ensuring the operational continuity of the system even under adverse conditions.

[0125] Finally, the present invention includes module-level security mechanisms, such as the integration of blockchain technology for tamper resistance and AI-based anomaly detection for detecting security breaches. This has the advantage of ensuring both the data integrity and the operational reliability of the modular agent systems at a high level, even in safety-critical application areas.

[0126] According to a further aspect of the present invention, the hybrid cloud solution is integrated as a central technical element into the method for distributed and cooperative problem-solving, combining edge computing and cloud computing in a synergistic manner. The agents utilize locally installed edge computing units for real-time data processing and decision-making, while global coordination tasks and data-intensive analyses are outsourced to the cloud. This has the advantage that latency-critical processes can be processed directly on-site, while the scalability and long-term optimization of the system are ensured through cloud-supported analyses and model training.

[0127] According to a further aspect of the present invention, the agent systems are modular in design, whereby the communication, sensor, and computing units can be dynamically and flexibly interconnected. The communication modules allow the agents to switch between different protocols such as 5G, Wi-Fi, or point-to-point connections depending on network conditions. This enables adaptive and interference-free communication between the agents, even in infrastructure-poor or error-prone environments. Combining these communication modules with the hybrid cloud architecture creates a robust network that enables both local real-time communication and cross-regional data coordination via the cloud. This has the advantage of significantly increasing the reliability and resilience of the agent systems.

[0128] Another aspect of the present invention relates to the synergistic interaction of the modular sensor units with the edge computing units. The agents combine different sensor technologies, such as multispectral optical cameras, infrared, and lidar sensors, for precise detection and analysis of the environment. The sensor data is processed in real time directly on the agents' edge computing units, enabling rapid decision-making in dynamic environments. At the same time, aggregated and optimized sensor data is transferred to the cloud for long-term pattern analysis and model adjustments. This has the advantage that the combination of real-time reactions and global optimization significantly improves the overall performance of the system in demanding application areas such as autonomous driving or disaster relief.

[0129] According to a further aspect of the present invention, decision-making and task allocation within the agent network are controlled by AI-based mechanisms. The hybrid cloud solution enables AI models to be trained at the cloud level, which learn from historical data and optimize strategic decisions. These models are then transferred to the agents' edge units, where they are used for real-time analysis and local decision-making. For example, agents can optimize the allocation of subtasks based on their computing power, energy efficiency, and physical proximity to the deployment site. This has the advantage of dynamically adjusting load distribution within the network and efficiently coordinating agents. Another aspect of the present invention comprises the redundant modularity of the hybrid cloud architecture, which ensures high fault tolerance.Should an edge computing unit fail, neighboring agents can temporarily take over its tasks or outsource data processing to the cloud. At the same time, communication is maintained through adaptive network protocols that automatically switch to alternative connections such as cellular or Bluetooth. This has the advantage of ensuring that the overall system remains stable and functional even in critical situations, which is particularly important in safety-critical applications such as search and rescue operations.

[0130] According to a further aspect of the present invention, the hybrid cloud solution enables continuous improvement of agent AIs through iterative feedback between the edge and cloud levels. While local agents perform tasks and collect results in real time, the resulting data is transferred to the cloud, where it is used to update the underlying AI models. The optimized models are then distributed back to the agents, allowing them to continuously refine their decision-making processes. This has the advantage that the system learns adaptively and continuously adapts to new requirements and environments, increasing the long-term efficiency and performance of the agent systems.

[0131] In summary, the integration of the hybrid cloud solution into the distributed problem-solving process optimizes the synergy between modular communication, sensor, and computing units. The combination of real-time edge computing with cloud-based analytics ensures scalability, fault tolerance, and adaptive decision-making, making the overall system highly robust and flexible in a wide variety of deployment scenarios.

[0132] The method is generally implemented on a computer, with the tasks or subtasks being performed in the real world, i.e., with physical interaction. This does not preclude the possibility that some tasks or subtasks may also be implemented virtually or using software.

[0133] The object is also achieved by a system arrangement for distributed and cooperative problem-solving in real-world space by autonomous agents with automated establishment of communication channels, comprising an interface unit configured to provide a real-world task profile; an identification unit configured to identify potentially executing agents based on their respective registration information and / or by means of a continuous readout of imaging sensors by known executing agents; a comparison unit configured to compare the identified, potentially executing agents with the provided task profile, wherein the task profile is broken down into subtasks with selection criteria and the agents provide selection information regarding the selection criteria via a communication channel;a selection unit configured to select agents to be executed based on the comparison of the selection criteria and the provided selection information; an assignment unit configured to assign the respective subtask to a respective agent to be executed via the communication channel; at least one sensor unit configured to optically detect a result after the respective subtask has been executed by at least one agent; and an adaptation unit configured to adapt the real-world task profile depending on the detected result.

[0134] The problem is also solved by a computer program product with control commands which implement the proposed method or operate the proposed device.

[0135] According to the invention, it is particularly advantageous that the method can be used to operate the proposed devices and units. Furthermore, the proposed devices and units are suitable for implementing the method according to the invention. Thus, each device implements structural features suitable for executing the corresponding method. However, the structural features can also be configured as method steps. The proposed method also provides steps for implementing the function of the structural features. Furthermore, physical components can also be provided virtually or in a virtualized form.

[0136] Further advantages, features and details of the invention will become apparent from the following description, in which aspects of the invention are described in detail with reference to the drawings. The features mentioned in the claims and in the description can each be essential to the invention individually or in any combination. Likewise, the features mentioned above and those further explained here can each be used individually or in groups in any combination. Parts or components with similar functions or that are identical are sometimes provided with the same reference numerals. The terms “left”, “right”, “top” and “bottom” used in the description of the exemplary embodiments refer to the drawings in an orientation with normally legible figure designations or normally legible reference numerals.The embodiments shown and described are not intended to be exhaustive, but rather are exemplary in nature to illustrate the invention. The detailed description is intended to inform those skilled in the art; therefore, known circuits, structures, and methods are not shown or explained in detail in order not to obscure the understanding of the present description. The figures show:

[0137] Figure 1: a schematic flow diagram of the proposed method for distributed and cooperative problem solving according to one aspect of the present invention;

[0138] Figure 2: another schematic flow diagram of the proposed

[0139] A method for distributed and cooperative problem solving according to another aspect of the present invention;

[0140] Figure 3: another schematic flow diagram of the proposed

[0141] A method for distributed and cooperative problem solving according to another aspect of the present invention;

[0142] Figure 4: another schematic flow diagram of the proposed

[0143] A method for distributed and cooperative problem solving according to another aspect of the present invention, which illustrates agent selection; and

[0144] Figure 5: another schematic flow diagram of the proposed

[0145] Method for distributed and cooperative problem solving according to another aspect of the present invention, which clarifies the selection criteria.

[0146] Figure 1 shows a schematic flow diagram of a method for distributed and cooperative problem solving in real-world space by autonomous agents with automated communication channel setups, comprising providing 100 a real-world task profile; identifying 101 potentially executing agents based on their respective registration information and / or by means of a continuous readout of imaging sensors by known executing agents; comparing 102 the identified 101 potentially executing agents with the provided task profile, wherein the task profile is broken down 103 into subtasks with selection criteria and the agents provide selection information regarding the selection criteria via a communication channel 104; selecting 105 agents to be executed based on the comparison of the selection criteria and the provided selection information;Assigning 106 the respective subtask to an agent to be executed via the communication channel; optically capturing 108 a result after the execution 107 of the respective subtask by at least one agent; and adjusting 109 the real-world task profile depending on the captured result.

[0147] Figure 2 shows a further schematic flow diagram of the proposed method for distributed and cooperative problem solving according to another aspect of the present invention. The individual method steps are described as follows:

[0148] Method step 200: an optical recording of the current situation by one or more agents;

[0149] Process step 201: signal processing after all data has been viewed;

[0150] Process step 202: a data analysis is carried out which divides the tasks and analyses available agents and their skills;

[0151] Process step 203: based on the data, it is selected which agents should perform which task or subtask; and

[0152] Process step 204: the individual agents each fulfill their task and thus solve the posed problem or task collectively.

[0153] Figure 3 shows a further schematic flow diagram of the proposed method for distributed and cooperative problem solving according to another aspect of the present invention. The individual method steps are described as follows:

[0154] Process step 300: AI-based communication mechanisms are used to establish a communication channel with agents;

[0155] Process step 301: One or more camera units of agents record the current situation of the task to be solved, and additional agents are recorded; Process step 302: A high-performance AI evaluates the data and records boundary conditions and additional data;

[0156] Process step 303: automated transmission of information to decision-making agents takes place;

[0157] Process step 304: decision-making takes place in real time, for example by negotiation or majority voting of the deciding agents;

[0158] Process step 305: work processes are optimized by assigning individual agents to carry out their tasks; and

[0159] Process step 306: the result is reviewed and the tasks are adjusted if necessary so that a new iteration of the process can start.

[0160] Figure 4 shows another schematic flowchart of the proposed method for distributed and cooperative problem solving according to another aspect of the present invention, illustrating agent selection. The flowchart describes a detailed and methodical process in which artificial intelligence (AI) is used to analyze situations, identify proposed solutions, and make decisions. This process is useful not only for understanding the functionality of AI applications, but also for understanding how AI can contribute to problem solving in practical scenarios. A more detailed description of the individual steps follows:

[0161] AI receives data: The process begins with the AI ​​receiving data from various sources. This data can be in the form of user input, sensor data, database information, or other formats. At this point, the quality and relevance of the data is crucial, as it forms the basis for all subsequent analyses and decisions.

[0162] Analysis of the situation: After receiving the data, the trainer begins the analysis. This analysis involves evaluating the data in the context of the task or problem to be solved. The trainer uses machine learning algorithms to recognize patterns, understand connections, and gain insights relevant to the situation. Identification of solutions / suggestions: Based on the situation analysis, the trainer identifies potential solutions or suggested courses of action. This step often requires the use of complex algorithms capable of drawing conclusions from the available data and generating creative solutions.

[0163] Decision point: At this critical point in the process, a decision must be made as to whether a suitable solution has been found.

[0164] Solution available: If the client has identified a feasible solution, the process continues to communicate this solution.

[0165] No solution: If no suitable solution is available, the AI ​​recognizes that more data is needed to make an informed decision. This initiates a cycle of data collection, after which the process of analysis and solution search is repeated.

[0166] Selection of relevant communication partners: Once a solution has been found, the AI ​​selects the relevant agents with whom the solution should be shared or the necessary actions should be performed. The selection of communication partners depends on the type of solution and the actors involved.

[0167] Transmission of instructions: The selected instructions or solutions are then transmitted to the identified stakeholders. This can be in the form of automated actions, notifications, or reports, depending on the scope of the class.

[0168] Task accomplished?: Finally, it is evaluated whether the measures taken have led to a solution to the problem.

[0169] Yes: If the goal has been achieved, the process is successfully completed.

[0170] No: If the problem persists or the solution hasn't achieved the desired effect, the process moves into an adjustment phase. Here, the instructions are modified based on the feedback received and further analysis to find an improved solution. The cycle of evaluation and adjustment is repeated until the task is completed satisfactorily.

[0171] This iterative process allows AI to flexibly adapt to new information and continuously optimize solutions. It is an example of how AI systems are capable of tackling complex problems by combining data analysis, creative problem-solving, and adaptive learning.

[0172] Figure 5 shows another schematic flowchart of the proposed method for distributed and cooperative problem solving according to another aspect of the present invention, illustrating the selection criteria. The flowchart shows a complex decision-making process performed by an artificial intelligence (AI). It consists of several steps and decision points based on criteria. Here is a detailed description of the process:

[0173] Data analysis and decision-making: The AI ​​begins with the collection and analysis of data. This phase is crucial for gaining a comprehensive understanding of the situation, upon which decisions can be made.

[0174] The class selects communication partners or agents: After analyzing the data, the class selects the communication partners relevant to the situation. This step is divided into several subsections based on specific criteria:

[0175] Technical know-how: Here the class makes a selection based on the technical understanding or expertise necessary for the problem or task.

[0176] Capability: In this branch of the diagram, a selection is made that focuses on the capabilities and suitability of the possible communication partners or agents.

[0177] Priority: This path prioritizes partner selection based on the urgency of the task or problem. Agents or agent types can also be prioritized.

[0178] Task: Here, communication partners are selected based on their relevance or responsibility for the specific task.

[0179] Preparing the information: After selecting the communication partners, the class prepares the necessary information to be communicated. This involves compiling instructions, explanations, or solutions tailored to the situation. Communicating the instructions: The carefully prepared information is then communicated to the selected communication partners. This step is critical because it determines how the solution is communicated and implemented.

[0180] Feedback required?: After communicating the instructions, the class assesses whether feedback is necessary. This may mean waiting for feedback to verify whether the solution has been accepted or whether adjustments are necessary.

[0181] Yes: If feedback is required, the process enters a feedback loop. Here, the feedback received is analyzed, and the instructor adjusts their instructions accordingly. This might mean going back and reevaluating information, collecting additional data, or changing the communication strategy. This step ensures that the approach is iterative and responsive.

[0182] No: If no feedback is necessary because the task is considered completed, the process ends.

[0183] The flowchart illustrates a cyclical and adaptive decision-making process, highlighting the importance of proper selection and preparation of information, as well as the need for feedback to optimize solution finding.

Claims

Patent claims 1. A method for distributed and cooperative problem solving in real-world space by autonomous, machine agents with automated facilities for communication channels, comprising: - providing (100) a real-world task profile; - identifying (101) potentially executing agents based on their respective registration information and / or by means of a continuous readout of imaging sensors by known executing agents; - comparing (102) the identified (101), potentially executing agents with the provided task profile, wherein the task profile is broken down into subtasks with selection criteria (103) and the agents provide selection information regarding the selection criteria via a communication channel (104); - selecting (105) agents to be executed based on the comparison of the selection criteria and provided selection information; - assigning (106) the respective subtask to an agent to be executed via the communication channel; - optically capturing (108) a result after execution (107) of the respective subtask by at least one agent; and - an adjustment (109) of the real-world task profile depending on the recorded result.

2. Method according to claim 1, characterized in that in a preparatory method step agents are provided which execute control commands for carrying out the method.

3. Method according to claim 1 or 2, characterized in that control commands for carrying out the method are executed in a distributed manner among several agents.

4. Method according to one of the preceding claims, characterized in that, in order to carry out the method, the agents act autonomously without a central control unit.

5. Method according to one of the preceding claims, characterized in that the potentially executing agents are stored and / or are recognized by means of optical information from at least one other agent.

6. Method according to one of the preceding claims, characterized in that for each agent at least one communication channel is stored which is used for communication and in the event of a failed communication attempt the agent is no longer taken into account.

7. Method according to one of the preceding claims, characterized in that the method is carried out using at least one neural network.

8. Method according to one of the preceding claims, characterized in that the task profile is broken down into subtasks with selection criteria, using a predefined metric and / or a neural network.

9. Method according to one of the preceding claims, characterized in that the selection criteria and / or the selection information is read out optically, measured, read out by means of an interface and / or determined empirically.

10. Method according to one of the preceding claims, characterized in that potentially executing agents are recognized by means of image recognition, an optical code, pattern recognition and / or using an image processing method.

11. Method according to one of the preceding claims, characterized in that the optical detection is carried out by at least one optical sensor of at least one agent.

12. Method according to one of the preceding claims, characterized in that the selection of the agents to be executed is carried out on the basis of a metric which determines which agent can perform the subtask particularly well.

13. System arrangement for distributed and cooperative problem solving in real-world Space by autonomous agents with automated facilities of Communication channels, comprising: - an interface unit configured to provide (100) a real-world task profile; - an identification unit configured to identify (101) potentially executing agents based on their respective registration information and / or by means of a continuous readout of imaging sensors by known executing agents; - a comparison unit configured to compare (102) the identified (101), potentially executing agents with the provided task profile, wherein the task profile is broken down into subtasks with selection criteria (103) and the agents provide selection information regarding the selection criteria via a communication channel (104); - a selection unit configured to select (105) agents to be executed based on the comparison of the selection criteria and provided selection information; - an assignment unit configured to assign (106) the respective subtask to a respective agent to be executed via the communication channel; - at least one sensor unit configured to optically detect (108) a result after execution (107) of the respective subtask by at least one agent; and - an adaptation unit set up to adapt (109) the real-world task profile depending on the recorded result.

14. A computer program product comprising instructions which, when the program is executed by at least one computer, cause the computer to carry out the steps of the method according to any one of claims 1 to 12.

15. A computer-readable storage medium comprising instructions which, when executed by at least one computer, cause the computer to perform the steps of the method according to any one of claims 1 to 12.

Images