Assistance system for a motor vehicle and procedures

Abstract

The invention relates to an assistance system (10) for a motor vehicle, by which a multitude of vehicle functions can be controlled depending on predefined rules, comprising a vehicle data module which is configured to acquire vehicle data and / or status information of the motor vehicle, an input module which is configured to receive inputs from a driver via at least one user interface, and a communication module (20) which is configured to establish a data connection to external entities, wherein an AI control module (40) is arranged which is configured to derive an abstract driver goal from natural language inputs, vehicle data provided by the vehicle data module and / or context data determined by the communication module (20) using a domain-specific trained large-language model (LLM).and to divide this driver objective into coordinated subtasks for several vehicle functions, depending on an emotional state of the driver determined from sensor data, and to forward the respective subtasks to specialized function modules (60) for execution by means of a planning module (50). Furthermore, the invention relates to a method.

Description

[0001] The invention relates to an assistance system for a motor vehicle according to the preamble of claim 1. Furthermore, the invention relates to a method for operating an assistance system for a motor vehicle.

[0002] Such assistance systems for a motor vehicle use a modular architecture to address and / or coordinate multiple vehicle functions simultaneously, connecting a user interface, vehicle-side data sources and external entities.

[0003] US 2006 / 0036356 A1 discloses a system-side policy approach that provides rule-based control logic between in-vehicle applications and / or external devices, thereby enabling functions depending on predefined rules.

[0004] The object of the invention is to provide personalized assistance by means of an assistance system for a motor vehicle, whereby natural language inputs, vehicle data and / or context data are transformed into coordinated subtasks for several vehicle functions.

[0005] This problem is solved by means of an assistance system for a motor vehicle with the features of claim 1, and by means of a method according to the invention for operating an assistance system for a motor vehicle. Advantageous embodiments of the assistance system for a motor vehicle according to the invention are to be regarded as advantageous embodiments of the method according to the invention for operating an assistance system for a motor vehicle, wherein the means of the assistance system can be used to carry out the respective method steps. Furthermore, advantageous developments of the invention are described by the dependent claims, the following description, and by figures.

[0006] One aspect of the invention relates to an assistance system for a motor vehicle, in particular a passenger car, by which a multitude of vehicle functions can be controlled according to predefined rules, comprising a vehicle data module configured to acquire vehicle data and / or vehicle status information, an input module configured to receive input from a driver via at least one user interface, and a communication module configured to establish a data connection to external entities. The assistance system thus comprises a modular basic structure in which a vehicle data module provides sensor and / or control unit information, an input module acquires natural language input and / or other input, and a communication module provides a connection to external entities, in particular applications and / or devices.The term "entities" refers to external applications and / or devices. The relationship between the vehicle data module and / or input module and / or communication module is established via an internal data connection, thus enabling downstream processing by other modules.

[0007] To solve the problem of the invention and thus provide personalized assistance, the invention provides for an AI control module which is configured to derive an abstract driver goal from natural language inputs, vehicle data provided by a vehicle data module, and / or context data determined by a communication module, using a domain-specific trained large-language model (LLM). The AI ​​control module provides, in particular, the semantic interpretation, whereby an abstract driver goal, for example, "relaxed drive" and / or "quick appointment," is generated from heterogeneous inputs.In addition, it is planned that this driver objective will be divided into coordinated subtasks for several vehicle functions depending on the driver's emotional state determined from sensor data, thereby enabling, in particular, situation-adapted and / or context-sensitive control.

[0008] The respective subtasks are forwarded to specialized functional modules for execution via a planning module, whereby in particular an IT-side interface between the planning module and / or the functional modules provides an orderly and reproducible transfer of the subtasks.

[0009] This specific design ensures that several vehicle functions can be addressed simultaneously and / or in a coordinated manner, resulting in a particularly personalized and comprehensible assistance system with modular task division.

[0010] In particular, the system is designed to initiate a relaxation function, for example. The AI ​​control module derives a driver objective, such as "relaxation in traffic jams," and translates this into several coordinated subtasks. The planning module can then pass these sub-objectives on to function modules that address various vehicle functions, such as adjusting the interior lighting, activating calming audio content, and / or controlling fragrance systems. This combined execution of several related subtasks ensures a consistent vehicle response to the detected driver state. The system allows for flexible and context-sensitive configuration, for example, by integrating external applications via the communication module to suggest or activate additional relaxation measures.In this way, a dynamic adaptation of the vehicle's behavior to the emotional and situational states of the driver is enabled, whereby the vehicle actively contributes to emotional stabilization and comfort.

[0011] In a particularly advantageous embodiment of the invention, a policy module is provided which is configured to subject the subtasks forwarded by the planning module to the function modules to a context-sensitive safety and / or authorization check before they are released for execution. The policy module evaluates, in particular, vehicle states and / or driver authorizations and / or rules, thereby rejecting impermissible subtasks. The technical interaction is such that the policy module is integrated as a checkpoint between the planning module and / or function modules, thus creating a clear separation between goal derivation and / or execution. An application example is a subtask for changing a chassis setting, which is only released when the vehicle is at a suitable speed and / or valid authorization is present.

[0012] In a further advantageous embodiment of the invention, a learning module is provided which is configured to generate an adaptive behavioral model based on driver reactions, emotional patterns, and / or contextual data. This model personalizes and continuously adapts future control decisions of the AI ​​control module. For example, the learning module records repetitions of driver preferences and / or feedback from execution, creating a model that adapts future subtasks and / or parameter selection. The connection between the learning module and / or the AI ​​control module is established via a model update, allowing the LLM and / or the planning logic to take updated weightings into account. A technical advantage is demonstrated, for example, by the automatic suggestion of a suitable interior configuration and / or route selection in the case of recurring stress patterns.

[0013] In a further advantageous embodiment of the invention, a prioritization module is provided, which is configured to evaluate and weight competing subtasks depending on safety criteria, driver state, and / or contextual data in order to ensure prioritized and conflict-free execution of vehicle functions. For example, the prioritization module subordinates audio prompts and / or climate control adjustments to safety-relevant vehicle dynamics interventions, thereby reducing conflicts between simultaneous requirements. The interface between the prioritization module and / or planning module and / or function modules is designed such that priorities and / or blocking conditions are incorporated into the execution sequence, thus achieving a defined processing pattern.A practical scenario is the simultaneous request for navigation announcements and / or fatigue warnings, where the warning is given priority.

[0014] Another aspect of the invention relates to a method for operating an assistance system for a motor vehicle, in which a control of a multitude of vehicle functions is possible depending on predefined rules, in which vehicle data and / or status information of the motor vehicle are recorded by means of a vehicle data module, in which inputs from a driver are received via at least one user interface by means of an input module, and in which a data connection to external entities is established by means of a communication module.The method is characterized by the execution of a domain-specific trained large-language model via an AI control module. This model derives an abstract driver goal from natural language input, vehicle data provided by the vehicle data module, and / or context data determined by the communication module. Depending on the driver's emotional state determined from sensor data, this derived driver goal is divided into coordinated subtasks for multiple vehicle functions. These subtasks are then forwarded to specialized function modules for execution via a planning module. Interaction between process steps and / or modules occurs via defined data interfaces, thus ensuring a clear separation between data acquisition, interpretation, and / or execution.This creates a process that is particularly comprehensible and implementable for a specialist in vehicle design.

[0015] In other words, the driver assistance system for a motor vehicle is intended to include not only the modules mentioned in the preceding sections, but also, optionally, a logging module, a fallback module, and / or an interface control for external entities. A logging module is designed to record execution states and / or subtasks over time, thereby providing, in particular, diagnostics and / or traceability of system activities. A fallback module may be provided, which essentially ensures the rule-based continuation of function control in the event of limited connectivity and / or computing resources. Additionally, an entity interface control may be provided, enabling granular release of data streams and / or function calls between third-party applications and vehicle-internal modules.The combination of these features leads to a particularly robust solution suitable for complex vehicle architectures, thereby further developing personalized assistance compared to the known state of the art, especially since an abstract driver goal is derived at the level of the AI ​​control module and / or translated into a coordinated execution via planning, prioritization and / or policy mechanisms.

[0016] In other words, modern vehicles already possess centralized control systems and programming interfaces that provide access to numerous vehicle functions. Nevertheless, existing driver assistance systems, such as voice-based systems like Siri and Alexa, or vehicle-specific solutions like MBUX, are deterministic and operate with predefined commands. These familiar systems are essentially limited to comfort or infotainment functions and do not allow for dynamic, context-sensitive control. In particular, they lack mechanisms that would allow for... - Recognize and interpret complex user goals from natural language, - translate these goals in real time into coordinated subtasks for multiple vehicle functions, - Consider dynamic external data sources, e.g., traffic information and / or calendar data, - detect the driver's emotional state and actively incorporate it into the steering strategy, and / or - Integrate third-party applications securely and in a controlled manner into the vehicle control system without compromising the safety or integrity of the vehicle.

[0017] Furthermore, the current state of the art lacks a safety logic that ensures adaptive controls and / or external applications do not negatively impact safety-critical states. Existing solutions invariably address only partial aspects, such as speech processing, emotion recognition, or API access, without linking them in a unified architecture. While systems from GM, for example, already provide voice-based control of individual vehicle functions, they lack emotional context processing. Other solutions work with rule-based emotion recognition, but without AI-driven target composition. Still other systems allow the integration of third-party apps, but without intelligent planning or prioritization logic. Even newer developments, such as combinations of MBUX and ChatGPT, enable dialogue-based interactions but offer no connection to the vehicle's internal control functions.

[0018] The present invention differs fundamentally from this, as it employs a domain-specific trained large-language model that understands language, emotion, and situational context, derives abstract goals from them, and translates these into coordinated actions using a hybrid planning architecture. The integration of multimodal data streams, e.g., from third-party apps, into goal derivation, as well as the dynamic prioritization of competing vehicle functions, are key features of the invention.

[0019] Furthermore, the solution according to the invention offers a secure interface logic which, through a policy layer and an AI-orchestrated access system, ensures that external entities can only access vehicle functions to the authorized extent. A continuously learning model also enables ongoing personalization and adaptation of the assistance functions to individual driver preferences. This synergistic combination of semantic target recognition, emotional context assessment, safety-tested third-party integration, and planning-based function control is neither described nor suggested in the known prior art and constitutes the inventive step of the present solution.

[0020] Further advantages, features, and details of the invention will become apparent from the following description of a preferred embodiment and from the drawing(s). The features and combinations of features mentioned above in the description, as well as the features and combinations of features mentioned below in the figure description and / or shown in the figure(s) alone, can be used not only in the combinations specified, but also in other combinations or individually, without departing from the scope of the invention.

[0021] This shows: Fig. 1. An architecture to illustrate a possible embodiment of an assistance system for a motor vehicle.

[0022] In the figure, identical or functionally equivalent elements are provided with the same reference symbols.

[0023] Fig. Figure 1 shows an architecture to illustrate a possible embodiment of an assistance system 10 for a motor vehicle, through which a control of a multitude of vehicle functions depending on predefined rules can be carried out.

[0024] The assistance system 10 comprises a vehicle data module 12, which is designed to record vehicle data and / or status information of the motor vehicle, an input module 14, which is designed to receive input from a driver via at least one user interface, and a communication module 20, which is designed to establish a data connection to external entities.

[0025] Furthermore, an AI control module 40 is arranged, which is trained to derive an abstract driver goal from natural language inputs, vehicle data provided by the vehicle data module 12 and / or context data determined by the communication module 20 using a domain-specific trained Large Language Model (LLM).

[0026] This driver objective is divided into coordinated subtasks for several vehicle functions depending on the driver's emotional state determined from sensor data, whereby the respective subtasks are forwarded to specialized function modules 60 for execution by means of a planning module 50.

[0027] The purpose and technical effect of the assistance system 10 is to provide context-sensitive and personalized control of several vehicle functions, jointly evaluating voice inputs, vehicle data and the driver's emotional state.

[0028] This makes it possible to automatically translate complex driving objectives, such as creating a relaxed driving atmosphere in traffic jams, into coordinated subtasks and execute them safely.

[0029] The context presented in Fig. Figure 1 shows the case where a traffic jam has been detected and the driver or user is experiencing an increased stress level.

[0030] The figure shows the communication module 20, referred to as CalmDrive app 22 and API gateway 24, which is designed to establish S1 data connections to external entities in a first step and to transmit control requests, for example to activate a “Zen mode”.

[0031] Furthermore, a Policy Module 30, also referred to as the Policy Layer, is provided, which is designed to subject incoming tax requests to a context-sensitive security and / or authorization check.

[0032] In a second step, S2 sends a request with Auth + Scope = comfort / multi to the policy module 30 via the API gateway 24.

[0033] In a third step, the policy module 30 returns a release or approval to the API gateway 24 in S3.

[0034] Subsequently, in a fourth step, S4 transmits the target context “Zen mode in case of traffic jam” to the AI ​​control module 40, also known as the LLM agent.

[0035] The AI ​​control module 40 is trained to derive an abstract driver goal, for example in a fifth step S5 “Relaxation in traffic jams”, using a domain-specific trained LLM and to transmit this as SetGoal (“Relaxation → Zen mode”) to the planning module 50, also known as the Planning Agent.

[0036] Planning module 50 is designed to divide the driver's objective into coordinated subtasks for several vehicle functions and to forward the respective subtasks to specialized function modules 60, also known as Comfort Agents. For this purpose, in a sixth step, S6 subtasks are transmitted, for example, Subgoals = [light = blue, scent = on, audio = waves].

[0037] The function modules 60 are trained to transmit actuator commands to vehicle systems 70 in a seventh step S7, which execute the corresponding vehicle functions and respond in an eighth step S8 with feedback “Ack” to the function modules 60 and in a ninth step S9 with “done” to the planning module 50.

[0038] In a tenth step, the planning module 50 then transmits a status message "Status OK" to the KL control module 40.

[0039] In an eleventh step, the Kl control module 40 sends feedback S11 "Result: Zen mode activated" to the API gateway 24.

[0040] Finally, in a twelfth step, the API gateway 24 transmits a status message S12 to the CalmDrive app 22, for example {“status”: “executed”, “mood”: “relaxed”}.

[0041] In addition, a logging module 80, also referred to as a logging service, is provided, which stores events, context information and / or third-party IDs for traceability.

[0042] In a thirteenth step, S13 transmits an event entry, including third-party ID and context, from the function module 60 to the logging module 80.

[0043] The in Fig. The architecture shown in Figure 1 serves only to illustrate a possible system structure and does not limit the scope of protection of the invention.

[0044] In summary, the invention proposes a multimodal AI vehicle control system with specialized LLM planning, emotion recognition, and secure third-party integration. QUOTES INCLUDED IN THE DESCRIPTION

[0000] This list of documents cited by the applicant was automatically generated and is included solely for the reader's convenience. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions. Cited patent literature

[0000] US 2006 / 0036356 A1

[0003]

Claims

[1] Assistance system (10) for a motor vehicle, by which it is possible to control a large number of vehicle functions depending on predefined rules, - with a vehicle data module designed to record vehicle data and / or vehicle condition information, - with an input module designed to receive input from a driver via at least one user interface, and - with a communication module (20) which is trained to establish a data connection to external entities, characterized by, that an AI control module (40) is arranged which is trained to derive an abstract driver goal from natural language inputs, vehicle data provided by the vehicle data module and / or context data determined by the communication module (20) using a domain-specific trained large-language model (LLM), and to divide this driver goal into coordinated subtasks for several vehicle functions depending on an emotional state of the driver determined from sensor data, and to forward the respective subtasks to specialized function modules (60) for execution by means of a planning module (50). [2] Assistance system (10) according to claim 1, characterized by, that a policy module (30) is arranged which is designed to subject the subtasks forwarded by the planning module (50) to the function modules (60) to a context-sensitive security and / or authorization check before they are released for execution. [3] Assistance system (10) according to any of the preceding claims, characterized by , that a learning module is provided which is trained to generate an adaptive behavioral model depending on driver reactions, emotion patterns and / or context data, by means of which future control decisions of the AI ​​control module (40) can be personalized and continuously optimized. [4] Assistance system (10) according to any of the preceding claims, characterized by, that a prioritization module is provided which is designed to evaluate and weight competing subtasks depending on safety criteria, driver condition and / or context data, and to provide a prioritized and conflict-free execution of vehicle functions. [5] Method for operating an assistance system (10) for a motor vehicle, - in which it is possible to control a large number of vehicle functions depending on predefined rules, - in which vehicle data and / or condition information of the motor vehicle are recorded by means of a vehicle data module, - in which inputs from a driver are received via at least one user interface using an input module, - in which a data connection to external entities is established by means of a communication module (20), characterized by, that a domain-specific trained Large Language Model (LLM) is executed by means of an AI control module (40), which derives an abstract driver goal from natural language inputs, the vehicle data provided by the vehicle data module and / or the context data determined by the communication module (20), wherein the derived driver goal is divided into coordinated subtasks for several vehicle functions depending on an emotional state of the driver determined from sensor data, and wherein the subtasks are forwarded to specialized function modules (60) for execution by means of a planning module (50).

Images