Signal processing module of a platform in the maritime environment

The signal processing module with digital twins and reinforcement learning optimizes tactical decisions for maritime platforms, addressing the reliance on human experience in situational awareness, enhancing tactical planning and computation efficiency.

WO2026238629A1PCT designated stage Publication Date: 2026-11-19ATLAS ELEKTRONIK GMBH +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
PCT/EP2026/063575
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2025-05-15
Filing Date
2026-05-11
Publication Date
2026-11-19

AI Technical Summary

Technical Problem

Current situational awareness prediction in maritime environments relies heavily on human experience, lacking technical means to support commanders in developing tactical strategies.

Method used

A signal processing module that generates digital twins of relevant objects using pre-trained templates, simulates scenarios based on these twins, and employs a meta-model to optimize decisions for mission success, using reinforcement learning to train both primary and meta-models.

Benefits of technology

Enables proactive tactical planning by providing navigational, effector control, and communication information to platforms, enhancing situational awareness and reducing computation time for scenario simulations.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure EP2026063575_19112026_PF_FP_ABST
    Figure EP2026063575_19112026_PF_FP_ABST
Patent Text Reader

Abstract

The invention relates to a signal processing module (20) of a platform (22) in the maritime environment, having the following features: - an input interface (24), which is designed to receive sensor data for creating a positional image (30) or an already created positional image (30); - an output interface (26), which is designed to output information for the platform (22); - a signal processing unit (28), which is designed to generate digital twins (31) for relevant objects in the positional image (30) and, proceeding from the positional image (30), to simulate different scenarios (32) using the digital twins (31); - wherein the digital twins (31) each have a mathematical meta-model which is designed to change a decision of an associated digital twin (31) into a meta-decision if the meta-decision is worse for this digital twin (31) but better for a result of the current scenario (32), or to make a supplementary meta-decision; - wherein the signal processing unit (28) is designed to provide the information for the platform (22) at the output interface (26) on the basis of a selected simulated scenario (32) or a plurality of selected simulated scenarios (32).
Need to check novelty before this filing date? Find Prior Art

Description

[0001] Signal processing module of a platform in a maritime environment

[0002] Description

[0003] The invention relates to the simulation of situational awareness in a maritime environment in order to provide information for a platform in a maritime environment, in particular for naval warfare.

[0004] Situational awareness is crucial for executing missions, such as warfare, both on land and at sea. A situational awareness picture can depict the positions, as well as other characteristics of objects and, in particular, targets. To develop tactics for individual units or a formation—for example, to repel one or more attackers—it is essential to predict how the situational awareness picture will develop. Currently, this prediction requires experienced commanders who formulate tactics based on their experience. However, it would be desirable to support commanders with technical means or to have them develop a proposed tactic.

[0005] The object of the present invention is therefore to create an improved concept for the simulation of future situational awareness scenarios.

[0006] The problem is solved by the subject matter of the independent patent claims. Further advantageous embodiments are the subject matter of the dependent patent claims.

[0007] Exemplary embodiments show a signal processing module for a platform in a maritime environment. The signal processing module comprises an input interface, an output interface, and a signal processing unit. The input interface allows the signal processing module to receive sensor data for creating a situational awareness picture or to receive a previously generated situational awareness picture. If only the sensor data for creating the situational awareness picture is received, the signal processing unit can, for example, generate the situational awareness picture from the sensor data. The output interface allows the signal processing module to output information to the platform. Alternatively, the platform with the signal processing module is also disclosed.

[0008] The signal processing unit creates digital twins, or more precisely, instances of digital twins, for relevant objects in the situational awareness picture. The signal processing unit can access a wide variety of objects in the maritime environment, particularly vessels or aircraft, to generate the digital twins relevant to the current situational awareness picture, for example, through instantiation. Thus, it is possible to generate two digital twins for two identical objects in the situational awareness picture, based on the same template.

[0009] Advantageously, the templates for the digital twins are already pre-trained. The digital twins can also be referred to as agents, especially when trained using reinforcement learning. In principle, however, a digital twin is understood to be a digital representation of an object that is as detailed as possible. That is, the object's properties are mapped in the digital twin. Examples of possible properties of an object include its speed of movement, type and / or number of weapons, turning radius, etc. Based on the properties of the objects, future scenarios can be simulated by modeling physical laws within the simulation.

[0010] Based on the properties of the objects, and thus the digital twins, different scenarios can now be simulated using the digital twins, starting from the situational awareness picture. The scenarios differ primarily in the various decisions made by the digital twins, some of which can be predetermined by an operator or other means. Decisions include, for example, selecting an action such as choosing and / or deploying a weapon, taking defensive measures against an attacking weapon, or requesting support. It is also possible to simulate different boundary conditions, such as changing weather conditions. Furthermore, new situational awareness pictures can be used as the basis for further simulations. A scenario can, for example, end after a predetermined time period or upon achieving a predetermined outcome (e.g., all units of one side eliminated).

[0011] Each digital twin possesses a mathematical meta-model designed to modify a decision made by a corresponding digital twin into a meta-decision if the meta-decision, while detrimental to that digital twin, is more favorable to the outcome of the current scenario. Alternatively, the meta-model can also make a supplementary meta-decision for a digital twin. Both the supplementary meta-decision and the modification of the decision into a meta-decision are rarely made.

[0012] Typically, a maximum of 25%, preferably a maximum of 15%, more preferably a maximum of 10%, or most preferably a maximum of 5% of a digital twin's decisions are meta-decisions. The mathematical meta-model is a model that is advantageously trained using machine learning. However, the meta-model differs from (primary) classical mathematical models in its training method and, consequently, in its perspective on changes in the situational picture during the simulation. Optionally, the digital twin can also have a primary mathematical model to optimize its decisions. This primary mathematical model is typically trained to optimize the decisions for the individual digital twin. Therefore, the primary mathematical model can make the decisions for the scenario simulation unless it is modified, i.e., overruled, by the meta-model.Alternatively, a classic trial-and-error approach can be used to map a majority of possible permutations of the decisions.

[0013] With regard to machine learning of the mathematical model using reinforcement learning, the digital twin can, for example, receive a positive reward for making a good decision and a negative reward for making a bad one. The negative reward is also referred to as a penalty. In a combat scenario between two parties, an individual decision might be to attack an opponent with a weapon. If a weapon is selected that inflicts lasting damage on the opponent or, in a maritime combat scenario, sinks them, the digital twin can receive a positive reward. If the selected weapon is insufficient to inflict lasting damage on the opponent, or if the opponent is missed due to an incorrect setting, this will result in a negative reward for the digital twin.

[0014] The meta-model, however, is advantageously trained not on the individual success of a digital twin, but on the success of a party. To return to the previous example, an attack on the opponent might elicit a response from the opponent that the attacking party cannot reciprocate. For example, the attacked party might have superior firepower or be able to request reinforcements. Although the digital twin's mathematical model would launch the attack because its firepower is sufficient to eliminate an opponent, the meta-model can override the mathematical model's decision and order a different one to avoid jeopardizing the mission's success. For example, the meta-model might order a request for reinforcements before the attack itself. Advantageously, the same meta-model is used on all allied platforms.Unlike the primary mathematical model, the meta-model does not undergo individual adaptation. Besides optimizing for mission success rather than individual platform success, the meta-model can also learn reactions from enemy platforms and thus make predictions.

[0015] Thus, at the output interface, information can be provided to the platform depending on one or more selected simulated scenarios. Specifically, the information is provided to the platform based on the evolution of the situational picture in the different scenarios, particularly in a selected scenario. The selected scenario could, for example, be the one that achieved a desired result most quickly or the one that achieved the best result within a given timeframe.

[0016] The information for the platform can be of various kinds. For example, the platform information can include one or more navigational pieces of information to navigate the platform according to one or more selected scenarios. Advantageously, in this case, the platform is mobile, such as a watercraft or an aircraft. The navigational information can specify or consist of a speed and / or heading for the platform. That is, the output interface provides a suggestion for, for example, the speed and / or heading of the platform using the navigational information. Alternatively, the navigational information can also be used directly to control the platform, for example, a drive and control unit. The navigational information can be extracted from the digital twins of the simulation of the selected scenario.The navigation information that the platform's digital twin adopts during the selected scenario can be transferred to the platform or suggested for the platform's navigation. Furthermore, it is possible to transmit the navigation information of other digital twins to corresponding platforms. This transmission can be carried out, for example, via a communication unit of the platform.

[0017] Furthermore, the platform can include an effector, in particular a weapon. The information for the platform can include information for controlling the effector in order to proactively react to a future situation predicted by a selected scenario. This means that the effector can, for example, adjust itself to an approaching object, specifically align itself, before the object comes within range of a sensor for adjusting the effector. Adjusting or aligning the effector can include, for example, rotating a gun or setting an angle of attack. In particular, the signal processing module can output a sequence of objects in which they come within range of the platform's effector. Additionally or alternatively, it is possible to output the position of the objects at which they come within range of the sensor for adjusting the effector. The sensor for adjusting the effector can, for example, be an infrared sensor.In particular, the platform can feature a variety of effectors, especially a variety of weapons. Information for the platform can be output, for example, a suggested defense sequence, such as preparing a sensor and / or effector chain. The defense sequences can include any selection of actions such as laying mines, preparing missiles, rotating guns, adjusting a gun's angle of attack, deploying an anti-torpedo torpedo, etc.

[0018] The sensor for adjusting the effector can also be referred to as a target tracking unit. A fire control device or fire control system is particularly suitable as a target tracking unit. Depending on the information provided to the platform, the signal processing unit can control a weapon's target tracking unit to prepare it to proactively track one of the relevant objects in the situational awareness picture and engage the target at the earliest possible opportunity. This means the target tracking unit can, for example, adjust to an approaching object, and in particular align itself, before the object even comes within range of the target tracking unit.

[0019] In particular, the signal processing module can output a sequence of objects in which they come within range of the target tracking unit. Additionally or alternatively, it is possible to output the position of the objects at which they come within range of the target tracking unit.

[0020] Additionally or alternatively, the platform can include a communication unit. Depending on the information provided to the platform, the signal processing unit can control the communication unit to request support from a base station, a vessel, or an aircraft in the maritime environment. For example, support can be requested if the scenarios do not yield a satisfactory result.

[0021] Furthermore, the signal processing unit can output information to the platform for coordinating the platform's crew. For example, the signal processing unit can issue the order for the crew to go to battle stations and / or to remain silent, for example, on a submarine. Additionally or alternatively, the instruction to man the medical station and / or weapons can be issued. For this purpose, the signal processing unit can control a platform communication medium, in particular the platform's loudspeaker system, warning systems such as warning lights, or radio-based communication, for example, with individual or all crew members on the platform. It is also possible to suggest a tactic for a selection of platforms, especially friendly platforms, as information to the platform. The selection of friendly platforms can be, for example,The scenario can include a group of manned and / or unmanned watercraft. Furthermore, it is possible to select a multitude of scenarios instead of just one. The variety of possible scenarios can differ, particularly in the opposing party's reaction. Optionally, further scenarios can be simulated based on the resulting situational awareness. Assigning a tactic to each scenario creates a tactics tree. This allows for the proactive preparation of tactics for many eventualities, which can be accessed based on the evolving situational awareness. Optionally, the signal processing unit can control the communication unit to send the tactics to the selected allied platforms.

[0022] Advantageously, the allied platforms and the signal agreement act according to the requested or transmitted tactics. This can be automatic, especially with unmanned and / or autonomous platforms, while manned platforms allow the crew to manually deviate from the tactics.

[0023] One approach is to use a mathematical meta-model that monitors the digital twins for simulating situational awareness and, if necessary, overrides—that is, modifies or corrects—the decisions of the digital twins. It should be noted that simulating situational awareness implies that the digital twins of the objects make independent decisions and react to the decisions of other digital twins. The digital twins can simulate objects within the situational awareness picture and potentially beyond, for example, if a digital twin has requested reinforcements. Simulation differs in this respect from situational awareness prediction, which attempts to determine the most probable situational awareness picture at a predetermined time in the future from a multitude of learned, usually real-world, missions. In other words, prediction involves training a mathematical model to forecast the future situational awareness picture.In the simulation, the properties of objects are mapped to digital twins, allowing each digital twin to act independently. The simulation thus makes it possible, for example, to simulate an opponent's reaction to different decisions or tactics (as a sum of decisions).

[0024] In some implementation examples, the primary mathematical model of the digital twin is trained using reinforcement learning. For reinforcement learning, the positive and negative rewards can be selected based on the specific scenario. Examples include the probability of an underwater vehicle being discovered, the achievement of assigned tasks, reaching a navigational target, etc. It should be noted that a trained primary mathematical model typically improves the decision-making of the digital twins, for example, compared to a mathematical model based on random decisions or trying out numerous permutations of decisions. In any case, it reduces the computation time required to simulate relevant scenarios. The advantage of reinforcement learning over other training methods is that real-world data is not necessarily required for training.For example, untrained digital twins can simulate a variety of scenarios. During the simulation of these scenarios, the digital twins can be trained by positively and negatively rewarding good and bad decisions, respectively.

[0025] In particular, it is possible to consider a large number of decisions for training purposes, for example, all decisions of a scenario together, and to form a resulting sum of positive and negative rewards that is used for training the digital twins.

[0026] In further embodiments, the meta-model is trained using reinforcement learning. Training the meta-model using reinforcement learning is particularly advantageous when the primary mathematical model has already been trained using reinforcement learning. For example, the meta-model can then be trained using the sum or mean of the rewards of the digital twins.

[0027] Further examples show that the digital twins are divided into a party with friendly platforms and a party with hostile platforms. This makes it possible to simulate, for example, attack and defense scenarios in a battle, as well as surveillance tasks, such as monitoring critical infrastructure or a port facility. Advantageously, the meta-model is then trained only using the decisions of the digital twins of the opposing party. However, it is also possible to train the meta-model on the decisions of individual digital twins or the entirety of the opposing party's digital twins in order to predict the behavior of the opposing party's digital twins. For this purpose, the meta-model can be integrated into each digital twin from which it is to learn via an interface.The interface also allows you to specify which information from the digital twins is used to train the meta-model.

[0028] Alternatively, a communication interface can be implemented in the digital twins and the meta-model. Using communication commands, the digital twin can send messages to the meta-model. The responses from the meta-model then return to the digital twin as input variables in the next iteration step. This option has the advantage that an operator, i.e., a human, can also interact with the digital twins and / or the meta-model. That is, the operator can give instructions and / or the digital twins or the meta-model can provide the operator with information, for example, explaining decisions.

[0029] Similarly, a method for providing information to a platform in a maritime environment is disclosed, comprising the following steps: a1) receiving sensor data to create a situational awareness picture and creating the situational awareness picture, or a2) receiving an already created situational awareness picture, b) generating digital twins for relevant objects in the situational awareness picture, c) simulating different scenarios based on the situational awareness picture using the digital twins, each digital twin having a mathematical meta-model, d) changing a decision of an associated digital twin into a meta-decision if the meta-decision is worse for that digital twin but better for a result of the current scenario, e) providing the information to the platform depending on a selected simulated scenario or several selected simulated scenarios.Accordingly, a computer program is disclosed, comprising instructions which, when the program is executed by a computer, cause it to perform the procedure.

[0030] Furthermore, a method for training a digital twin to simulate situational awareness based on an existing situational awareness picture is disclosed. The method comprises the following steps a) to c). In step a), the digital twin for an object of the situational awareness picture is modeled by implementing a primary mathematical model, in particular a mathematical model trained using machine learning, and taking into account the properties of the object. That is, the properties of the object are assigned to the digital twin so that a realistic representation of the object in a virtual environment is achieved, at least with regard to the properties of the object. An optical representation of the object, e.g., for visualizing the simulations, is possible but not required. By means of a simulation that, for example,Running on a graphics engine, the digital twin can perform actions based on physical laws, creating a virtual representation of the object's actions. For example, the digital twin can navigate, evaluate sensor data, and / or operate effectors, particularly weapons. Simulation can be used to determine, for instance, whether a weapon will eliminate an opponent or how the digital twin should navigate to achieve optimal results. During the simulation, the primary mathematical model can make decisions for the digital twin and execute the actions. Advantageously, the primary mathematical model is trained using machine learning methods.

[0031] In step b), the digital twin is trained by positively rewarding positive actions and negatively rewarding negative actions of the respective digital twin in a simulation environment. This means that the digital twin is advantageously trained using reinforcement learning.

[0032] In addition, step c) involves training a meta-model of the digital twin by positively rewarding actions of the digital twin that positively influence the outcome of a scenario and negatively rewarding actions of the digital twin that negatively influence the outcome. The meta-model is trained on the overall success of a party and can therefore intervene if a decision by the primary mathematical model, or the lack thereof, is detrimental to the party's overall success.

[0033] For the sake of completeness, it should be noted that multiple digital twins can be simulated in a single simulation environment. In this case, the steps of the procedure are performed for each digital twin.

[0034] Analogously, a computer program is disclosed, comprising instructions which, when the program is executed by a computer, cause it to perform the procedure.

[0035] In particular, the primary mathematical model, or a multitude of primary mathematical models, can be used to simulate various scenarios within a given situation using artificial intelligence. The artificial intelligence comprises the primary model, which optionally includes a plurality of, in particular, different, sub-models. A (sub-)model describes the architecture of an algorithm. Most common, well-known algorithms are suitable, especially artificial neural networks, Bayesian classifiers, support vector machines (SVMs), and linear regression, but deep learning approaches such as transformer-based artificial neural networks and / or convolutional artificial neural networks are preferred. If the primary model comprises several sub-models, similar algorithms (e.g., different artificial neural networks) as well as different algorithms (e.g.,An artificial neural network and linear regression can be combined. Submodels can optionally be used, for example, for communication and / or the use of weapons and / or the navigation of the digital twins in the simulation. In particular, it has been found that no special algorithms need to be used to create the primary model. It is sufficient to train commercially available algorithms with appropriate data. Such algorithms are also referred to as COTS (Components-Off-The-Shelf). It has been found that for reinforcement learning agents, the exact network architecture, i.e., the number of neurons or the depth of the layers, is largely irrelevant to the learning success, as long as a minimum level of model complexity is achieved.Agents that achieve excellent results in public reinforcement learning benchmarks are particularly well-suited. Examples include the Atari 2600 games in the Arcade Learning Environment (ALE) of OpenAI Gym or the continuous control tasks (e.g., Hopper, Walker) in the MuJoCo simulator. Model examples that perform strongly in these benchmarks are based on algorithms such as DQN (Deep Q-Network), PPO (Proximal Policy Optimization), or SAC (Soft Actor-Critic).

[0036] The following revelations contain exemplary descriptions of such models:

[0037] Volodymyr Mnih, Koray Kavukcuoglu, David Silver et al - Human-level control through deep reinforcement learning. Nature, 2015.

[0038] John Schulman, Filip Wolski, Prafulla Dhariwal et al - Proximal Policy Optimization Algorithms. arXiv: 1707.06347 [cs.LG], 2017.

[0039] Tuomas Haarnoja, AurickZhou, Pieter Abbeel u. a. - Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor.

[0040] arXiv: 1801.01290 [cs.LG], 2018.

[0041] In particular, the meta-model allows the simulation of various scenarios using artificial intelligence to be monitored within a predefined situational picture and optimized through user-defined decisions. The artificial intelligence comprises the meta-model, which optionally includes multiple, and especially different, sub-models. A (sub-)model describes the architecture of an algorithm. Most common, well-known algorithms are suitable for the meta-model, especially artificial neural networks, Bayesian classifiers, support vector machines (SVMs), and linear regression. However, deep learning approaches such as transformer-based artificial neural networks and / or convolutional artificial neural networks are preferred. If the primary model comprises several sub-models, similar algorithms (e.g., different artificial neural networks) as well as different algorithms (e.g.,An artificial neural network and linear regression are combined. Submodels can optionally be used, for example, for communication and / or the use of weapons and / or the navigation of the digital twins in the simulation.

[0042] The same algorithms are suitable for the meta-model as for the primary mathematical model.

[0043] Preferably, at least one (partial) model of the primary model and / or the meta-model is implemented as an artificial neural network. The artificial neural network can, for example, be based on Transformer technology and / or be or comprise a convolutional artificial neural network.

[0044] Preferred embodiments of the present invention are explained below with reference to the accompanying drawings. They show:

[0045] Fig. 1: a schematic representation of a signal processing module.

[0046] Before exemplary embodiments of the present invention are explained in detail below with reference to the drawings, it should be noted that identical, functionally equivalent or equivalent elements, objects and / or structures in the different figures are provided with the same reference numerals, so that the description of these elements shown in different exemplary embodiments is interchangeable or can be applied to one another.

[0047] Fig. 1 shows a schematic diagram of a signal processing module 20 of a platform 22, here exemplified by a manned or unmanned ship, in a maritime environment. The signal processing module 20 comprises an input interface 24, an output interface 26, and a signal processing unit 28.

[0048] The input interface 24 can receive sensor data to create a situational awareness picture or, as shown here, an already created situational awareness picture 30. The output interface 26 can output information for the platform 22.

[0049] The signal processing unit 28 generates digital twins 31 for relevant objects in the situational awareness picture 30. The circles in the situational awareness picture 30 can symbolize the individual objects of the situational awareness picture 30. Based on the situational awareness picture 30, different scenarios 32, here purely as examples the three scenarios 32a, 32b, 32c, are simulated using the digital twins 31. The result of a scenario can be a simulated situational awareness picture resulting from decisions made by the digital twins 31 in the simulation. Typically, more than 10, preferably more than 100, and even more preferably more than 1000 scenarios can be simulated. The number of digital twins 31 can vary in the scenarios, for example, by eliminations or by requesting reinforcements.

[0050] The digital twins 31 can each have a mathematical meta-model and optionally a (primary) mathematical model. The mathematical meta-model can change a decision of an associated digital twin into a meta-decision if the meta-decision is worse for that digital twin 31, but better for an outcome of the current scenario.

[0051] Additionally or alternatively, the meta-model can make a supplementary meta-decision. However, the digital twin 31 makes its regular decisions based on the mathematical model or using another logic, for example, randomly.

[0052] Depending on one or more selected scenarios 32, the signal processing unit 32 now provides the information for the platform 22 at the output interface 26.

[0053] Although some aspects have been described in connection with a device, it is understood that these aspects also represent a description of the corresponding process, so that a block or component of a device can also be understood as a corresponding process step or as a feature of a process step. Similarly, aspects described in connection with or as a process step also represent a description of a corresponding block, detail, or feature of a corresponding device.

[0054] Depending on the implementation requirements, embodiments of the invention can be implemented in hardware or in software. The implementation can be carried out using a digital storage medium, for example, a magnetic or optical storage medium, on which electronically readable control signals, e.g., a computer program, are stored. These signals can interact with a programmable computer system (CPU and / or GPU) in such a way that the respective method is carried out. For this reason, the digital storage medium should be computer-readable. Embodiments can therefore include a data carrier that has electronically readable control signals capable of interacting with a programmable computer system in such a way that one of the methods described herein is carried out.

[0055] In some embodiments, a programmable logic device (for example, a field-programmable gate array, an FPGA) can be used to perform some or all of the functionalities of the methods described herein. In some embodiments, a field-programmable gate array can interact with a microprocessor to perform one of the methods described herein. Generally, in some embodiments, the methods are performed by any hardware device. This can be general-purpose hardware such as a computer processor (CPU) or a graphics processing unit (GPU), or hardware specific to the method, such as an ASIC. Distributed execution across the CPU and GPU is also possible.

[0056] The embodiments described above merely illustrate the principles of the present invention. It is understood that modifications and variations of the arrangements and details described herein will be obvious to other people skilled in the art. Therefore, it is intended that the invention be limited only by the scope of protection set forth in the following claims and not by the specific details presented herein by way of description and explanation of the embodiments. Reference numerals:

[0057] 20 Signal processing module

[0058] 22 Platform

[0059] 24 input interfaces

[0060] 26 Output interface

[0061] 28 Signal processing unit

[0062] 30 original / actual situation picture 31 digital twins

[0063] 32 scenarios

Claims

Patent claims 1. Signal processing module (20) of a platform (22) in a maritime environment with the following features: - an input interface (24) which is configured to receive sensor data for the creation of a situational image (30) or a situational image (30) that has already been created; - an output interface (26) configured to output information to the platform (22); - a signal processing unit (28) which is designed to generate digital twins (31) for relevant objects in the situational image (30) and to simulate different scenarios (32) using the digital twins (31) based on the situational image (30); - wherein the digital twins (31 ) each have a mathematical meta-model trained to change a decision of an associated digital twin (31) into a meta-decision if the meta-decision is worse for that digital twin (31) but better for a result of the current scenario (32), or to make a supplementary meta-decision; - wherein the signal processing unit (28) is configured to provide information to the platform (22) at the output interface (26), depending on one or more selected simulated scenarios (32).

2. Signal processing module (20) according to claim 1, wherein the digital twins (31) each have a primary mathematical model to make decisions for the simulation of the scenarios (32).

3. Signal processing module (20) according to claim 2, wherein the primary mathematical model is trained using reinforcement learning.

4. Signal processing module (20) according to one of the preceding claims, wherein the signal processing unit (28) is configured to provide information to the platform depending on a further development of the situation picture in the different scenarios (32), in particular a selected scenario.

5. Signal processing module (20) according to one of the preceding claims, wherein the meta-model is trained using reinforcement learning.

6. Signal processing module (20) according to one of the preceding claims, wherein the digital twins (31) are divided into a party with friendly platforms and a party with hostile platforms.

7. Signal processing module (20) according to one of the preceding claims, wherein the scenarios (32) are simulated based on different boundary conditions or an initial predetermined decision.

8. Signal processing module (20) according to one of the preceding claims, wherein the information for the platform (22) comprises navigation information to navigate the platform (22) according to a selected scenario (32).

9. Signal processing module (20) according to claim 8, wherein the navigation information includes or consists of a speed and / or a course for the platform (20).

10. Signal processing module (20) according to one of the preceding claims, wherein the platform (20) comprises an effector, in particular a weapon, wherein the information for the platform includes information for controlling the effector in order to proactively respond to a future situational picture predicted by a selected scenario (32).

11. Signal processing module (20) according to one of the preceding claims, wherein the platform has a communication unit, wherein the signal processing unit (28) is configured to control the communication unit depending on the information for the platform in such a way that support is requested from a base station or a watercraft or an aircraft in the maritime environment.

12. Signal processing module (20) according to one of the preceding claims, wherein the signal processing unit (28) is configured to control a target tracking unit of a weapon depending on the information for the platform (22) in order to prepare the target tracking unit to proactively track one of the relevant objects of the situational picture (30) as a target and to be able to engage the target at the earliest possible time.

13. Signal processing module (20) according to one of the preceding claims, wherein the signal processing unit (28) is configured to output information for the platform for the coordination of a crew of the platform.

14. Signal processing module (20) according to one of the preceding claims, wherein the signal processing unit (20) is configured to suggest a tactic for a selection of friendly platforms as information for the platform.

15. Signal processing module (20) according to claim 14, wherein the platform has a communication unit, wherein the signal processing unit (28) is configured to control the communication unit such that the tactics are sent to the selection of friendly platforms.

16. Procedure for providing information to a platform (22) in a maritime environment comprising the following steps: a1) Receiving sensor data to create a situational picture (30) and creating the situational picture (30); or a2) Receiving an already created situation picture (30); b) Generating digital twins (31) for relevant objects in the situational picture (30); c) Simulating different scenarios (32) based on the situational picture (30) using the digital twins, each digital twin having a mathematical meta-model; d) Changing a decision of an associated digital twin (31) into a meta-decision if the meta-decision is worse for that digital twin but better for a result of the current scenario; e) Providing the information to the platform depending on one or more selected simulated scenarios.

17. Method for training a digital twin to simulate situational awareness based on an existing situational awareness picture, comprising the following steps: a) Modeling a digital twin for an object of the situational awareness by implementing a mathematical model and taking into account properties of the object; b) Training the digital twin by positively rewarding positive actions and negatively rewarding negative actions of the respective digital twin in a simulation environment; c) Training a meta-model of the digital twin by positively rewarding actions of the digital twin that positively affect the outcome of a scenario and negatively rewarding actions of the digital twin that negatively affect the outcome of the scenario.

18. Computer program comprising instructions which, when the program is executed by a computer, cause it to execute the method according to claim 16 or claim 17.