Proceedings

Abstract

Method for the adaptive control of a vehicle convoy, wherein vehicle data from several vehicles of the vehicle convoy are acquired, wherein traffic data from an infrastructure and environmental information are acquired, wherein a common context model for all vehicles is built for synchronized decision-making, wherein the vehicle data, the traffic data and the environmental information are analyzed using a Large Language Model for pattern recognition and derivation of adaptive driving strategies, wherein the vehicles are coordinated by a multi-agent system, wherein the multi-agent system comprises vehicle agents in the individual vehicles, a convoy coordination agent for controlling an overall convoy strategy and a higher-level Large Language Model agent for analyzing traffic patterns, wherein distances,The speeds and lane selection of vehicles in the vehicle convoy are dynamically adjusted based on driving strategies derived from the Large Language Model to optimize traffic flow and energy efficiency, with the Large Language Model being used for decision-making in complex convoy scenarios, cooperative decisions being made between the convoy agents through intelligent communication, and a safety logic being implemented for the controlled dissolution of the vehicle convoy in case of disruptions.

Description

[0001] The invention relates to a method for the adaptive control of a vehicle convoy.

[0002] The increasing connectivity of modern vehicles enables new forms of cooperative driving strategies. Particularly in the area of ​​automated and assisted driving, this opens up potential for optimizing traffic flow, energy efficiency, and safety through coordinated vehicle groups. One concrete application is the dynamic formation of vehicle convoys on highways or expressways. By coordinating driving in vehicle convoys, air resistance can be reduced, more consistent driving profiles achieved, and bottlenecks in traffic flow avoided.

[0003] Current approaches to convoy formation are often based on rigidly defined rules or simple vehicle-to-vehicle communication protocols. However, these methods reach their limits when it comes to adaptively optimizing complex traffic situations, energy states, and driving strategies of multiple vehicles simultaneously. In particular, there is a lack of flexible, adaptive systems that can recognize patterns in real time, interpret contextual information, and develop cooperative strategies based on this understanding.

[0004] The current state of the art in convoy formation and control is characterized by the challenges outlined below. These include limited context processing, as rule-based systems can only partially represent complex traffic situations and cannot flexibly react to dynamic changes; a lack of holistic optimization, as distances, speed, lane selection, and energy consumption are not considered in an integrated optimization approach; limited communication, as information exchange between vehicles is often restricted to simple driving commands or sensor data, without higher-level logical coordination; insufficient energy optimization, as strategies such as slipstreaming or smooth acceleration are not systematically used; and a lack of safety logic for adaptive convoy dissolution in the event of disruptions such as sudden braking, traffic jams, or technical problems.

[0005] German patent DE 10 2023 004 727 A1 describes a method for operating a vehicle in which information transmitted via an in-vehicle communication network and / or information processed in the vehicle is examined for anomalies, and in which a reaction is triggered as a result of detecting at least one anomaly. In this method, an in-vehicle processing unit checks whether at least one parameter included in the information lies outside a defined permissible range of values, and if so, replaces the parameter value with an artificially generated value within the permissible range.

[0006] Against this background, one object of the present invention is to provide an improved method.

[0007] Accordingly, a method for the adaptive control of a vehicle convoy is proposed, wherein vehicle data from several vehicles of the vehicle convoy are acquired, traffic data from an infrastructure and environmental information are acquired, wherein a common context model for all vehicles is built for synchronized decision-making, wherein the vehicle data, the traffic data and the environmental information are analyzed using a Large Language Model for pattern recognition and derivation of adaptive driving strategies, wherein the vehicles are coordinated by a multi-agent system, wherein the multi-agent system comprises vehicle agents in the individual vehicles, a convoy coordination agent for controlling an overall convoy strategy and a higher-level Large Language Model agent for analyzing traffic patterns, including distances,The speeds and lane selection of vehicles in the vehicle convoy are dynamically adjusted based on driving strategies derived from the Large Language Model to optimize traffic flow and energy efficiency, with the Large Language Model being used for decision-making in complex convoy scenarios, cooperative decisions being made between the convoy agents through intelligent communication, and a safety logic being implemented for the controlled dissolution of the vehicle convoy in case of disruptions.

[0008] This enables a holistic and adaptive optimization of vehicle convoys that goes beyond conventional rule-based systems and can flexibly handle complex traffic situations.

[0009] In this context, a "vehicle convoy" is a coordinated group of vehicles traveling together at a defined distance and with harmonized driving parameters. "Vehicle data" refers to information about the current status and parameters of individual vehicles. "Traffic data" encompasses information about the current traffic conditions, traffic density, and traffic flow. "Environmental information" refers to data about external conditions and circumstances that can influence driving.

[0010] A "shared context model" is a unified data representation of all relevant information available to all vehicles. A "Large Language Model" is a machine learning-based system for processing and analyzing natural language and complex data structures. "Adaptive driving strategies" are dynamically adaptable driving styles that can adjust to changing conditions. A "multi-agent system" is an architecture consisting of several autonomous software components that communicate and cooperate with each other.

[0011] "Vehicle agents" are local software components in individual vehicles for controlling vehicle-specific functions. A "convoy coordination agent" is a higher-level software component for controlling the overall convoy strategy. "Intelligent communication" is an adaptive exchange of information between system components based on contextual analysis. A "safety logic" is a set of rules for ensuring vehicle safety in critical situations.

[0012] According to one embodiment, the vehicle data includes vehicle position, speed, energy state, and sensor data. This ensures comprehensive recording of all relevant vehicle parameters for precise convoy coordination. In this context, "vehicle position" refers to the spatial location of a vehicle, typically determined by GPS coordinates. "Energy state" includes information about a vehicle's current energy supply and energy consumption. "Sensor data" refers to measurements from sensors used to detect the vehicle's surroundings and its condition.

[0013] According to another embodiment, the sensor data includes radar and camera data. This enables reliable environmental perception and object recognition for safe convoy driving. "Radar and camera data" refers to specific sensor data for capturing the vehicle's surroundings using radar technology and optical image acquisition.

[0014] According to another embodiment, the environmental information includes weather information. This enables weather-dependent adjustments to driving strategies to increase safety and efficiency. In this context, "weather information" includes data on meteorological conditions such as precipitation, visibility, wind speed, and road conditions.

[0015] According to another embodiment, the Large Language Model is combined with Retrieval-Augmented Generation, incorporating technical vehicle information, traffic regulations, and situational data. This improves decision-making through additional contextual information and ensures compliance with traffic regulations. Retrieval-Augmented Generation is a technique for extending Large Language Models with additional information sources and databases. In this context, "traffic regulations" refers to legal and regulatory rules for road traffic. "Situational data" comprises contextual information about the current traffic situation.

[0016] According to another embodiment, energy consumption models and safety restrictions are taken into account during the analysis using the Large Language Model. This enables an optimal balance between energy efficiency and safety requirements in convoy control. "Energy consumption models" are mathematical models for predicting and optimizing the energy demand of vehicles. "Safety restrictions" are limitations and specifications to ensure vehicle safety.

[0017] According to another embodiment, intelligent communication between convoy agents takes place via V2V and V2X communication channels. This ensures reliable and standardized data transmission between the vehicles and the infrastructure. "V2V communication channels" enable direct communication between vehicles (vehicle-to-vehicle). "V2X communication channels" encompass communication between vehicles and the entire traffic infrastructure (vehicle-to-everything).

[0018] According to another embodiment, the adaptive driving strategies include slipstreaming, smooth acceleration profiles, and coordinated lane changes. This leads to a significant reduction in energy consumption and an improvement in traffic flow. "Slipstreaming" is a driving strategy for reducing air resistance by driving in the slipstream of vehicles ahead. "Smooth acceleration profiles" are continuous and harmonious changes in speed for energy optimization. "Coordinated lane changes" are synchronized lane changes by several vehicles for traffic flow optimization.

[0019] According to another embodiment, the safety logic includes a dynamic dissolution or reorganization of the vehicle convoy in the event of technical defects, traffic obstructions, and sudden braking maneuvers. This ensures the safety of all convoy participants even in critical situations. "Technical defects" include malfunctions or failures of vehicle components. "Traffic obstructions" are objects or situations that impede the normal flow of traffic. "Sudden braking maneuvers" are unforeseen, severe decelerations of vehicles.

[0020] According to another embodiment, fleet-wide coordination is achieved using a cloud connection and real-time communication between convoy agents. This enables large-scale optimization of traffic flow beyond individual vehicle convoys. "Fleet-wide coordination" refers to the alignment between different vehicle groups or fleets. A "cloud connection" enables access to cloud-based services and data processing systems. "Real-time communication" is the immediate data transmission without any significant delay.

[0021] The term "one" here should not necessarily be understood as restricting the number to exactly one element. Rather, it can also refer to multiple elements, such as two, three, or more. Similarly, every other numerical word used here should not be interpreted as requiring a precise restriction to exactly the corresponding number of elements. Rather, numerical deviations, both higher and lower, are possible.

[0022] Other possible implementations of the invention also include combinations of features or embodiments described previously or subsequently with regard to the exemplary embodiments, even if not explicitly mentioned. In such cases, the person skilled in the art will also add individual aspects as improvements or additions to the respective basic form of the invention.

[0023] Further advantageous embodiments and aspects of the invention are the subject of the dependent claims and the exemplary embodiments of the invention described below. The invention will be explained in more detail below with reference to preferred embodiments.

[0024] The increasing connectivity of modern vehicles enables new forms of cooperative driving strategies. Particularly in the area of ​​automated and assisted driving, this opens up potential for optimizing traffic flow, energy efficiency, and safety through coordinated vehicle groups. One concrete application is the dynamic formation of vehicle convoys on highways or expressways. By coordinating vehicle driving in groups, air resistance can be reduced, more consistent driving profiles achieved, and traffic bottlenecks avoided.

[0025] Previous approaches to convoy formation are often based on rigidly defined rules or simple vehicle-to-vehicle communication protocols. However, these methods reach their limits when it comes to adaptively optimizing complex traffic situations, energy states, and driving strategies of multiple vehicles simultaneously.

[0026] In particular, there is a lack of flexible, adaptive systems that can recognize patterns in real time, interpret contextual information, and develop cooperative strategies based on this. Large Language Models (LLMs) with multi-agent architectures offer new possibilities in this area.

[0027] The current state of the art in convoy formation and control is characterized by the following challenges.The challenges include limited context processing; in particular, rule-based systems can only represent complex traffic situations to a limited extent and do not react flexibly to dynamic changes; a lack of holistic optimization; in particular, distances, speed, lane selection, and energy consumption are not considered in an integrated optimization approach; limited communication; in particular, the exchange of information between vehicles is often limited to simple driving commands or sensor data without higher-level logical coordination; insufficient energy optimization; in particular, strategies such as slipstreaming or smooth acceleration are not used systematically; and a lack of safety logic for adaptive convoy dissolution in the event of disturbances, in particular sudden braking, traffic jams, and / or technical problems.

[0028] The present method comprises an LLM-based multi-agent system for the dynamic, context-sensitive control of vehicle convoys. Vehicles, particularly passenger cars, autonomously form convoys, whose distances, speeds, lane selection, and energy consumption are optimized by cooperating agents. A higher-level LLM agent analyzes traffic data, vehicle states, and environmental information, recognizes patterns, and develops adaptive driving strategies. The method combines approaches from traffic flow optimization and fleet coordination. Through intelligent communication between convoy agents, cooperative decisions are made that improve traffic flow and reduce energy consumption.

[0029] A holistic, LLM-based approach combines traffic flow optimization and fleet coordination. LLM is used for pattern recognition in traffic data and for decision-making in complex convoy scenarios. Distances, speed, and lane selection are dynamically adjusted based on cooperative agent logic. Energy-efficient driving strategies are implemented, such as slipstreaming, smooth acceleration profiles, and coordinated lane changes. A safety logic is in place for the controlled disbanding of the vehicle convoy in the event of disruptions, such as technical defects, traffic obstructions, or sudden braking maneuvers. Fleet-wide coordination is achieved through cloud connectivity and real-time communication between convoy agents.

[0030] The technical implementation is explained below. Data acquisition and context modeling are achieved by collecting relevant data such as vehicle position, speed, energy status, sensor data (radar, camera), traffic data from the infrastructure, and weather information. A common context model is built for all vehicles in the convoy to enable synchronized decision-making.

[0031] The multi-agent architecture includes vehicle agents, in particular local agents in each vehicle for implementing driving maneuvers and communicating with other vehicles, a convoy coordination agent, in particular for controlling the overall convoy strategy (distances, speed, lane selection) based on LLM recommendations, and a higher-level LLM agent, in particular for analyzing traffic patterns, recognizing relevant situations, deriving driving strategies, and coordinating multiple vehicle convoys.

[0032] The Decision Layer (LLM Reasoning) includes the use of an LLM trained on traffic scenarios and fleet strategies to generate adaptive driving decisions, the combination with Retrieval-Augmented Generation (RAG) to incorporate technical vehicle information, traffic rules and situational data, and the consideration of energy consumption models and safety restrictions.

[0033] Communication and security includes the use of V2V and V2X communication channels for agent coordination, the implementation of security logic for dynamic resolution or reorganization of the vehicle convoy in case of disruptions, and ensuring redundancy and stability through decentralized decision-making mechanisms.

[0034] In existing systems, the decision logic aspect is rule-based and therefore limited. In the invention, however, it is LLM-based and thus context-sensitive. The optimization scope aspect is limited to individual aspects (e.g., distances or speed) in existing systems. In contrast, the invention is holistic (distances, speed, lane selection, energy). The communication aspect is limited to simple driving commands in existing systems. In the invention, however, it is based on cooperative agent communication.

[0035] The aspect of energy efficiency is hardly considered in existing systems, whereas the invention implements systematic strategies such as slipstreaming. The aspect of disruption management in existing systems is reactive and therefore simple. The invention implements an adaptive safety logic for convoy dissolution. The aspect of fleet coordination in existing systems involves individual vehicles or rigid groups, while the invention enables dynamic, LLM-controlled convoy formation and management. 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] DE 10 2023 004 727 A1

[0005]

Claims

[1] Method for adaptively controlling a vehicle convoy, wherein vehicle data from several vehicles of the vehicle convoy are acquired, wherein traffic data from an infrastructure and environmental information are acquired, wherein a common context model for all vehicles is built for synchronized decision-making, wherein the vehicle data, the traffic data and the environmental information are analyzed using a Large Language Model for pattern recognition and derivation of adaptive driving strategies, wherein the vehicles are coordinated by a multi-agent system, wherein the multi-agent system comprises vehicle agents in the individual vehicles, a convoy coordination agent for controlling an overall convoy strategy and a higher-level Large Language Model agent for analyzing traffic patterns, wherein distances,The speeds and lane selection of vehicles in the vehicle convoy are dynamically adjusted based on driving strategies derived from the Large Language Model to optimize traffic flow and energy efficiency, with the Large Language Model being used for decision-making in complex convoy scenarios, cooperative decisions being made between the convoy agents through intelligent communication, and a safety logic being implemented for the controlled dissolution of the vehicle convoy in case of disruptions. [2] Method according to claim 1, wherein the vehicle data includes vehicle position, speed, energy state and sensor data. [3] Method according to claim 2, wherein the sensor data includes radar and camera data. [4] Method according to any of the preceding claims, wherein the environmental information includes weather information. [5] Method according to any of the preceding claims, wherein the Large Language Model is combined with Retrieval-Augmented Generation, so that technical vehicle information, traffic rules and situation data are included. [6] Method according to one of the preceding claims, wherein energy consumption models and safety restrictions are taken into account in the analysis using the Large Language Model. [7] Method according to any of the preceding claims, wherein the intelligent communication between the convoy agents takes place via V2V and V2X communication channels. [8] Method according to any of the preceding claims, wherein the adaptive driving strategies include slipstreaming, uniform acceleration profiles and / or coordinated lane changes. [9] Method according to any of the preceding claims, wherein the safety logic comprises a dynamic dissolution or reorganization of the vehicle convoy in the event of technical defects, traffic obstructions and / or sudden braking maneuvers. [10] Method according to any of the preceding claims, wherein cross-fleet coordination is carried out using a cloud connection and / or real-time communication between convoy agents.