Vehicle auxiliary driving system

Abstract

The invention provides a vehicle auxiliary driving system which comprises a surrounding layer and a core layer, the surrounding layer comprises a function safety module and an information safety module, and the core layer comprises an SoC performance domain, an MCU real-time domain and a host end. The architecture design of the vehicle auxiliary driving system is carried out according to business layering, domain classification and the like, a software design based on layered decoupling and service dynamic management and a core design oriented to service and middleware management are provided based on an SoC performance domain, and the coupling between software and a hardware system is reduced; and in combination with the global surrounding design of function security and information security and task shunting at the host end, the coupling and communication load of the system are reduced on the premise that the system is safe and reliable, and the scene adaptability is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle intelligent control systems, and in particular relates to a vehicle assisted driving system. Background Art

[0002] Domestic vehicle assisted driving systems are generally distributed, with dedicated sensing integrated devices for forward vision, blind spot detection, surround view, and other functions. Each integrated device sends sensing target-level information to the MCU via traditional CAN communication. The MCU runs the control algorithm and controls the vehicle's execution-end devices (drive motor, steering auxiliary source, brake, etc.) through CAN communication. Researchers focus on how to use distributed control and traditional CAN bus technology to improve the stability and real-time performance of assisted driving. At the same time, they have explored multi-sensor fusion and functional safety for special working conditions of commercial vehicles. However, the following problems still exist: (1) fragmented computing power, high communication latency, poor functional scalability and versatility; (2) low electrical system integration, high overall cost, and low functional safety level; (3) high communication load, high information repetition rate, and information security issues.

[0003] Foreign vehicle assisted driving systems generally integrate key perception, fusion, decision-making, and planning functions into a single high-performance computing platform, with the software architecture utilizing the AP AUTOSAR platform. Foreign systems, however, place greater emphasis on domain centralization, deep learning, and hardware collaborative scheduling. These systems, based on the AP AUTOSAR (Adaptive Platform AUTOSAR) platform and paired with high-performance foreign system-on-chips (SoCs) and real-time control strategies, enable efficient response to complex assisted driving scenarios. However, the system's high complexity and extremely high overall cost have limited its practical application in the domestic commercial vehicle sector. Summary of the Invention

[0004] The purpose of the present invention is to address the shortcomings of the existing technology and provide a vehicle assisted driving system that reduces system coupling and communication load while improving scene adaptability under the premise of system safety and reliability.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A vehicle assisted driving system includes a surround layer and a core layer, wherein the surround layer includes a functional safety module and an information security module, and the core layer includes a SoC performance domain, an MCU real-time domain, and a host end; the surround layer and the core layer are in a parallel relationship;

[0007] The SoC performance domain is divided from bottom to top into a hardware base layer, a board support package layer, a middleware layer, an abstraction layer, a service layer, and a first application layer;

[0008] The hardware base layer comprises sensor hardware, and provides a data stream basis;

[0009] The board-level support package layer operates system development and hardware platform matching optimization;

[0010] The middleware layer unifies the communication framework of the SoC performance domain, and the communication framework is based on distributed real-time message middleware and communication serialization, and performs communication optimization and scene expansion.

[0011] The abstract layer writes independent plug-ins for each sensor type, defines a unified data structure for each sensor type, and realizes loading and unloading of the plug-in, sensor registration, error state monitoring, and sensor data distribution.

[0012] The service layer integrates the core services required by the first application layer, obtains the data stream required by the first application layer from the abstract layer through the middleware layer, performs data filtering and data integration processing, and provides the first application layer.

[0013] The MCU real-time domain is responsible for cyclic redundancy check and signal rationality detection of sensor signals based on double redundancy design, and the configuration direct memory access realizes data exchange of the MCU real-time domain in the SPI slave mode and the SoC performance domain.

[0014] The host end comprises an integrated development environment building module and a deep learning model development module.

[0015] The application designs the architecture of the vehicle auxiliary driving system according to business layering and domain classification, proposes a software design based on hierarchical decoupling and service dynamic management based on the SoC performance domain, and a core design facing services and middleware management, thereby reducing the coupling of the software and the hardware system; one service of the application can be reused by multiple upper layer functions in real time, thereby reducing the communication and computing power load required for function implementation; in combination with the global surrounding design of functional safety and information safety and the task shunting of the host end, the application reduces the system coupling and the communication load under the premise of system safety and reliability, and improves the scene adaptability.

[0016] Further, the sensor hardware comprises at least one of a camera, a radar, a laser, an ultrasonic wave, and a positioning module.

[0017] Further, the functions of the functional safety module include hardware redundancy, dual-core lockstep, degradation protection, and monitoring and diagnosis; and the functions of the information safety module include data security, communication security, identity authentication, and intrusion detection.

[0018] The present invention integrates multiple functional modules based on the functional safety of the surround layer, facilitates the implementation of a unified security strategy, supports global status monitoring and rapid fault location, and can perform global security response and dynamic degradation protection in a timely manner; information security based on the surround layer can efficiently achieve data flow security and intrusion protection through centralized security encryption and unified access control; and reduces system coupling and communication load.

[0019] Furthermore, the abstract layer includes a sensor hardware abstraction module, a sensor data abstraction module, and an abstract management module. The sensor hardware abstraction module writes an independent plug-in for each sensor type, and the sensor data abstraction module defines a unified data structure for each sensor type; the abstract management module implements the loading and unloading of plug-ins, sensor registration, error status monitoring, and distribution of sensor data.

[0020] Furthermore, the first application layer includes a heterogeneous sensor perception module, a perception fusion framework module, a planning and decision module, and a communication and data module.

[0021] The present invention integrates the architecture based on rule-based algorithms and data-driven technology to improve the adaptability of scenarios.

[0022] Furthermore, the MCU real-time domain is divided from bottom to top into a microcontroller abstraction layer, a basic software layer, an operation service layer, and a second application layer; the functions of the second application layer include assisted driving, network communication, body control, chassis control, fault handling, and power management.

[0023] Furthermore, the functions of the integrated development environment building module include cross-compilation environment construction, multimodal sensor calibration, visual debugging and performance analysis;

[0024] The functions of the deep learning model development module include training dataset management, deep learning model data annotation, deep learning model development and training, deep learning model verification and reasoning, and deep learning model quantization and deployment.

[0025] Furthermore, the vehicle assisted driving system includes a central domain controller and multiple sub-domain systems. The central domain controller receives data and status of the sub-domain systems through the communication module and controls the sub-domain systems.

[0026] The central domain controller includes a SoC processor and an MCU microcontroller, and the SoC processor and the MCU microcontroller communicate through two serial interfaces;

[0027] SoC processors are used for multi-sensor perception processing, heterogeneous perception information fusion, trajectory prediction, environmental model building, communication and data processing;

[0028] MCU microcontroller is used for controlling, managing and monitoring chassis and vehicle body.

[0029] The SoC of the application cooperates with the MCU, realizes high-performance computing and high real-time control under the premise of controlling cost, improves system computing power, sensing performance and communication efficiency, etc.

[0030] The application divides multiple sub-domain systems according to functions, and each domain can be controlled and executed according to components and system position distribution, and components and systems in each domain can be flexibly tailored, added and upgraded according to characteristics such as vehicle type and individual function; through central domain integrated management and in-domain nearest principle, wiring harness and computing power cost are reduced, and system coupling and communication load are reduced.

[0031] Further, the sub-domain system includes a perception positioning domain system, a chassis domain system, a vehicle body domain system and a cabin domain system.

[0032] Further, the perception positioning domain system includes at least one of a camera, a radar, a laser, an ultrasonic wave and a positioning module.

[0033] The chassis domain system includes at least one of a driving system, an energy storage system, a braking system, a steering system and a suspension system.

[0034] The vehicle body domain system includes a vehicle body domain gateway and a vehicle body control module.

[0035] Compared with the prior art, the application has the following advantages:

[0036] The application designs the architecture of the vehicle assisted driving system according to business layering and domain classification, proposes a software design based on hierarchical decoupling and service dynamic management based on the SoC performance domain, and a core design oriented to service and middleware management, thereby reducing the coupling of software and hardware systems; one service of the application can be reused by multiple upper layer functions in real time, thereby reducing the communication and computing power load required for function implementation; the application further combines the global surrounding design of functional safety and information safety, and the host task shunting, thereby reducing the system coupling and communication load under the premise of system safety and reliability, and improving the scene adaptability.

[0037] The SoC of the application cooperates with the MCU, realizes high-performance computing and high real-time control under the premise of controlling cost, improves system computing power, sensing performance and communication efficiency, etc.

[0038] The application divides multiple sub-domain systems according to functions, and each domain can be controlled and executed according to components and system position distribution, and components and systems in each domain can be flexibly tailored, added and upgraded according to characteristics such as vehicle type and individual function; through central domain integrated management and in-domain nearest principle, wiring harness and computing power cost are reduced, and system coupling and communication load are reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a schematic diagram of the overall architecture of the central domain control platform according to an embodiment of the present invention;

[0040] Figure 2 A schematic diagram of a central domain controller hardware platform according to an embodiment of the present invention;

[0041] Figure 3 Schematic diagram of the software architecture of a vehicle assisted driving system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The present invention will be described in detail below with reference to the embodiments. It should be noted that the embodiments and features of the embodiments of the present invention can be combined with each other without conflict.

[0043] Example

[0044] Considering the commercial vehicle field's demand for the versatility and scalability of the overall control platform, while strengthening the integrated management between components and reducing the cost of wiring harnesses and computing power waste, this embodiment provides a central domain control platform. The overall architecture of the central domain control platform is as follows: Figure 1 shown.

[0045] The central domain control platform includes five domains: central domain, perception and positioning domain, chassis domain, body domain, and cockpit domain:

[0046] The central domain controller is the core of the platform's algorithm control and integrated management. It receives data and status collected from the four domains via CAN / vehicle Ethernet / dedicated interfaces and performs coordinated control of the four domains.

[0047] The perception and positioning domain includes multimodal perception sensors such as cameras, radars, lasers, and ultrasound, positioning modules such as inertial navigation / GPS, and components required for perception and positioning of commercial vehicle assisted driving / autonomous driving systems;

[0048] The chassis domain includes core execution systems related to vehicle control, such as drive, energy storage, braking, steering, and suspension;

[0049] The body domain includes the body domain gateway and its body control modules. The body domain gateway is responsible for integrating and feeding back information module message data and distributing central domain commands. Each body control module collects signals from surrounding electrical components and executes output based on proximity.

[0050] The cockpit domain is designed with a human-computer interaction domain controller in mind due to its diverse data types, large data transmission volume, and lower data real-time requirements (compared with other domain data). While meeting the real-time requirements of human-computer interaction, it achieves collaborative work among various components in the cockpit domain through standardized data interface division and big data storage, and can independently interact with cloud platforms through 5G networks.

[0051] As described above, the central domain control platform architecture combines functional domains with location domains, dividing the system into five functional domains. Within each domain, control and execution are performed based on the location of components or systems. This platform combines the characteristics of both functional and location domains, offering the following advantages:

[0052] The five domains under this platform architecture and the components / systems within each domain can be flexibly tailored, added and upgraded according to the characteristics of commercial vehicle models, personalized functions and other requirements; taking commercial vehicle buses as an example, the number of sensors, hardware driver interfaces and actuators required for a 6-meter bus is less than that of a 10-meter bus. This platform architecture can reduce the cost of the entire vehicle by reducing the number of sensors, body modules and wiring harnesses while meeting the functions; the same model has different high and low configurations, and most intelligent functions and some control functions are optional. This platform can achieve the purpose of flexible selection by adding or reducing sensors, actuators and even entire functional domains within the domain.

[0053] The platform architecture greatly reduces wiring and computing power costs through central domain integrated management and the principle of proximity within the domain.

[0054] This embodiment also provides a central domain controller hardware system, such as Figure 2 shown.

[0055] The central domain controller can be awakened through Ethernet, CAN and the first gear power of the vehicle when connected to a 9-32V power supply, and can interact with the vehicle's cockpit domain, chassis domain, body domain, etc. for data signals based on the central domain platform architecture.

[0056] The core chip modules of the central domain controller are divided into the performance domain (AI-SoC) and the real-time domain (Safety-MCU). The SoC processor features high clock speed and strong computing power. It allocates hardware resources through system kernel management and is primarily used for high-computing tasks in the performance domain, such as multi-sensor perception processing, heterogeneous perception information fusion, trajectory prediction, environmental model building, and high-speed vehicle Ethernet communication. The MCU microcontroller, with its high stability and strong real-time performance, primarily controls, manages, and monitors the chassis and body domains of the vehicle, which require high reliability and safety.

[0057] Theoretically, for some application scenarios, a single SoC master control chip can complete MCU chip-related tasks through a built-in safety island or lock core. However, compared with traditional plug-in MCUs using mature processes, the former's built-in MCU chip has limited memory, which affects the planning of control algorithm models and the deployment of MCU system software. Moreover, a single SoC solution currently cannot meet the automotive-grade functional safety and information security requirements of the vehicle end. Therefore, the SoC and MCU design solution of this embodiment adopts intra-board inter-chip communication and functional decoupling.

[0058] In terms of intra-board data link design, this central domain controller selects two serial peripheral interfaces (SPI) as the communication bridge between the SoC and MCU based on the participants, indicator parameters, security level and other requirements of internal communication. The two SPIs are mutually master-slave, which can fully demonstrate the SPI's low-cost, high real-time and strong flexibility in data transmission, while also providing communication redundancy and avoiding the risk of controller failure caused by the SPI master-slave limitation.

[0059] In terms of off-board data link design, this central domain controller quantifies and classifies various types of signal interfaces and communication interfaces based on technical indicators such as communication rate, communication data volume, real-time performance, and stability, and arranges them on the SoC and MCU chips respectively based on this basis (for example, in-vehicle Ethernet is centrally deployed on the SoC end due to its high data transmission efficiency and fast transmission rate; CAN is centrally deployed on the MCU end due to its stable data transmission and high degree of standardization, making it suitable for closed-loop data flow between the system and the execution domain).

[0060] This embodiment also provides a software architecture for an assisted driving system. Figure 3 As shown in the figure, the assisted driving software architecture is divided into a surround layer and a core layer at the business level. The surround layer horizontally covers the entire core layer.

[0061] The surround layer is an integrated system that separates the two business modules of functional safety and information security that will not affect the execution of core functions from the software architecture. The surround layer covers the entire core layer horizontally.

[0062] Functional safety includes hardware redundancy, dual-core lockstep, degradation protection, monitoring and diagnosis, and other aspects. Due to the limitations of the distributed domain controller system software framework, each independent controller needs to be configured with redundant hardware (such as dual MCUs), which increases system cost and power consumption. The functional safety of high-level active safety applications at level L2 (semi-autonomous driving or partially automated driving) and above is difficult to reach ASIL-D level; however, this embodiment integrates multiple functional modules based on the functional safety of the surround layer, which facilitates the implementation of a unified safety strategy, supports global status monitoring and rapid fault location, and can perform global safety response and dynamic degradation protection in a timely manner.

[0063] ASIL (Automotive Safety Integrity Level) is defined in the ISO 26262 standard and is used to assess and categorize the safety requirements of automotive electrical and electronic systems. There are four ASIL levels: A, B, C, and D. D is the highest level, indicating that a system or component failure could cause a serious accident and pose a significant threat to human safety.

[0064] Information security encompasses data security, communication security, identity authentication, intrusion detection, and other aspects. In a distributed architecture, multiple independent controllers must communicate via a bus (such as CAN), making cross-node security coordination difficult. Furthermore, each node independently implements security measures (such as different encryption protocols), resulting in inconsistent overall protection and vulnerability to man-in-the-middle attacks (such as counterfeiting vehicle electronic control unit (ECU) instructions). Therefore, data flow information security based on a distributed architecture has always been a pain point in the industry. Information security based on the surround layer, on the other hand, can efficiently achieve data flow security and intrusion protection through centralized security encryption and unified access control solutions.

[0065] The core layer refers to the functional system that implements core business logic such as perception, regulation, and algorithm execution. It is divided into two major functional domain modules based on real-time performance, system complexity, and computing resource requirements: the MCU-based hardware and software system is regarded as the real-time domain; the SoC and the hardware, software, and operating system running on it are collectively referred to as the performance domain.

[0066] The core layer is the system's "brains" and "muscles," while the surround layer serves as the "immune system" and "protective net." The core layer's functional execution relies on the secure environment provided by the surround layer. The surround layer continuously monitors the core layer's status, inputs, outputs, and resource consumption. If it detects abnormal core behavior, resource depletion, or security risks, it has the highest priority to intervene and can request the core layer to downgrade, suspend, or shut down specific functions.

[0067] The core layer and the surrounding layer have a symbiotic and cooperative relationship of "monitoring-execution-guarantee".

[0068] 1. SoC performance domain architecture design

[0069] As a software platform architecture standard for high-performance computing, AP AUTOSAR architecture has been used in combination with SoC in the past. However, due to technical limitations such as its complex development system, highly dependent development tool chain, and conflicting real-time requirements, its actual implementation cost is extremely high.

[0070] To address these issues, this embodiment retains the AP AUTOSAR service discovery mechanism and utilizes a self-designed layered software architecture, a Portable Operating System Interface (POSIX) operating system, and a core design style of service-oriented + middleware management. Its key features are as follows:

[0071] Service and abstract interface: The service implementation of the function and the user interface are separated. Service implementation refers to the processing, calculation, storage, etc. of internal data. The user interface refers to the interface for the lower layer to provide data and the upper layer to obtain results. Separating the two allows the service to iterate independently without affecting the call of the user interface, truly decoupling the service and the interface.

[0072] Dynamic service loading: Services are integrated into environment variables in the form of dynamic libraries, supporting distributed updates and iterations of the software architecture. Application layer functions can be implemented by dynamically configuring and loading different services into random access memory (RAM).

[0073] Middleware-based message publishing and subscription: The communication mechanism between functional components adopts message-based publishing and subscription. After sending a message, the message publisher no longer owns the memory ownership of the message data, and interested receivers receive and process it asynchronously.

[0074] Based on the above core design style, this embodiment divides the SoC performance domain into multiple dimensional layers from bottom to top, namely: hardware foundation layer, board support package layer (BSP), middleware layer, abstraction layer, service layer, and application layer (first application layer):

[0075] The hardware foundation layer is the data flow foundation of the software architecture, which meets the requirements of commercial vehicle assisted driving functions, performance redundancy, scalability and other indicators; the hardware foundation layer includes sensor hardware, and the sensor hardware includes at least one of the camera, radar, laser, ultrasonic, and positioning modules.

[0076] The board support package (BSP) layer sits between the hardware foundation layer and the middleware layer. Its core purpose is to integrate a common operating system kernel (Linux / QNX) with the board support package (BSP), optimizing operating system development for hardware platform compatibility. Both the operating system and the BSP support secondary development and iterative optimization are supported. The BSP layer includes a multitasking scheduling module, a high-performance foundational library, and a system boot loader.

[0077] The middleware layer embodies the SoC's distributed design philosophy and unifies the communication framework of the SoC software architecture. This communication framework is based on distributed real-time messaging middleware and communication serialization, enabling communication optimization and scenario expansion. Functions implemented in the middleware layer are implemented as functional components, including data serialization, service / request components, environment monitoring components, log management components, thread / coroutine management components, module management components, and data caching components.

[0078] Communication framework standardization refers to the unified planning, design and standardization of multiple different communication frameworks to achieve consistency and compatibility of communication frameworks in terms of functions, interfaces, protocols, etc., thereby improving the maintainability, scalability and interoperability of the system.

[0079] The levels of vehicle-assisted driving and functional requirements in different commercial vehicle projects vary greatly. This difference significantly increases the software architecture viscosity and development difficulty of the upper-level application and the underlying hardware interface. The design of the abstraction layer can effectively solve this problem. The abstraction layer consists of three parts: the first is the sensor hardware abstraction module (Sensor HAL), which writes independent plug-ins for each specific sensor model that needs to be supported. The plug-in implements the specific hardware abstraction interface defined by that type of sensor, achieving the purpose of forming a unified hardware abstraction interface for different sensors, different manufacturers, different interfaces, and different communication protocols; the second is the sensor data abstraction module (Sensor Abstraction), which defines a unified data structure for each sensor type, allowing upper-level applications to be easily decoupled from the hardware base layer; the third part is the abstraction management module (Sensor Manager) based on the above two parts, which implements the loading and unloading of plug-ins, sensor registration, error status monitoring, and distribution of sensor data.

[0080] As a service-oriented core layer, the service layer integrates all core services required by the application layer. Under the management of service configuration, it can obtain the data streams required by the application layer from the abstract layer through the middleware layer. It then performs front-end data processing such as data screening and integration before providing it to the upper layer. The service layer includes forward-view perception service modules, CAN communication service modules, SPI communication service modules, surround view perception service modules, in-cabin perception service modules, and radar perception service modules.

[0081] The application layer (first application layer) belongs to the uppermost layer of the SoC performance domain software framework, and classifies intelligent functions such as auxiliary driving perception, planning, etc. Considering the current commercial vehicle technical capability, overall cost, training data set quality, auxiliary driving function level application scene and regulatory index, the overall software architecture based on rule algorithm driving + multi-level deep learning software framework based on data driven application layer composite software architecture form is finally selected to improve the algorithm potential, software adaptability and iterability while ensuring the stability, safety and explainability of the overall software architecture of the SoC performance domain.

[0082] The application layer (first application layer) includes heterogeneous sensor perception modules, perception fusion framework modules, planning decision modules, communication and data modules. The functions of the heterogeneous sensor perception module include forward target recognition, blind area monitoring, in-cabin target recognition, target tracking, etc. The functions of the perception fusion framework module include fusion positioning, time synchronization, global target recognition, etc. The functions of the planning decision module include logic / environment model, trajectory prediction, global planning, decision management, etc. The functions of the communication and data module include data caching, inter-chip communication, cross-domain communication, data processing, etc. The data flow between the modules in the application layer is: heterogeneous sensor perception -> perception fusion -> planning decision -> communication and data.

[0083] 2. MCU real-time domain software architecture design

[0084] The MCU real-time domain software architecture design is based on the Classic AUTOSAR architecture of ISO 26262 functional safety ASIL-D level, and the following designs are made:

[0085] The MCU real-time domain is divided into a microcontroller abstraction layer, a basic software layer, a running service layer, and a control application layer (second application layer) from bottom to top. The functions of the control application layer (second application layer) include auxiliary driving, network communication, power management, vehicle body control, fault handling, chassis control, etc.

[0086] The key sensor signals are checked by cyclic redundancy check (CRC) and signal rationality detection based on double redundancy design to ensure data integrity.

[0087] Direct memory access (DMA) is configured to realize high-speed data exchange between the MCU real-time domain in SPI slave mode and the SoC performance domain.

[0088] 3. Host software structure design

[0089] The focus of resources, computing power and acceleration optimization of the Soc performance domain and MCU real-time domain are placed on the realization of their respective core functions. The host side is designed separately as an external interaction window between developers and embedded platforms to realize the two major functions of integrated development environment construction and deep learning model development.

[0090] The integrated development environment includes the following sub-functions:

[0091] Cross-compilation environment establishment: Package the required applications, environment, and dependencies in the SoC performance domain into a portable SDK image. Create a Docker container based on the image on the host side, and then use the GNU compilation suite, Cmake compilation tool, Python, Protobuf, and other compilation tool chains and Secure Shell (SSH) services to implement the SoC performance domain-host cross-compilation environment establishment.

[0092] Multimodal sensor calibration: The host calibrates vision and radar sensors using a visual host computer and transmits the eXtensible Markup Language (XML) calibration configuration file to the SoC performance domain / MCU real-time domain via SSH.

[0093] Visual debugging and performance analysis: Develop dedicated debugging tools to synchronously receive, visualize, and compare sensor raw data, vehicle dynamics data, CPU / GPU / NPU occupancy rates, and other data, and integrate the Valgrind dynamic analysis tool for software memory and thread management.

[0094] Deep learning model development includes the following sub-functions:

[0095] Training data set management: The host side synchronously reuses the collected debugging data and training data sets, sets data collection trigger conditions based on edge scenarios, and eliminates invalid data through automated scripts while ensuring that the data covers typical assisted driving scenarios.

[0096] Model data labeling: The host side builds a multi-level labeling system including the basic labeling layer and the scene semantic layer to meet the requirements of commercial vehicle assisted driving functions, and uses a semi-automatic labeling tool (Label Studio) to improve labeling efficiency.

[0097] Model development and training: To balance model accuracy, resource computing power, and the real-time requirements of embedded deployment, the host side uses channel pruning, compressed model parameters, and multi-task joint training to implement a "lightweight network + full coverage dataset" model development solution.

[0098] Model verification and reasoning: To address the poor interpretability of deep learning models, the host side built a modular testing framework based on scenario datasets to conduct real-vehicle scenario reasoning tests on application functions such as automatic emergency braking system AEB, lane keeping system LKA, adaptive cruise control ACC, and blind spot assist system BSM, and output reasoning performance reports.

[0099] Model quantization and deployment: The host side performs hardware adaptation for the model output dimensions, activation functions, quantization parameters, and customized operators supported by the neural network processing unit (NPU) in the SoC performance domain to reduce precision loss. Finally, quantization optimization and embedded deployment are performed through the model quantization deployment tool chain.

[0100] This method has the following advantages:

[0101] (1) Platform architecture innovation: A new central domain platform architecture is proposed, which takes into account the requirements of the commercial vehicle field for the versatility and scalability of the overall control platform, while strengthening the integrated management between various components and reducing the waste costs of wiring harnesses and computing power.

[0102] (2) Hardware innovation: We have made innovations in the design of core chip modules and data links inside and outside the board. We have proposed a dual-core collaborative architecture of SoC and MCU, a design scheme of inter-chip communication within the board + functional decoupling, and a data link design idea of ​​communication redundancy + quantitative classification. This allows us to balance high-performance computing and high real-time control while controlling platform costs. Compared with traditional single-core distributed solutions of the same cost, the comprehensive improvement in system computing power, perception performance and communication efficiency is more than 5%.

[0103] (3) Software architecture innovation: Based on the SoC performance domain, a software design paradigm based on layered decoupling + service dynamic management and a core design style oriented to service + middleware management are proposed. Combined with the global surround design of functional safety and information security and host-side task diversion, the system coupling and communication load are significantly reduced while the system is safe and reliable.

[0104] (4) Innovation in technical paths: The system integrates the architecture of rule-based algorithms and data-driven technical paths, improving the ability to adapt to scenarios while ensuring safety, and providing new ideas for commercial vehicle assisted driving technical paths.

[0105] The assisted driving software and hardware systems based on the commercial vehicle central domain control system provided in this embodiment solve the following problems:

[0106] (1) Multiple independent controllers / hosts in a distributed architecture have high costs, low functional safety levels, and their computing resources cannot be coordinated and scheduled, resulting in waste of computing power.

[0107] (2) Distributed assisted driving systems have problems such as large communication delay, high communication load, high information repetition rate, and inability to ensure information security;

[0108] (3) The distributed assisted driving system architecture is discrete, and the software and hardware scalability and versatility are poor;

[0109] (4) The system-level integration of the two technical routes of rule-based algorithms and data-driven in the field of commercial vehicle assisted driving.

[0110] The contents illustrated in the above embodiments should be understood as these embodiments are only used to more clearly illustrate the present invention, and are not used to limit the scope of the present invention. After reading the present invention, various equivalent modifications of the present invention by those skilled in the art shall fall within the scope defined by the claims attached to this application.

Claims

1. A vehicle assisted driving system, characterized in that: It includes a surrounding layer and a core layer, wherein the surrounding layer includes a functional safety module and an information security module, and the core layer includes a SoC performance domain, an MCU real-time domain, and a host end; the surrounding layer and the core layer are in a parallel relationship; The SoC performance domain is divided from bottom to top into a hardware base layer, a board support package layer, a middleware layer, an abstraction layer, a service layer, and a first application layer; The hardware foundation layer includes sensor hardware and provides the data flow foundation; The board-level support package layer performs operating system development and hardware platform matching optimization; The middleware layer unifies the communication framework of the SoC performance domain. The communication framework is based on distributed real-time messaging middleware and communication serialization to optimize communication and expand scenarios. The abstract layer writes independent plug-ins for each sensor type, defines a unified data structure for each sensor type, and implements plug-in loading and unloading, sensor registration, error status monitoring, and sensor data distribution; The service layer integrates the core services required by the first application layer, obtains the data stream required by the first application layer from the abstract layer through the middleware layer, performs data screening and data integration processing, and then provides it to the first application layer; The MCU real-time domain is responsible for performing cyclic redundancy check and signal rationality detection on sensor signals based on dual redundant design, and configuring direct memory access to realize data exchange between the MCU real-time domain and the SoC performance domain in SPI slave mode; The host side includes an integrated development environment building module and a deep learning model development module.

2. The vehicle assisted driving system according to claim 1, characterized in that: The sensor hardware includes at least one of a camera, radar, laser, ultrasonic wave, and positioning module.

3. The vehicle assisted driving software system according to claim 1, characterized in that: The functions of the functional safety module include hardware redundancy, dual-core lockstep, degradation protection, and monitoring and diagnosis; the functions of the information security module include data security, communication security, identity authentication, and intrusion detection.

4. The vehicle assisted driving system according to claim 1, characterized in that: The abstract layer includes a sensor hardware abstraction module, a sensor data abstraction module, and an abstract management module. The sensor hardware abstraction module writes an independent plug-in for each sensor type, and the sensor data abstraction module defines a unified data structure for each sensor type; the abstract management module implements the loading and unloading of plug-ins, sensor registration, error status monitoring, and distribution of sensor data.

5. The vehicle assisted driving system according to claim 1, characterized in that: The first application layer includes a heterogeneous sensor perception module, a perception fusion framework module, a planning and decision module, and a communication and data module.

6. The vehicle assisted driving system according to claim 1, characterized in that: The MCU real-time domain is divided from bottom to top into a microcontroller abstraction layer, a basic software layer, an operation service layer, and a second application layer; the functions of the second application layer include assisted driving, network communication, body control, chassis control, fault handling, and power management.

7. The vehicle assisted driving system according to claim 1, characterized in that: The functions of the integrated development environment building module include cross-compilation environment construction, multi-modal sensor calibration, visual debugging and performance analysis; The functions of the deep learning model development module include training dataset management, deep learning model data annotation, deep learning model development and training, deep learning model verification and reasoning, and deep learning model quantization and deployment.

8. The vehicle assisted driving system according to claim 1, characterized in that: It includes a central domain controller and multiple sub-domain systems. The central domain controller receives data and status of the sub-domain systems through the communication module and controls the sub-domain systems. The central domain controller includes a SoC processor and an MCU microcontroller, and the SoC processor and the MCU microcontroller communicate through two serial interfaces; SoC processors are used for multi-sensor perception processing, heterogeneous perception information fusion, trajectory prediction, environmental model building, communication and data processing; MCU microcontrollers are used to control, manage and monitor the chassis and body.

9. The vehicle assisted driving system according to claim 8, characterized in that: The subdomain systems include a perception and positioning domain system, a chassis domain system, a body domain system, and a cockpit domain system.

10. The vehicle assisted driving system according to claim 8, characterized in that: The perception and positioning domain system includes at least one of a camera, a radar, a laser, an ultrasonic wave, and a positioning module; The chassis domain system includes at least one of a drive system, an energy storage system, a braking system, a steering system, and a suspension system; The body domain system includes a body domain gateway and a body control module.

Images