High-safety scene-driven decision-making control integrated automatic driving system with high-precision positioning support

Abstract

The invention discloses a high-precision positioning-supported high-safety scene-driven decision control integrated automatic driving system, which comprises a satellite platform, a cloud platform and a vehicle-end platform, and is characterized in that the vehicle-end platform uploads an observation sequence and positioning information to the cloud platform; the cloud platform communicates with the vehicle end platform and the satellite platform, the cloud platform carries a data safety monitoring framework, a scene causal reasoning framework and a fast and slow knowledge management framework, and the data safety monitoring framework performs data safety monitoring; the scene causal reasoning framework is used for summarizing and managing scene knowledge at the cloud; the fast and slow knowledge management framework comprises a fast system and a slow system; and the satellite platform performs satellite communication with the cloud platform, complements the cellular network through satellite communication, and provides communication support.

Description

Technical Field

[0001] The present invention relates to the fields of vehicle engineering and transportation engineering, and relates to a high-precision positioning-supported, high-security, scenario-driven, decision-making and control-integrated automatic driving system. Background Art

[0002] High-level autonomous driving is the technological commanding height of smart cars. High-level autonomous driving, represented by navigation assisted driving (NOA), achieves safe and efficient point-to-point autonomous driving through real-time environmental perception and path planning technology. High-precision positioning and stable vehicle-backend / cloud interconnection can greatly enable the improvement of autonomous driving levels. Centimeter-level positioning accuracy can not only ensure that the vehicle can accurately identify its own position in complex driving environments, but also provide support for path planning, environmental perception, and driving control of autonomous vehicles through close integration with high-precision map information. With the continuous advancement of high-precision positioning technology, the capabilities of high-level autonomous driving will continue to improve, achieving a higher level of smart travel.

[0003] However, the existing autonomous driving system based on cellular network vehicle networking technology (CV2X) relies on the distribution and density of ground base stations for its communication coverage, lacks backup communication means, and has weak communication signals and coverage blind spots in remote areas where base stations cannot fully cover, and cannot provide stable and reliable positioning services. At the same time, the autonomous driving system based on cellular network vehicle networking technology (CV2X) lacks sufficient real-time communication security monitoring and response mechanisms, and there is a risk of sensitive information leakage, which limits its safe and effective application in various environments and driving scenarios. Summary of the invention

[0004] In order to address the deficiencies in the prior art, this application proposes a high-security scenario-driven decision-control integrated autonomous driving system supported by high-precision positioning. Through the integration of low-orbit satellite signals and cellular networks, a seamless coverage, high-reliability ground-to-earth communication system is formed to solve the problems of network coverage and communication stability. Based on reliable and high-precision positioning under the ground-to-earth communication system, an integrated decision-making and control framework driven by high-security scenarios is constructed to solve information security issues, providing a new solution for autonomous driving in large-scale scenarios.

[0005] The technical solution adopted by the present invention is as follows:

[0006] High-security scenario-driven decision-making and control integrated autonomous driving system supported by high-precision positioning, including satellite platform, cloud platform and vehicle-side platform.

[0007] The vehicle-side platform uploads the observation sequence and positioning information to the cloud platform;

[0008] The cloud platform communicates with the vehicle-side platform and the satellite platform respectively. The cloud platform is equipped with a data security monitoring framework, a scene causal reasoning framework, and a fast and slow knowledge management framework. The data security monitoring framework performs data security monitoring; the scene causal reasoning framework is used to summarize and manage scene knowledge in the cloud; the fast and slow knowledge management framework includes a fast system and a slow system;

[0009] The satellite platform performs satellite communications with the cloud platform, complements the cellular network through satellite communications, and provides communication support.

[0010] Furthermore, the data security monitoring framework performs a large language model data analysis process, a data transmission behavior baseline setting process, and a data transmission activity qualitative analysis process.

[0011] Furthermore, the large language model data analysis process is to use the large language model to analyze the data content, extract key information to establish a data information library, and realize data classification and sensitivity rating.

[0012] Furthermore, the data transmission behavior baseline setting process performs anomaly detection on the behavior data and adjusts the baseline setting based on the anomaly detection result.

[0013] Furthermore, the qualitative analysis process of data transmission activities adopts a transformer-based causal reasoning model to perform qualitative analysis on time-series data transmission activities.

[0014] Furthermore, the scenario causal reasoning framework includes a cross-scenario training sequence collection process, a scenario causal reasoning model training process, and a scenario experience pool construction process, which summarizes and manages scenario knowledge in a cloud platform.

[0015] Furthermore, the cross-scenario training sequence collection process uploads the cross-scenario training historical sequences of the intelligent connected vehicles in the vehicle-side group to the cloud platform.

[0016] Furthermore, the scene causal reasoning model training process trains the scene causal reasoning model based on the historical training process uploaded by the vehicle-side group.

[0017] Furthermore, in the process of constructing the scenario experience pool, the observation sequence and high-precision positioning information of the intelligent connected vehicles in the vehicle-side group are used as input, and the process training scenario causal reasoning model is used to predict the scenario factors of the driving scenario in which the intelligent connected vehicles are located to construct the scenario experience pool.

[0018] Furthermore, the fast system pushes the decision-making control integrated neural network parameters corresponding to the predicted scene causality and high-precision positioning query scene to the intelligent connected vehicle through the scene-driven experience push based on federated learning;

[0019] The slow system monitors the vehicle dynamic sequence for real-time safety risk assessment and emergency strategy switching.

[0020] Beneficial effects of the present invention:

[0021] (1) Through the integration of low-orbit satellite signals and cellular networks, a highly reliable space-ground integrated communication system with seamless connection and global coverage has been formed. Under this system, the transmission of parameters of the federated learning neural network effectively reduces the risk of exposure during the communication process, solves the problem of information islands caused by privacy awareness, and realizes a highly secure communication process.

[0022] (2) Through the classification of communication data content based on the large language model (LLM), the identification of highly sensitive data transmission activities, and the qualitative analysis of data transmission activities, a high-security data transmission process is ensured; by building a scenario experience pool, the unified integration and storage of high-precision scenario experience is achieved, ensuring a high-security data management process.

[0023] (3) Construct a fast and slow knowledge management framework, push scenario-driven experience through the fast system, quickly retrieve the scenario experience pool based on high-precision positioning, and extract the best practices that have been verified in the corresponding scenarios, providing an efficient and flexible system access and scenario strategy switching method for intelligent connected vehicles.

[0024] (4) Maintain and store high-security emergency response strategies in the scene experience pool, monitor the dynamic sequence of vehicles through the slow system, and realize real-time safety risk assessment and emergency strategy switching; based on vehicle-side sensor data and macro traffic information, through the slow system incremental continuous learning, synchronous training optimization, evaluation and screening of autonomous driving models, and continuously update the scene experience pool based on high-precision positioning, realizing high-security strategy updates supported by high-precision positioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 Schematic diagram of the high-safety scenario-driven decision-making and control integrated autonomous driving system proposed in the present invention with high-precision positioning support.

[0026] Figure 2 Schematic diagram of the intelligent connected vehicle data flow used in the present invention.

[0027] Figure 3 Schematic diagram of the fast and slow knowledge management framework based on federated learning proposed in the present invention. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0029] Combined with Figure 1-3 The present invention proposes a high-security scenario-driven decision-making and control integrated automatic driving system supported by high-precision positioning, including a satellite platform, a cloud platform and a vehicle-side platform. The functions of each part are as follows:

[0030] (1) Satellite platform

[0031] The satellite platform includes high-orbit and low-orbit satellites and satellite ground receiving stations (SGS). Satellite communications are carried out between the high-orbit and low-orbit satellites and the satellite ground receiving stations. Satellite communications complement the cellular network to form a seamless coverage, high-reliability integrated space-ground communication network, providing communication support for scene-driven decision-making and control integration.

[0032] (2) Cloud Platform

[0033] The cloud platform communicates with the vehicle-side platform and the satellite platform respectively; the cloud platform is equipped with a data security monitoring framework, a scenario causal reasoning framework, and a fast and slow knowledge management framework; it monitors data communication behavior through big data analysis, provides a high-security data communication process, and provides knowledge summary and management for scenario-driven decision-making and control integration.

[0034] (2.1) Data security monitoring framework

[0035] The data security monitoring framework uses a large language model (LLM) to perform data security monitoring and build a high-security data communication system. The data security monitoring framework performs a large language model (LLM) data analysis process, a data transmission behavior baseline setting process, and a data transmission activity qualitative analysis process.

[0036] Furthermore, the LLM data analysis process uses the LLM to analyze the data content, extract key information to establish a data information library, and realize data classification and sensitivity rating. The data content includes but is not limited to vehicle-related data (such as positioning information, driving logs, fault diagnosis information, etc.), passenger-related data (such as personal information, payment information, communication information, behavior data, etc.), and transportation infrastructure data (monitoring data, road network data, etc.). The extraction of key information uses the natural language understanding ability of the LLM to perform semantic analysis and context understanding on the input data, identify key elements such as entities, relationships, events in the text, and understand the overall meaning and background of the data. Finally, based on the understanding of the data, an output containing key information is generated. The data classification process constructs a multi-layer classification system through the attributes, uses, formats, and sources of the data content, and divides the data into vehicle-related data, passenger-related data, and transportation infrastructure data according to major categories, and then subdivides them according to major categories to form a detailed classification link. The sensitivity rating is first set manually according to the impact of data leakage, with low, medium and high sensitivity ratings, and the sensitivity rating results are fitted through the data-driven large language model (LLM). In the process of data classification and rating, the large language model (LLM) undertakes data analysis, information condensation, and automatic classification and labeling.

[0037] Furthermore, the qualitative analysis process of data transmission activities mainly adopts a transformer-based causal reasoning model to achieve The transmission activity was qualitatively analyzed.

[0038] The timing data transmission activities are:

[0039]

[0040] in, represents the sequential data transmission activity of the nth intelligent connected vehicle, IP represents the address of the data receiving method, t T Indicates the type of data transmitted at time T, b T represents the size of the data packet transmitted at time T, f T represents the data transmission frequency at time T, d T represents the data transmission duration at time T, s T Indicates the sensitivity rating of the data transmitted at time T.

[0041] The qualitative analysis of the time series data transmission activities mainly uses a manual labeling method to characterize the time series data transmission activities into data transmission activities that require intervention and data transmission activities that do not require intervention, and fits the qualitative analysis results of the time series data transmission activities through a causal reasoning model.

[0042] (2.2) Scenario Causal Reasoning Framework

[0043] The scenario causal reasoning framework is mainly used to summarize and manage scenario knowledge in the cloud. The scenario causal reasoning framework includes a cross-scenario training sequence collection process, a scenario causal reasoning model training process, and a scenario experience pool construction process.

[0044] Furthermore, the cross-scenario training sequence collection process is to upload the cross-scenario training history sequence of the intelligent connected vehicles in the vehicle-side group to the cloud through V2N communication for unified management and processing. The training history sequence includes the historical training process of the vehicle-side intelligent connected vehicles:

[0045]

[0046] in, represents the historical training process of the nth intelligent connected vehicle, o T represents the state of the intelligent connected car at time T, a T represents the control output of the intelligent connected vehicle at time T, r T Represents the reward function feedback of the environment to the intelligent connected vehicle at time T.

[0047] Furthermore, the scene causal reasoning model training process mainly constructs a scene causal reasoning model and trains it based on the historical training process uploaded by the vehicle-side group. The causal reasoning model adopts a transformer-based cross-attention structure:

[0048] Q=w q ′ q,K=w k ′ f,V=w v ′ f

[0049]

[0050] Q′=w″(AV)

[0051] Where Q, K and V represent query, key and value respectively, w q ′,w′ k , w v ′ and w″ denote learnable parameters, A denotes the attention map, h denotes the number of multi-head attentions, and Q′ denotes the output query.

[0052] Furthermore, the process of constructing the scenario experience pool is mainly to predict the scenario factors of the driving scenario in which the intelligent networked vehicles are located through the trained scenario causal reasoning model, based on the observation sequence of the intelligent networked vehicles in the vehicle-side group, combined with high-precision positioning information, to construct a scenario experience pool, and store scenario experiences covering positioning scenarios, scenario factors, and corresponding decision-making and control integrated neural network parameters. The scenario experience:

[0053] E=[p lat ,p long ,ε,φ * ]

[0054] Among them, p lat Indicates the current vehicle's latitude, p long represents the current vehicle's longitude, ε represents the scene factor of the current driving scene of the intelligent connected vehicle, φ * Represents the decision-making and control integrated neural network parameters corresponding to the current environment of the intelligent connected vehicle.

[0055] (2.3) Fast and Slow Knowledge Management Framework

[0056] The fast and slow knowledge management framework mainly realizes the rapid access of intelligent connected vehicles to the proposed high-safety scenario-driven decision-making and control integrated autonomous driving system supported by high-precision positioning based on the fast system under high-precision positioning; it adapts to newly emerging driving scenarios and changing driving environments based on the slow system, and continuously updates the scenario experience pool.

[0057] The fast system receives the observation sequence and high-precision positioning information uploaded by the intelligent connected vehicle through the cloud. The fast system predicts the scene factors of the driving scene in which the intelligent connected vehicle is located based on the scene causal reasoning model, and combines the decision-making control integrated neural network parameters corresponding to the high-precision positioning query scene. Through the scene-driven experience push based on federated learning, the decision-making control integrated neural network parameters are pushed to the corresponding intelligent connected vehicle. The experience push transmits the decision-making control integrated neural network parameters through federated learning combined with the ground-ground integrated communication system:

[0058]

[0059] where φ n represents the decision-making and control integrated neural network parameters of the nth intelligent connected vehicle, It represents the decision-making and control integrated neural network parameters that the cloud transmits to the nth intelligent connected vehicle based on the experience in the scene experience pool.

[0060] The slow system maintains and stores high-security emergency response strategies in a scenario experience pool based on the slow system's long-term data analysis and scenario understanding, and realizes real-time safety risk assessment and emergency strategy switching by monitoring the observation sequence uploaded by the vehicle-side platform through the slow system.

[0061] The safety risk assessment process mainly uses the intelligent chassis feedback information contained in the vehicle-side dynamic sequence. According to the degree of state change of the vehicle during driving, the three state quantities most obviously perceived by the human body, namely lateral acceleration, yaw angular acceleration and longitudinal acceleration, are used as measurement indicators, and the sensitive intervals of human perception are drawn out in turn. The weighted square sum of the three intervals is used as a quantitative indicator to establish a quantified index of intelligent chassis comfort, and the safety risk of vehicle-side autonomous driving is evaluated. When the quantified index of intelligent chassis comfort exceeds the human comfort threshold, an emergency strategy is switched, and the vehicle-side autonomous driving strategy is switched to a high-safety emergency response strategy maintained and stored in the scene experience pool.

[0062] The intelligent chassis comfort quantification index:

[0063]

[0064] in, represents the quantitative index of intelligent chassis comfort, a i (t) represents the lateral acceleration, yaw angular acceleration, and longitudinal acceleration of the intelligent chassis at time t, w i Represents the weighted parameters of the lateral acceleration, yaw acceleration, and longitudinal acceleration of the intelligent chassis, i=0 represents lateral, i=1 represents yaw, and i=2 represents longitudinal.

[0065] The emergency response strategy is mainly a verified high-safety emergency strategy that minimizes the quantitative index of intelligent chassis comfort among all scenario experiences in the scenario experience pool.

[0066] Based on vehicle-side sensor data and macro traffic information, the slow system continuously learns incrementally, trains new knowledge based on the original scene experience to adapt to emerging driving scenarios and changing driving environments, and uses screening and evaluation based on intelligent chassis feedback to store the newly generated scene experience in the scene experience pool. For experience updates generated by local changes in the scene, the slow system synchronizes communications under the integrated space-ground communication network, and performs real-time updates and maintenance of the existing scene experience in the scene experience pool based on high-precision positioning.

[0067] The screening and evaluation process based on intelligent chassis feedback is mainly to select the neural network that minimizes the intelligent chassis comfort quantitative index to participate in the neural network parameter aggregation, aggregate the new autonomous driving strategy neural network parameters and store them in the scene experience pool. The neural network parameter aggregation process:

[0068]

[0069] Among them, φ n ′ represents the neural network parameter φ of the autonomous driving strategy used by the nth intelligent connected vehicle n The neural network parameters φ of the autonomous driving strategy used by another intelligent connected car m , aggregate the newly generated autonomous driving strategy neural network parameters.

[0070] (3) Vehicle-side platform

[0071] The vehicle-side platform is composed of multiple vehicle-side groups, each of which is composed of several intelligent connected vehicles. The intelligent connected vehicles communicate with each other through V2V (vehicle-to-vehicle) communication technology; the vehicle-side platform is used to upload observation sequences and high-precision positioning information, quickly access the system, and continuously provide new scene knowledge through incremental training.

[0072] The above embodiments are only used to illustrate the design ideas and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, any equivalent changes or modifications made based on the principles and design ideas disclosed by the present invention are within the protection scope of the present invention.

Claims

1. A high-security scenario-driven decision-making and control integrated autonomous driving system supported by high-precision positioning, characterized by: Including satellite platform, cloud platform and vehicle-side platform, The vehicle-side platform uploads the observation sequence and positioning information to the cloud platform; The cloud platform communicates with the vehicle-side platform and the satellite platform respectively. The cloud platform is equipped with a data security monitoring framework, a scene causal reasoning framework, and a fast and slow knowledge management framework. The data security monitoring framework performs data security monitoring; The scenario causal reasoning framework is used to summarize and manage scenario knowledge in the cloud; the fast and slow knowledge management framework includes a fast system and a slow system; The satellite platform performs satellite communications with the cloud platform, complements the cellular network through satellite communications, and provides communication support.

2. The high-security scene-driven decision-making and control integrated automatic driving system supported by high-precision positioning according to claim 1 is characterized in that: The data security monitoring framework includes a large language model data analysis process, a data transmission behavior baseline setting process, and a data transmission activity qualitative analysis process.

3. The high-security scene-driven decision-making and control integrated automatic driving system supported by high-precision positioning according to claim 2 is characterized in that: The large language model data analysis process is to use the large language model to analyze the data content, extract key information to establish a data information library, and realize data classification and sensitivity rating.

4. The high-security scene-driven decision-making and control integrated automatic driving system supported by high-precision positioning according to claim 2 is characterized in that: The data transmission behavior baseline setting process performs anomaly detection on the behavior data and adjusts the baseline setting based on the anomaly detection result.

5. The high-security scene-driven decision-making and control integrated automatic driving system supported by high-precision positioning according to claim 2 is characterized in that: The qualitative analysis process of data transmission activities adopts a transformer-based causal reasoning model to perform qualitative analysis on time-series data transmission activities.

6. The high-precision positioning supported high-security scene-driven decision-making and control integrated automatic driving system according to claim 1, characterized in that: The scenario causal reasoning framework includes a cross-scenario training sequence collection process, a scenario causal reasoning model training process, and a scenario experience pool construction process, which summarizes and manages scenario knowledge in a cloud platform.

7. The high-precision positioning supported high-security scene-driven decision-making and control integrated automatic driving system according to claim 6 is characterized in that: The cross-scenario training sequence collection process uploads the cross-scenario training historical sequences of the intelligent connected vehicles in the vehicle-side group to the cloud platform.

8. The high-precision positioning supported high-security scene-driven decision-making and control integrated automatic driving system according to claim 6, characterized in that: The scene causal reasoning model training process trains the scene causal reasoning model based on the historical training process uploaded by the vehicle-side group.

9. The high-precision positioning supported high-security scene-driven decision-making and control integrated automatic driving system according to claim 6, characterized in that: In the process of constructing the scenario experience pool, the observation sequence and high-precision positioning information of the intelligent connected vehicles in the vehicle-side group are taken as input, and the process training scenario causal reasoning model is used to predict the scenario factors of the driving scenario in which the intelligent connected vehicles are located to construct the scenario experience pool.

10. The high-precision positioning supported high-security scene-driven decision-making and control integrated automatic driving system according to claim 1, characterized in that: The fast system pushes the decision-making control integrated neural network parameters corresponding to the predicted scene causality and high-precision positioning query scene to the intelligent connected vehicle through the scene-driven experience push based on federated learning; The slow system monitors the vehicle dynamic sequence for real-time safety risk assessment and emergency strategy switching.

Images