A multi-vehicle AI agent natural language negotiation emergency passage control method based on C-V2X PC5

Abstract

The application discloses a kind of based on C-V2X PC5's multi-vehicle AI Agent natural language negotiation emergency passage control method, it is related to intelligent transportation and vehicle networking communication technical field, including the following steps: sending end vehicle passes through physical trigger button or voice recognition trigger emergency passage demand, and obtains current road environment information and vehicle operating state information;Based on emergency passage demand and current road environment information and vehicle operating state information, AI Agent in sending end vehicle calls preset speech technique template and generates natural language text, and natural language text is encapsulated as UDP message after light coding, further encapsulated as IPv4 broadcast message.The application realizes communication isolation by fixed UDP port, and realizes low-latency collaboration by combining vehicle end light AI;Interoperability is improved by standardized natural language template and UTF-8 coding;Modular architecture is used to enhance compatibility and expansion capability, meet the demand of vehicle networking emergency collaborative control.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation and vehicle-to-everything (V2X) communication technology, specifically to a multi-vehicle AI Agent natural language negotiation emergency traffic control method based on C-V2X PC5. Background Technology

[0002] With the continuous improvement of the vehicle-road-cloud integrated system and the rapid development of advanced autonomous driving technology, full-scenario unmanned driving has become an important development direction for future intelligent transportation. Unmanned vehicles rely heavily on all-domain environmental perception, vehicle-road cooperative interaction, and multi-vehicle group collaboration capabilities during operation. Single-vehicle intelligence is no longer sufficient to handle complex road conditions, extreme congestion, and emergency passage scenarios. Against this backdrop, real-time interconnection, intent communication, autonomous negotiation, and collaborative control between vehicles have become crucial for the large-scale application of unmanned driving. C-V2X, as a next-generation cellular vehicle-to-everything (V2X) technology, includes Uu cellular communication and PC5 direct communication. PC5 Sidelink communication, based on the 5.9GHz intelligent transportation frequency band, supports distributed resource scheduling, does not rely on base stations and core networks, and possesses low latency, high reliability, and self-organizing network capabilities, making it suitable for real-time interaction between short-range vehicle clusters. Simultaneously, with the development of onboard AI agents and edge-side large-scale model technology, vehicles now possess local semantic understanding, natural language processing, and autonomous decision-making capabilities, providing the foundation for intelligent collaboration among multiple vehicles.

[0003] However, existing C-V2X PC5 communication systems primarily rely on standardized binary structured messages (such as BSM, RSI, SPAT, etc.) for data transmission. They only support one-way broadcasting of basic data such as vehicle location, speed, and road information. The data format is fixed, the interaction dimension is limited, and semantic parsing and intent expression capabilities are lacking. This makes bidirectional communication and negotiation between vehicles impossible, and insufficient to meet the dynamic coordination needs of complex scenarios such as emergency passage. Furthermore, existing emergency passage solutions largely depend on centralized scheduling via Uu cellular links or roadside units (RSUs), heavily relying on public network communication. In environments such as tunnels and weak networks, latency jitter, packet loss, and even communication interruptions are common, making it difficult to guarantee real-time performance and reliability in autonomous driving scenarios. In addition, existing in-vehicle AI agents are mainly limited to single-vehicle decision-making or human-machine interaction, lacking cross-vehicle collaborative mechanisms based on PC5 direct communication. They have not yet developed natural language interaction, request-response, and collaborative decision-making capabilities among multiple vehicle AI agents, resulting in low efficiency and insufficient intelligence in vehicle group collaborative driving, making it difficult to support the efficient and safe emergency passage requirements of future unmanned transportation systems.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-vehicle AI Agent natural language negotiation emergency passage control method based on C-V2X PC5 to solve the problems in the background art mentioned above.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a multi-vehicle AIAgent natural language negotiation emergency passage control method based on C-V2X PC5, comprising the following steps: The sending vehicle triggers the emergency passage request through a physical trigger button or voice recognition, and obtains current road environment information and vehicle operating status information; Based on emergency passage needs, current road environment information, and vehicle operation status information, the sending end vehicle in AIAgent calls a preset script template to generate natural language text, and after lightweight encoding of the natural language text, it is encapsulated into UDP packets, and then further encapsulated into IPv4 broadcast packets. The IPv4 broadcast message is broadcast to the outside world through the C-V2X PC5 direct communication channel. The receiving vehicle receives the IPv4 broadcast message and parses it to obtain natural language text. The receiving vehicle's in-vehicle AI Agent performs semantic recognition and intent parsing on the natural language text, and generates natural language response text and corresponding autonomous driving control commands based on the parsing results. At the same time, it sends feedback messages through the C-V2X PC5 direct communication channel. After receiving the feedback message, the sending vehicle updates the route plan and generates collaborative control instructions based on the feedback content. It then drives the vehicle to execute emergency passage control through the vehicle control link, thereby realizing collaborative passage between multiple vehicle AI Agents based on natural language negotiation.

[0007] Preferably, a unified identification mechanism for multi-source triggering information is constructed, integrating physical input and speech semantic processing paths to form a consistent emergency demand output logic. The steps are as follows: The signal level change generated by the physical trigger button is collected, and the signal is sampled by the analog-to-digital converter unit to convert the continuous analog signal into a discrete digital signal and generate the corresponding trigger status flag bit. The system acquires the voice input data stream, performs pre-emphasis, framing, and windowing processing on the voice signal, extracts Mel frequency cepstral coefficient features, outputs text results through a speech recognition model, and then extracts intent labels through a semantic parsing model. The trigger status flag and intent label are mapped to fields to build a unified data structure, including trigger source field, trigger type field and timestamp field, and data alignment is completed. A unified data structure is written to the shared cache, encapsulated into an emergency request data frame according to a preset message format, and pushed to the communication processing queue.

[0008] Preferably, a structured script generation method is introduced, combined with scene classification and parameter filling strategies, to form natural language interactive content with a unified expression format. The steps are as follows: Establish a script template data table. Each template includes a fixed statement segment, a variable parameter segment, and field identifiers. Assign template index numbers to different traffic scenarios. Read the current environment status identifier and match the template index, then write the vehicle IP address, latitude and longitude coordinates, speed value, and target action parameters into the corresponding field positions; Perform character-level processing on the filled text, including special character replacement, field order rearrangement, removal of empty fields, and pruning of redundant characters; The generated text is converted to a standard string format and written to the output buffer, along with a text length field and a validation flag.

[0009] Preferably, a unified text encoding processing flow is constructed, which, through character conversion and data structure organization, forms a data format suitable for communication transmission. The steps are as follows: Read standard string data and match it character by character according to the UTF-8 encoding rules, mapping each character to the corresponding byte sequence; Perform a concatenation operation on the byte sequence to generate a continuous byte stream, and record the start and end offset addresses; The length of the byte stream is counted and written to the data header, while boundary identifiers are added to mark the start and end positions of the data. Write the byte stream to the send buffer and generate a data description structure for subsequent encapsulation processes.

[0010] Preferably, the transport layer encapsulation process is constructed by organizing port identifiers and message structures to form identifiable data transmission units, and the steps are as follows: Initialize the UDP packet structure, allocate storage space for the header and data fields, and write a fixed application port number; Copy the encoded byte stream to the UDP data area and record the data area length and offset. Fill in the UDP header fields, including the source port number, destination port number, data length field, and checksum field; Perform checksum calculations on the entire UDP packet and generate a complete packet object, which is then stored in the network sending queue.

[0011] Preferably, a network layer broadcast mechanism is formed through address configuration and message construction to complete the multi-node data distribution process, as follows: Construct the IP packet structure and write the UDP packet as payload data into the data area; Set the IP header fields, including source IP address, broadcast destination address, protocol type field, and identifier field; Perform fragmentation flag configuration and header checksum calculation on IP packets to form complete IP data packets; The IP data packet is written into the physical layer transmit buffer and broadcast through the PC5 interface.

[0012] Preferably, a layered parsing process is constructed at the receiving end to recover the original semantic content through protocol parsing and data extraction, with the following steps: The radio frequency module receives wireless signals and performs demodulation processing, converting the signals into digital data frames; Perform link-layer deframe operation on the data frame to identify the frame header and trailer and extract the IP data segment; Parse the IP header fields and identify the protocol type; further parse the UDP header and locate the start of the data area. Perform UTF-8 decoding on the byte stream of the data area, restore it to string text, and write it to the semantic processing buffer.

[0013] Preferably, the IP header field is subjected to validation rule matching and the validity of the message structure is confirmed; the UDP header length field is subjected to consistency verification and the data area range is determined; the byte stream is subjected to boundary identifier recognition and continuity check processing; and the decoding result is subjected to character validity verification and written to the semantic processing buffer for subsequent semantic parsing.

[0014] Preferably, vehicle control information is generated by constructing a semantic-to-control command mapping process through intent recognition and parameter extraction, with the following steps: Perform word segmentation on the received text and build a word sequence, identifying action keywords and parameter keywords; Extract parameter fields, including target velocity value, lateral offset distance, execution time window, and priority indicator; The extracted parameters are input into the control decision model, and control trajectory data is generated through a path planning algorithm. The trajectory data is converted into a set of control commands, including acceleration, braking and steering commands, and output to the execution control interface.

[0015] Preferably, a multi-vehicle collaborative operation process is constructed through state updates and strategy adjustments to maintain continuous collaborative passage. The steps are as follows: It receives feedback messages from the communication link and parses out the status information of neighboring vehicles, including their position, speed, and action status. The status information of neighboring vehicles is fused with the perception data of this vehicle to update the environmental model data structure; The updated environment model is recalculated to replan the execution path and generate a new sequence of control parameters. The control parameter sequence is sent to the execution system and status broadcast data is generated synchronously and written into the communication sending queue for subsequent interaction.

[0016] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This invention achieves fine-grained isolation and identification of business data at the communication mechanism level by fixing a dedicated V2X UDP communication port and using the port number as a unique identifier for the application layer. During vehicle-to-vehicle direct communication, different business data are distinguished through the port, effectively avoiding interference problems present in traditional mixed data transmission and improving the stability and reliability of the communication link. Simultaneously, the fixed port design allows the receiving end to quickly locate the target data stream, reducing parsing complexity and processing latency. It maintains stable performance even in complex traffic environments and high-concurrency communication scenarios, further enhancing the security and controllability of V2X communication and providing a reliable data transmission foundation for emergency collaborative control.

[0017] This invention employs a lightweight model architecture, deploying key capabilities such as perception processing, semantic understanding, event recognition, dialogue generation, and decision mapping entirely within the vehicle-side domain controller. This eliminates the need to rely on large cloud-based models for inference, effectively avoiding latency fluctuations and uncertainties associated with public network communication. Rapid response is achieved through local computing on the device side. Combined with a fixed UDP port and IPv4 broadcast encapsulation mechanism, emergency requests can complete a closed-loop control process—from generation and transmission to parsing and execution—within a very short time. This significantly reduces the overall system response time, meeting the stringent requirements of low latency and high reliability in autonomous driving scenarios and significantly improving real-time coordination capabilities in emergency passage situations.

[0018] This invention constructs a unified vehicle-to-vehicle and vehicle-to-infrastructure (V2X) interaction protocol paradigm by designing a fixed-byte-length V2X interaction template and a standardized UTF-8 byte stream encapsulation format. This scheme ensures consistent communication content structure and clear semantics through template-based constraints and parameterized filling of natural language expressions, while balancing transmission and parsing efficiency. Different vehicle manufacturers, different domain control hardware platforms, and different perception algorithm systems can all send, receive, and parse data based on this unified standard, fundamentally solving problems such as inconsistent protocols, semantic incompatibility, and non-standard message formats in existing technologies. This significantly improves the compatibility and interoperability of vehicle networking systems, providing a standardized foundation for large-scale commercial applications.

[0019] This invention employs a modular architecture design with model layering, domain control deployment, port fixing, and template standardization. This achieves decoupling and collaboration among the perception, communication, decision-making, and control functional modules, giving the overall technical solution excellent scalability and adaptability. This architecture is compatible with mainstream perception systems, V2X communication modules, and autonomous driving domain controllers, supporting rapid deployment and flexible configuration across different vehicle platforms, reducing system integration difficulty and development costs. Furthermore, the technical solution reserves interface expansion capabilities during the design phase, supporting subsequent vehicle-to-cloud collaboration, enhanced information security, and the evolution of higher-level autonomous driving functions, providing a solid foundation for the continuous upgrading of future intelligent transportation systems. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0021] Figure 1 This is a schematic diagram of the distributed architecture of the multi-vehicle AI Agent and the C-V2X PC5 communication relationship of the present invention.

[0022] Figure 2 This is a schematic diagram of the natural language negotiation and collaborative control process of the multi-vehicle AI Agent in this invention. Detailed Implementation

[0023] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0024] This invention provides, for example Figure 1 and Figure 2 The following is a multi-vehicle AI Agent natural language negotiation emergency passage control method based on C-V2X PC5, with the following specific steps: The multi-vehicle AI Agent system mainly consists of a sending vehicle A and a receiving vehicle B. The two types of vehicles interact and coordinate control through a C-V2XPC5 direct communication link. The overall architecture adopts a vehicle-side distributed control system, forming a closed-loop link around perception, communication, computing, and execution.

[0025] The transmitting vehicle A mainly includes sensor A, an onboard communication unit, an autonomous driving control unit, a vehicle control unit, and actuators. Sensor A includes cameras, LiDAR, ultrasonic radar, and millimeter-wave radar, used for multi-dimensional real-time perception of the vehicle's surrounding environment, acquiring road structure information, traffic participant status information, and dynamic obstacle information. Cameras are used for visual semantic recognition and lane detection, LiDAR for high-precision spatial modeling, millimeter-wave radar for long-range target detection, and ultrasonic radar for short-range obstacle avoidance, thus forming the foundational data for multi-source perception fusion.

[0026] The onboard communication unit is an OBU, whose input sources include physical trigger buttons and voice recognition input. The physical trigger buttons are connected via hardwired connections, enabling rapid activation of emergency passage requests in urgent situations. Voice recognition input involves the cockpit system collecting user voice information, which is then semantically parsed to identify the intent of the emergency request. Sensor A and the OBU input sources are connected via wiring harnesses to ensure real-time and stable data transmission.

[0027] In terms of in-vehicle communication, the OBU and the autonomous driving control unit exchange data via CAN bus or Ethernet network. The autonomous driving control unit and the vehicle control unit also use CAN or Ethernet communication. The vehicle control unit and the actuators use CAN communication, realizing a step-by-step transmission link from information acquisition and decision generation to control execution.

[0028] The overall architecture of the receiving vehicle B is consistent with that of the sending vehicle A, also including sensors, OBU, autonomous driving control unit, vehicle control unit, and actuators. These units are interconnected via an onboard bus. The receiving vehicle B receives broadcast information from the sending vehicle A via C-V2X PC5 direct communication and parses and responds based on the received data.

[0029] At the vehicle's electronic and electrical architecture level, the OBU can be broken down into multiple functional controllers, including the cockpit domain controller and the connected terminal controller. The cockpit domain controller is responsible for receiving physical trigger button input and voice recognition input, performing signal processing and semantic parsing through internal software, and transmitting the input information to the OBU via CAN or Ethernet signals to achieve standardized conversion and reporting of input information.

[0030] The AI ​​Agent is deployed in a distributed manner on the vehicle side, and can be deployed in the cockpit domain controller and the autonomous driving control unit, each undertaking different functional roles. The AI ​​Agent in the cockpit domain controller is mainly used to realize human-machine interaction, voice understanding, and emergency demand recognition, serving as the input source for emergency passage requests; the AI ​​Agent in the autonomous driving control unit is mainly responsible for environmental understanding, decision generation, and control command output, used to support the execution of autonomous driving behavior and collaborative control.

[0031] In practical architecture, the vehicle control unit can be further subdivided into multiple sub-controllers, such as the vehicle gateway and the battery management system. The vehicle gateway is responsible for vehicle signal routing and protocol conversion, enabling data interaction between different buses; the battery management system is responsible for vehicle energy status monitoring and power management. Different controllers communicate with each other via CAN or Ethernet networks, forming a unified in-vehicle data interaction system.

[0032] Actuators can be categorized by function into powertrain actuators, braking system actuators, and steering system actuators. Powertrain actuators include engine controllers or motor controllers, used for acceleration and drive control; braking system actuators include electronic brake control units and anti-lock braking controllers, used for deceleration and braking control; steering system actuators include electronic power steering controllers or steer-by-wire controllers, used for vehicle direction adjustment. Different vehicle models may have different actuator configurations, and the overall structure can be adjusted according to actual development needs.

[0033] At the algorithm level, the AI ​​Agent integrates the capabilities of multiple large models to form a multi-model system for autonomous driving and cooperative control. The autonomous driving control unit integrates a visual scene understanding model, a radar point cloud perception model, and a multimodal feature fusion model to perform environmental modeling and target recognition; a semantic-control command mapping model is used to convert natural language interaction information into executable vehicle control commands; and a cooperative decision-making model is used for generating and dynamically adjusting cooperative strategies among multiple vehicles. The cockpit domain controller integrates a traffic scene semantic understanding model and a standardized natural language generation model to achieve semantic parsing and interactive script generation in emergency scenarios.

[0034] During the overall operation, after triggering an emergency passage request, vehicle A at the sending end acquires current road environment information through sensors. The AI ​​Agent then performs semantic analysis and scene understanding to generate compliant natural language interactive content. This content is encoded, encapsulated into a communication message through an OBU, and broadcast to surrounding vehicles via the C-V2X PC5 direct link.

[0035] After receiving the broadcast message, the receiving vehicle B completes the deframing of the physical layer and the data link layer through the OBU, and extracts the content of the application layer data. Subsequently, the AI Agent performs semantic parsing on the natural language content, identifies the emergency request type and the collaborative behavior requirement, makes decision generation by combining the environmental information obtained by the vehicle's own sensors, outputs the corresponding automatic driving control instructions, and generates feedback information to be transmitted back through the PC5 link.

[0036] In the continuous collaboration process, a dynamic negotiation mechanism is formed among multiple vehicles through natural language interaction, which continuously updates the collaboration strategy and control behavior, and realizes multi-vehicle linked control in emergency passage scenarios. The whole process does not rely on cellular networks or a centralized scheduling platform, and realizes low-latency and high-reliability collaborative control capability based on vehicle-to-vehicle direct communication and vehicle-side intelligent computing capability.

[0037] The technical solution realizes the direct mapping between natural language semantic interaction and automatic driving control through the integrated design of multi-source sensing, vehicle-mounted communication, distributed AI computing and execution control, forms a closed-loop collaborative control system, and still has stable operation capability under complex traffic environments and weak network conditions. Meanwhile, it has good scalability and adaptability, and can support the deployment requirements of different vehicle models and different control architectures.

[0038] S1, the transmitting vehicle A triggers an emergency passage demand according to a physical trigger button or voice recognition; During actual operation, multiple emergency triggering mechanisms are set inside the transmitting vehicle A to adapt to different usage scenarios. The physical trigger button is directly connected to the vehicle control system via a hard-wired mode, which features rapid response and high reliability, and is suitable for immediate triggering of sudden emergencies; the voice recognition mode collects voice input from the driver or passengers through the cockpit domain controller, extracts semantic information after parsing by the voice recognition model, so as to identify the intention of emergency passage. The voice recognition process includes three stages: voice collection, voice signal processing and semantic parsing. The keywords and context information are identified through the traffic scene semantic understanding model to ensure accurate identification of emergency requests. After triggering is completed, an emergency passage demand identifier in a unified format is generated inside the vehicle, which provides basic input for the subsequent data processing flow.

[0039] S2, the transmitting vehicle A identifies current road condition information according to the sensor A, and the current road condition information includes all traffic participants; Sensor A perceives its surroundings through multi-source fusion, including visual information from a camera, spatial point cloud information from LiDAR, long-range target information from millimeter-wave radar, and short-range obstacle information from ultrasonic radar. The data from these different sensors are processed uniformly in the autonomous driving control unit using a multimodal fusion algorithm to form a structured environment model. This model includes vehicle position, speed, acceleration, direction of travel, and road topology, while also encompassing the dynamic state information of surrounding traffic participants such as vehicles, pedestrians, and non-motorized vehicles. The environment modeling process also includes the identification of road types (such as urban roads, highways, and intersections) and traffic density, providing contextual support for subsequent semantic generation.

[0040] S3. Based on the emergency passage requirement, the passage requirement is transmitted to the OBU via hard-wired signals or analog-digital signals. The AI ​​Agent in the OBU calls the emergency dialogue template and generates natural language text in combination with the current road condition information. Emergency passage requests are transmitted from inside the vehicle to the On-Board Unit (OBU) via CAN bus or Ethernet network. The signal format can be digital or analog, and after protocol conversion by the gateway, it is input to the communication processing unit. An AI Agent is deployed inside the OBU, which integrates the emergency passage request with environmental information by calling a pre-defined emergency dialogue template library. The dialogue templates adopt a standardized structure design, including a fixed semantic framework and populated parameter fields, such as vehicle identification, current location, current speed, and requested action. The AI ​​Agent dynamically populates the templates based on real-time traffic information, generating natural language text that conforms to semantic specifications, while ensuring the text length meets communication link constraints to avoid redundant information affecting transmission efficiency.

[0041] S4, based on natural language text, is lightweightly encoded by the OBU; After natural language text is generated, it enters the encoding stage. The encoding method uses a unified character set format to convert the text into byte stream data. Data length is controlled during encoding to ensure high transmission efficiency during PC5 link transmission. The encoding rules also ensure parsing consistency between different vehicles, enabling the receiving end to quickly recover the original semantic content. The encoding result forms a standardized data format, providing a foundation for subsequent network layer encapsulation.

[0042] S5, according to lightweight coding, OBU encapsulates it into a UDP packet; The encoded data serves as the application layer payload and is encapsulated using a transport layer protocol. The UDP protocol is used to reduce communication latency and simplify the transmission process. During encapsulation, a fixed port number is set to distinguish service types, achieving isolation between emergency communication data and other data services. The UDP packet structure includes the source port, destination port, length, and checksum field; the data payload is the encoded natural language information.

[0043] S6. Based on the UDP packet, the OBU then encapsulates it into an IPv4 broadcast packet with the destination address being the broadcast address of the local network segment. UDP packets are further encapsulated at the network layer, constructing broadcast packets using the IPv4 protocol. The destination address is set to the local network segment's broadcast address, ensuring that all vehicles within the same communication range can receive the information. The IP header contains fields such as source address, destination address, and protocol type, ensuring compatibility between different devices through standard protocol formats. The broadcast method avoids the point-to-point communication establishment process, improving information distribution efficiency.

[0044] S7, based on the IPv4 broadcast message, the OBU broadcasts wirelessly to the outside world through the PC5 Sidelink broadcast channel; The encapsulated IPv4 broadcast message is sent via the PC5 Sidelink link. PC5 communication is based on a distributed resource scheduling mechanism, enabling direct communication between vehicles without relying on base stations. The broadcast channel operates in the 5.9GHz band, featuring low latency and high reliability, making it suitable for short-range, multi-node communication scenarios. Data is modulated at the physical layer and then transmitted to the air channel, enabling real-time reception by nearby vehicles.

[0045] S8, after the OBU in the receiving vehicle B receives the broadcast message, it performs message parsing; The receiving vehicle B listens to the broadcast channel via the PC5 interface. After receiving data, the physical layer demodulates it, the link layer performs frame parsing, and extracts the IP data packets. The OBU parses the IP header, identifies the UDP protocol data, extracts the application layer payload, and recovers the encoded byte stream data.

[0046] S9, the OBU in the receiving vehicle B performs semantic recognition and intent parsing on the message; The decoded natural language text is then passed to the AI ​​Agent for semantic processing. The semantic recognition process includes keyword extraction, syntactic analysis, and intent recognition. A traffic semantic model is used to determine the request type, such as slowing down, changing lanes, or yielding. Intent parsing simultaneously identifies request parameters, such as target speed, offset distance, and duration, providing input for subsequent decisions.

[0047] S10, based on semantic recognition and intent parsing, combined with road condition information and vehicle operating status collected by sensor B in the receiving vehicle B, generate corresponding natural language response text and autonomous driving cooperative control instructions. The receiving vehicle B generates decisions based on its own environmental perception data and analysis results. The AI ​​Agent assesses the current traffic environment and vehicle status, determines whether the conditions for collaborative operation are met, and generates corresponding control strategies. The control strategies include actions such as deceleration, lane changing, and maintaining a safe distance, while generating feedback natural language text indicating the execution intention and estimated completion time.

[0048] S11, based on the natural language response text and the autonomous driving cooperative control instructions, the OBU in the receiving vehicle B sends a PC5 UDP broadcast message back to the sending end, and at the same time, the receiving vehicle B performs an avoidance operation according to the autonomous driving control instructions. Feedback information is encoded and encapsulated using the same methods as the transmission process to form broadcast messages, which are then sent via the PC5 link. Simultaneously, control commands are transmitted to the actuators via the vehicle bus to achieve vehicle motion control. Environmental changes are continuously monitored during execution to ensure safe and reliable operation.

[0049] S12, after receiving the feedback message, vehicle A at the sending end combines the current road condition information and forms a path plan through the AI ​​Agent, and simultaneously broadcasts driving information to the outside world for cooperative driving. After receiving the feedback information, vehicle A at the sending end performs comprehensive analysis through the AI ​​Agent and updates its route planning strategy. Route planning considers the response status of surrounding vehicles, dynamically adjusting the driving trajectory and speed to achieve cooperative passage. The updated driving information is continuously broadcast through the communication link to maintain the cooperative state of multiple vehicles.

[0050] S13, according to the path planning, the OBU AI Agent in the sending vehicle A sends standardized control commands to the actuator through the vehicle's internal bus to complete the cooperative autonomous driving operation.

[0051] The path planning results are translated into specific control commands, which are transmitted to the actuators, including the powertrain, braking system, and steering system, via the onboard network. The actuators then perform vehicle control actions according to the commands, while simultaneously feeding back execution status information for closed-loop control, ensuring continuous and stable operation during cooperative passage.

[0052] I. Extended Implementation Methods for S1 In some implementations, S1 also includes: Based on the physical trigger button, the system is manually activated by the user with a single key press. After being input via a hardwired line, the system identifies the emergency passage requirement. The voice recognition algorithm analyzes the meaning of the user's language to identify the emergency passage requirement.

[0053] In this implementation, the emergency triggering mechanism adopts a dual-channel input structure to improve triggering reliability and applicability. The physical trigger button is directly connected to the vehicle control system via hardware circuitry, resulting in a short trigger path and fast response, suitable for drivers or passengers to quickly initiate emergency requests in emergency situations. The button signal is transmitted into the vehicle network via hardwired connection, converted into a standard control signal by the gateway node, and then transmitted to the communication unit, ensuring that the triggering information is not affected by software delays or system load.

[0054] The voice recognition triggering path utilizes the cockpit domain controller to complete voice acquisition, voice preprocessing, and semantic recognition. The voice signal first undergoes noise reduction and enhancement, then is converted into text information by a voice recognition algorithm, and finally, the semantic understanding model performs intent recognition. During recognition, key semantics such as emergency, yielding, and rescue are extracted, and semantic confirmation is performed in conjunction with contextual information to identify emergency passage requests. Both triggering methods are managed uniformly at the logic layer. When either triggering path identifies a valid emergency signal, a unified format emergency request identifier is generated and the process proceeds to the next stage.

[0055] II. Extended Implementation Methods for S3 In some implementations, S3 also includes: S31, based on the emergency script template, adopts a fixed script template for limited scenarios and a simplified filling mode for core parameters, standardizing, formatting and lightweighting the script template according to the scenario; S32, based on the AI ​​Agent, is deployed locally offline, does not rely on the public network, and can work offline.

[0056] In this implementation, the natural language generation mechanism adopts a template-driven model. The template structure consists of a fixed semantic framework and parameter-filled areas, such as fields for vehicle identification, current location, speed, and requested action. By designing constrained scenarios, the complexity and scope of expression are limited, ensuring semantic clarity and ease of parsing. Templates are categorized according to traffic scenarios, such as urban roads, highways, congested scenarios, and intersections, with each scenario corresponding to a different expression format, thereby improving the accuracy of semantic expression.

[0057] The core parameters are simplified and only essential information fields are retained to avoid redundant content that could increase communication overhead. Character length is controlled during text generation to ensure compliance with PC5 link transmission constraints. Standardization and format normalization processes ensure consistent semantic expression across different vehicles, avoiding parsing ambiguity.

[0058] The AI ​​Agent is deployed inside the vehicle's controller, operating locally offline and without relying on external network resources. Semantic generation and processing are both completed on the vehicle side, avoiding the latency and uncertainties associated with cloud communication. It can still operate normally in weak or no network environments, ensuring the continued effectiveness of emergency coordination functions.

[0059] III. Extended Implementation Methods for S4 In some implementations, S4 also includes: The lightweight encoding uses UTF-8 encoding format.

[0060] During the encoding process, natural language text is uniformly converted to UTF-8 encoding format, which features strong cross-platform compatibility and high encoding efficiency. The encoding process converts character sequences into byte stream data while maintaining semantic integrity. By controlling the encoding length, data consumes less bandwidth resources during transmission, improving communication efficiency.

[0061] The encoding rules are consistent across all participating vehicles, allowing the receiving end to directly decode using the same rules and recover the original text information. The encoded data structure is simple and clear, facilitating subsequent protocol encapsulation and parsing.

[0062] IV. Extended Implementation Methods for S10 In some implementations, S10 also includes: S101, according to the autonomous driving cooperative control instruction, this instruction is generated by the vehicle-side autonomous driving algorithm combining emergency passage needs and road condition information identified by the current sensors, and the final planning result is obtained after fusion. S102 adopts a fixed dialogue template and core parameter simplification filling mode based on the natural language response text, which standardizes, normalizes the format, and makes the dialogue template lightweight according to the scenario.

[0063] In this implementation, after completing semantic parsing, the receiving vehicle enters the collaborative decision-making stage. The autonomous driving control unit calculates the optimal response strategy based on the current environment model and the emergency request content. During the decision-making process, factors such as the vehicle's current speed, position, surrounding traffic density, and road structure are comprehensively considered, and an executable control scheme is generated through a path planning algorithm, such as deceleration, lane changing, or maintaining a safe distance.

[0064] After the control command is generated, the response content is constructed synchronously with the natural language feedback information. The natural language response text is also generated using a template-based method, with a structure consistent with the sender to ensure semantic symmetry. The text includes the executed action, response time, and status information, facilitating the sender's understanding of the response and further adjustments to its strategy.

[0065] V. Extended Implementation Methods for S13 In some implementations, S13 also includes: Based on the collaborative autonomous driving operation, the autonomous driving operation is carried out continuously and without interruption through information from C-V2X, single vehicle intelligence, cloud platform information and other aspects. It is not a single collaborative operation, but a continuous negotiation of emergency passage procedures based on actual road needs.

[0066] In this implementation, the collaborative control process employs a continuous interaction mechanism, rather than a one-time execution. During the execution of control commands, the vehicle continuously receives information from surrounding vehicles and updates the environmental model in real time. Status data is constantly exchanged via the C-V2X communication link to maintain the collaborative relationship between multiple vehicles.

[0067] The vehicle-mounted intelligent system provides real-time perception and local decision-making capabilities, with the onboard computing unit continuously processing information on environmental changes. When network conditions are available, the cloud platform can provide auxiliary information support, such as traffic status or global dispatch information. Multi-source information is integrated and processed internally to form dynamically updated collaborative strategies.

[0068] The coordinated passage process is continuously adjusted based on actual road conditions until the emergency vehicles complete their passage mission. Throughout the process, communication, perception, decision-making, and execution form a closed loop, ensuring the stability and reliability of the coordinated process and preventing it from being affected by single points of failure or network fluctuations.

[0069] Extended Explanation of Natural Language Text Generation, Transmission, and Parsing Methods: The AI ​​Agent of vehicle A at the sending end calls a predefined traffic emergency scenario restricted natural language dialogue template and fills in the vehicle information; Example of an emergency request: Emergency; This vehicle's IP: 192.168.88.10, current speed 60km / h, location longitude X, latitude Y, requesting neighboring vehicles to cooperate, reduce speed to 30km / h, move to the right by 0.5m, and continue to give way until this vehicle passes; Positive response template: Reply; Confirm receipt of emergency request. This vehicle's IP: 192.168.88.25. Will decelerate to 30km / h within 3 seconds, move 0.5m to the right, and notify simultaneously after completing the avoidance maneuver. The OBU of the sending vehicle A uses UTF-8 encoding to convert the emergency request content into a byte stream; The OBU of the sending vehicle A encapsulates the byte stream into UDP packets, specifies a fixed V2X application port when encapsulating the UDP packets, and executes the packet encapsulation initialization, IP header encapsulation, UDP header encapsulation, application layer data encapsulation, packet verification and sending process; the OBU of the sending vehicle A then encapsulates the UDP packets into IPv4 broadcast packets, with the destination address being the broadcast address of the local network segment. The OBU of vehicle A broadcasts IPv4 broadcast messages wirelessly through the Sidelink broadcast channel. The OBU of the receiving vehicle B listens to a fixed UDP port and broadcast messages across the network; the OBU hardware completes PC5 physical layer and link layer frame deframing and extracts standard IP data packets; the protocol stack parses the IP header and UDP header, strips off the protocol encapsulation, and restores the original natural language text; The AI ​​Agent of the receiving vehicle B directly reads the plaintext text and performs semantic recognition and intent understanding; The AI ​​Agent of the receiving vehicle B combines the current road condition information to generate natural language responses and autonomous driving control commands. The natural language responses are transmitted via PC5 unicast or broadcast UDP messages.

[0070] In this implementation, the natural language communication mechanism is built upon a unified design principle of templated expression, lightweight transmission, and rapid parsing. The script template library is predefined based on traffic emergency scenarios, with each template corresponding to a specific collaborative scenario, such as emergency yielding, lane avoidance, and speed adjustment. Each template consists of a fixed semantic structure and parameter-filled areas. Parameter fields include vehicle identification information, current location coordinates, real-time vehicle speed, target action parameters, and duration. This fixed structure and parameter-filling method ensures standardized characteristics in the generated text, facilitating rapid understanding and processing by different vehicles.

[0071] The emergency request content example uses a unified identifier prefix "emergency" to quickly distinguish business types, while embedding key parameters in a structured form within the natural language text. This approach ensures semantic clarity while balancing human readability and machine parsing. The affirmative response template uses a response identifier prefix to clearly define feedback attributes and includes execution plans and action confirmation information, thus forming a closed-loop semantic structure of request and response.

[0072] In the encoding stage, UTF-8 encoding is used to convert text into a byte stream. This encoding method features strong cross-platform consistency and high compatibility. During the encoding process, by controlling the character length and content structure, the generated data meets the PC5 link's constraints on message size. The byte stream, as the application layer data payload, enters the transport layer for encapsulation processing.

[0073] UDP packet encapsulation employs a fixed port design to isolate emergency communication services from other vehicle-mounted data services. The fixed port serves as a service identifier; the receiving end can quickly identify relevant data by listening on this port, thereby improving parsing efficiency. The encapsulation process includes steps such as initializing the packet structure, filling in IP header information, constructing UDP header fields, writing application layer data, and performing verification processing. This process ensures the integrity and reliability of data during transmission.

[0074] During the network layer encapsulation phase, UDP packets are further encapsulated into IPv4 broadcast packets. The broadcast address is set to the local network segment's broadcast address, ensuring that all vehicles within the same communication range can receive the information, avoiding the connection establishment delays associated with point-to-point communication. This broadcast mechanism enables rapid information dissemination and improves collaborative response efficiency.

[0075] Wireless transmission is accomplished via the PC5 Sidelink link, which employs a distributed resource scheduling approach, ensuring stable communication even in base station-less environments. The broadcast channel supports multi-node reception, making it suitable for multi-vehicle collaborative scenarios. The sending vehicle continuously broadcasts emergency information through this link, ensuring surrounding vehicles receive requested information promptly.

[0076] The receiving vehicle acquires data by listening to a fixed UDP port and broadcast address. The hardware layer performs signal demodulation and frame parsing, while the software protocol stack parses the IP and UDP headers to extract the original application layer data. The decoding process follows unified encoding rules to recover the natural language text content.

[0077] The AI ​​Agent directly reads plaintext and processes it using a semantic recognition model to extract key semantic information and behavioral intent. The semantic parsing process includes keyword recognition, syntactic structure analysis, and intent classification to determine the request type and execution requirements. Combined with current environmental information, the AI ​​Agent generates response text and control commands, which are then fed back through the communication link to achieve two-way interaction.

[0078] Extended Explanation of AI Agent Collaborative Workflow Methods: Understanding traffic scene semantics: By parsing sensor data, we can accurately understand the current traffic scene (such as urban roads, highways, congested road sections, and intersections), providing scene support for subsequent emergency event identification and control command generation. Understanding the semantics of emergency events: By parsing C-V2X interactive text and sensor data, we can accurately identify the type of emergency event (patient, rescue, accident) and extract core information about the emergency event (such as event location, needs, and priority) to provide a basis for script generation and control instructions.

[0079] Generate standardized natural language interaction scripts: Based on the script generation capabilities of the AI ​​Agent, combined with the semantics of emergency events and traffic scenarios, standardized and lightweight natural language interaction scripts are generated, which are adapted to PC5 sidechain UDP packet transmission (no redundant characters, and the number of bytes meets the limit), ensuring that the collaborative vehicle AI Agent can quickly parse them.

[0080] Transforming dialogue semantics into autonomous driving control commands: The AI ​​Agent transforms the semantics of V2X interaction dialogue (such as requesting a neighboring vehicle to slow down to 30km / h and shift to the right by 0.5m) into autonomous driving control commands (lane change, deceleration, distance control) that can be recognized by the on-board actuators, thereby achieving collaborative control by the AI ​​Agent.

[0081] In this collaborative process, traffic scene semantic understanding serves as a fundamental step, constructing an environmental model through the fusion of multi-source sensor data. This environmental model encompasses not only spatial information but also the behavioral characteristics of traffic participants and road attribute information. This comprehensive understanding of the environment provides the data foundation for subsequent semantic analysis and decision generation.

[0082] Semantic understanding of emergency events is achieved by fusing communication data and sensory data. Communication data provides event description information, while sensory data provides objective environmental conditions; combining the two improves recognition accuracy. By determining the event type and priority, response strategies and the degree of coordination can be identified.

[0083] Natural language generation is achieved through a combination of template-based and parameter-based methods. Templates ensure a consistent expression structure, while parameter filling ensures information completeness. Lightweight processing reduces redundant characters, making the text adaptable to communication bandwidth limitations. The generated text is not only used for communication but also participates in collaborative control as part of the decision-making input.

[0084] The mapping from semantics to control commands is a crucial step in the collaborative process. Through semantic parsing, abstract requests are transformed into concrete control actions. For example, a deceleration request corresponds to a speed control command, and a lane departure request corresponds to a steering control command. This mapping process relies on the vehicle control model, converting high-level semantics into low-level execution parameters, thereby enabling direct control of the autonomous driving system.

[0085] Throughout the collaborative process, the AI ​​Agent establishes a distributed collaborative relationship among different vehicles, achieving dynamic collaborative control among multiple vehicles through continuous information exchange and semantic understanding. Communication, perception, and control form a closed-loop structure, enabling the system to maintain stable operation in complex traffic environments.

[0086] This invention achieves fine-grained isolation and identification of business data at the communication mechanism level by fixing a dedicated V2X UDP communication port and using the port number as a unique identifier for the application layer. During vehicle-to-vehicle direct communication, different business data are distinguished through the port, effectively avoiding interference problems present in traditional mixed data transmission and improving the stability and reliability of the communication link. Simultaneously, the fixed port design allows the receiving end to quickly locate the target data stream, reducing parsing complexity and processing latency. It maintains stable performance even in complex traffic environments and high-concurrency communication scenarios, further enhancing the security and controllability of V2X communication and providing a reliable data transmission foundation for emergency collaborative control.

[0087] This invention employs a lightweight model architecture, deploying key capabilities such as perception processing, semantic understanding, event recognition, dialogue generation, and decision mapping entirely within the vehicle-side domain controller. This eliminates the need to rely on large cloud-based models for inference, effectively avoiding latency fluctuations and uncertainties associated with public network communication. Rapid response is achieved through local computing on the device side. Combined with a fixed UDP port and IPv4 broadcast encapsulation mechanism, emergency requests can complete a closed-loop control process—from generation and transmission to parsing and execution—within a very short time. This significantly reduces the overall system response time, meeting the stringent requirements of low latency and high reliability in autonomous driving scenarios and significantly improving real-time coordination capabilities in emergency passage situations.

[0088] This invention constructs a unified vehicle-to-vehicle and vehicle-to-infrastructure (V2X) interaction protocol paradigm by designing a fixed-byte-length V2X interaction template and a standardized UTF-8 byte stream encapsulation format. This scheme ensures consistent communication content structure and clear semantics through template-based constraints and parameterized filling of natural language expressions, while balancing transmission and parsing efficiency. Different vehicle manufacturers, different domain control hardware platforms, and different perception algorithm systems can all send, receive, and parse data based on this unified standard, fundamentally solving problems such as inconsistent protocols, semantic incompatibility, and non-standard message formats in existing technologies. This significantly improves the compatibility and interoperability of vehicle networking systems, providing a standardized foundation for large-scale commercial applications.

[0089] This invention employs a modular architecture design with model layering, domain control deployment, port fixing, and template standardization. This achieves decoupling and collaboration among the perception, communication, decision-making, and control functional modules, giving the overall technical solution excellent scalability and adaptability. This architecture is compatible with mainstream perception systems, V2X communication modules, and autonomous driving domain controllers, supporting rapid deployment and flexible configuration across different vehicle platforms, reducing system integration difficulty and development costs. Furthermore, the technical solution reserves interface expansion capabilities during the design phase, supporting subsequent vehicle-to-cloud collaboration, enhanced information security, and the evolution of higher-level autonomous driving functions, providing a solid foundation for the continuous upgrading of future intelligent transportation systems.

[0090] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A multi-vehicle AI Agent natural language negotiation emergency passage control method based on C-V2X PC5, characterized in that, Includes the following steps: The sending vehicle triggers the emergency passage request through a physical trigger button or voice recognition, and obtains current road environment information and vehicle operating status information; Based on emergency passage needs, current road environment information, and vehicle operation status information, the sending end vehicle in AIAgent calls a preset script template to generate natural language text, and after lightweight encoding of the natural language text, it is encapsulated into UDP packets, and then further encapsulated into IPv4 broadcast packets. The IPv4 broadcast message is broadcast to the outside world through the C-V2X PC5 direct communication channel. The receiving vehicle receives the IPv4 broadcast message and parses it to obtain natural language text. The receiving vehicle's in-vehicle AI Agent performs semantic recognition and intent parsing on the natural language text, and generates natural language response text and corresponding autonomous driving control commands based on the parsing results. At the same time, it sends feedback messages through the C-V2X PC5 direct communication channel. After receiving the feedback message, the sending vehicle updates the route plan and generates collaborative control instructions based on the feedback content. It then drives the vehicle to execute emergency passage control through the vehicle control link, thereby realizing collaborative passage between multiple vehicle AI Agents based on natural language negotiation.

2. The emergency passage control method based on C-V2X PC5 and multi-vehicle AI Agent natural language negotiation according to claim 1, characterized in that, A unified identification mechanism for multi-source trigger information is constructed, integrating physical input and speech semantic processing paths to form a consistent emergency demand output logic. The steps are as follows: The signal level change generated by the physical trigger button is collected, and the signal is sampled by the analog-to-digital converter unit to convert the continuous analog signal into a discrete digital signal and generate the corresponding trigger status flag bit. The system acquires the voice input data stream, performs pre-emphasis, framing, and windowing processing on the voice signal, extracts Mel frequency cepstral coefficient features, outputs text results through a speech recognition model, and then extracts intent labels through a semantic parsing model. The trigger status flag and intent label are mapped to fields to build a unified data structure, including trigger source field, trigger type field and timestamp field, and data alignment is completed. A unified data structure is written to the shared cache, encapsulated into an emergency request data frame according to a preset message format, and pushed to the communication processing queue.

3. The multi-vehicle AI Agent natural language negotiation emergency passage control method based on C-V2X PC5 according to claim 1, characterized in that, By introducing a structured script generation method and combining scene classification and parameter filling strategies, natural language interactive content with a unified expression format is formed. The steps are as follows: Establish a script template data table. Each template includes a fixed statement segment, a variable parameter segment, and field identifiers. Assign template index numbers to different traffic scenarios. Read the current environment status identifier and match the template index, then write the vehicle IP address, latitude and longitude coordinates, speed value, and target action parameters into the corresponding field positions; Perform character-level processing on the filled text, including special character replacement, field order rearrangement, removal of empty fields, and pruning of redundant characters; The generated text is converted to a standard string format and written to the output buffer, along with a text length field and a validation flag.

4. The multi-vehicle AI Agent natural language negotiation emergency passage control method based on C-V2X PC5 according to claim 1, characterized in that, A unified text encoding processing flow is constructed, which, through character conversion and data structure organization, forms a data format suitable for communication transmission. The steps are as follows: Read standard string data and match it character by character according to the UTF-8 encoding rules, mapping each character to the corresponding byte sequence; Perform a concatenation operation on the byte sequence to generate a continuous byte stream, and record the start and end offset addresses; The length of the byte stream is counted and written to the data header, while boundary identifiers are added to mark the start and end positions of the data. Write the byte stream to the send buffer and generate a data description structure for subsequent encapsulation processes.

5. The multi-vehicle AI Agent natural language negotiation emergency passage control method based on C-V2X PC5 according to claim 1, characterized in that, The transport layer encapsulation process is constructed by organizing port identifiers and message structures to form identifiable data transmission units. The steps are as follows: Initialize the UDP packet structure, allocate storage space for the header and data fields, and write a fixed application port number; Copy the encoded byte stream to the UDP data area and record the data area length and offset. Fill in the UDP header fields, including the source port number, destination port number, data length field, and checksum field; Perform checksum calculations on the entire UDP packet and generate a complete packet object, which is then stored in the network sending queue.

6. The multi-vehicle AI Agent natural language negotiation emergency passage control method based on C-V2X PC5 according to claim 1, characterized in that, A network layer broadcast mechanism is established through address configuration and message construction to complete the multi-node data distribution process. The steps are as follows: Construct the IP packet structure and write the UDP packet as payload data into the data area; Set the IP header fields, including source IP address, broadcast destination address, protocol type field, and identifier field; Perform fragmentation flag configuration and header checksum calculation on IP packets to form complete IP data packets; The IP data packet is written into the physical layer transmit buffer and broadcast through the PC5 interface.

7. The multi-vehicle AI Agent natural language negotiation emergency passage control method based on C-V2X PC5 according to claim 1, characterized in that, The receiving end layered parsing process is constructed, and the original semantic content is recovered through protocol parsing and data extraction. The steps are as follows: The radio frequency module receives wireless signals and performs demodulation processing, converting the signals into digital data frames; Perform link-layer deframe operation on the data frame to identify the frame header and trailer and extract the IP data segment; Parse the IP header fields and identify the protocol type; further parse the UDP header and locate the start of the data area. Perform UTF-8 decoding on the byte stream of the data area, restore it to string text, and write it to the semantic processing buffer.

8. The multi-vehicle AI Agent natural language negotiation emergency passage control method based on C-V2X PC5 according to claim 7, characterized in that, The system performs validation rule matching on the IP header fields to confirm the validity of the message structure, performs consistency checks on the UDP header length field to determine the data area range, performs boundary identifier recognition and continuity checks on the byte stream, and performs character validity checks on the decoding results and writes them to the semantic processing buffer for subsequent semantic parsing.

9. The multi-vehicle AI Agent natural language negotiation emergency passage control method based on C-V2X PC5 according to claim 1, characterized in that, The process of mapping semantics to control commands is constructed through intent recognition and parameter extraction to generate vehicle control information. The steps are as follows: Perform word segmentation on the received text and build a word sequence, identifying action keywords and parameter keywords; Extract parameter fields, including target velocity value, lateral offset distance, execution time window, and priority indicator; The extracted parameters are input into the control decision model, and control trajectory data is generated through a path planning algorithm. The trajectory data is converted into a set of control commands, including acceleration, braking and steering commands, and output to the execution control interface.

10. The multi-vehicle AI Agent natural language negotiation emergency passage control method based on C-V2X PC5 according to claim 1, characterized in that, A multi-vehicle collaborative operation process is constructed through status updates and strategy adjustments to maintain continuous collaborative passage. The steps are as follows: It receives feedback messages from the communication link and parses out the status information of neighboring vehicles, including their position, speed, and action status. The status information of neighboring vehicles is fused with the perception data of this vehicle to update the environmental model data structure; The updated environment model is recalculated to replan the execution path and generate a new sequence of control parameters. The control parameter sequence is sent to the execution system and status broadcast data is generated synchronously and written into the communication sending queue for subsequent interaction.

Images