Intelligent driving real-time task scheduling method based on multi-source information aggregation

Abstract

The invention relates to an intelligent driving real-time task scheduling method based on multi-source information aggregation, and belongs to the technical field of intelligent driving. Aggregating and verifying data of the message center; the task scheduler detects conditions; state transition of the finite-state machine; and the message publisher issues a control instruction. The performance and reliability of the automatic driving system can be improved, and the application requirement of the automatic driving system in a complex dynamic environment is met.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent driving technology, and in particular relates to an intelligent driving real-time task scheduling method based on multi-source information aggregation. Background Art

[0002] In the field of intelligent driving, the task scheduling system serves as the core hub connecting perception, planning, prediction, decision-making, and control. Its performance directly affects the safety and reliability of the autonomous driving system. Currently, mainstream task scheduling solutions mainly adopt the following two architectures:

[0003] 1) Centralized scheduling architecture

[0004] The currently open-source Executor model in ROS2 processes all tasks through a single event loop. A typical example is rclcpp::SingleThreadedExecutor. However, its drawbacks include fixed task priorities, an inability to adapt to dynamic scenarios, and a lack of ability to collaboratively process data from multiple algorithm modules.

[0005] 2) Distributed Scheduling Architecture

[0006] The multi-ECU collaboration solution in AUTOSAR uses the Tasking System for task allocation. However, it suffers from problems such as high cross-node communication overhead (data copying accounts for >30% of the time), insufficient sensor data time synchronization accuracy (typical error >50ms), and task blocking caused by resource competition (such as CAN bus conflicts).

[0007] The prior art discloses a real-time task scheduling method for cloud-edge collaboration of intelligent connected vehicles, whose technical features include: collecting real-time status data, task requirements and network conditions of the vehicle, performing multi-source data fusion processing, and generating comprehensive data; based on the comprehensive data, adjusting the task priority using a dynamic task priority optimization mechanism, and selecting the optimal path for task transmission using a shortest path algorithm; allocating tasks through collaboration between edge nodes to achieve dynamic migration and load balancing of tasks, identifying the scene type around the vehicle based on semantic analysis technology, and dynamically adjusting the task scheduling strategy, etc.

[0008] The above methods are based on reasoning of deep learning models, which may lead to insufficient timeliness of data fusion and increase the delay of task scheduling. Reinforcement learning algorithms may converge slowly or fail in dynamic environments. The high dynamics of vehicle status and network conditions lead to state space explosion, and traditional reinforcement learning is difficult to adapt quickly. Dynamic task migration may lead to frequent context switching, which reduces efficiency. It relies on load thresholds to trigger migration, but the threshold settings are fixed (such as preset thresholds) and cannot adapt to the performance differences of heterogeneous edge nodes. Global status monitoring requires continuous synchronization of vehicle, network, and edge node data, which occupies bandwidth. Dynamic policy adjustments (such as semantic analysis triggering rescheduling) may introduce additional delays. Therefore, the communication overhead of cloud-edge collaboration may offset the benefits of scheduling optimization. Summary of the Invention

[0009] The present invention aims to address the many defects and shortcomings in existing intelligent driving task scheduling systems and provide a new intelligent driving real-time task scheduling method based on multi-source information aggregation, so as to significantly improve the performance and reliability of the autonomous driving system and meet its application requirements in complex dynamic environments.

[0010] In order to achieve the above objectives, the technical solution of the present invention provides a method for real-time task scheduling of intelligent driving based on multi-source information aggregation, comprising the following steps:

[0011] Multi-source data input;

[0012] Message center data aggregation and verification;

[0013] Task scheduler detection conditions;

[0014] Finite state machine state transition;

[0015] The message publisher issues control instructions.

[0016] Preferably, the multi-source data includes vehicle sensor data, intelligent driving cabin data, MCU data, planning data and perception data.

[0017] Preferably, message center data aggregation and verification are thread-safe through shared memory and read-write lock mechanisms.

[0018] Preferably, the task scheduler detection conditions include driver behavior conditions, vehicle status conditions, environment perception conditions and system health conditions.

[0019] Preferably, the finite state machine state transition includes transitioning from an ACC standby state to an LCC active state.

[0020] Preferably, the message publisher issuing the control instruction includes sending the task instruction to the executor through the data distribution service.

[0021] Preferably, it also includes configuration management, which supports dynamic adjustment of parameters by loading thresholds from configuration files through a configuration manager.

[0022] Preferably, an exception handling mechanism is also included, which forces switching to a safe state and resetting related states when an abnormal condition is detected.

[0023] Preferably, the task scheduling method meets the ASIL-D real-time requirements, and the task scheduling period is ≤10ms.

[0024] Preferably, the message de-jitter technology is used to reduce the HMI instruction false trigger rate and improve system reliability.

[0025] Compared with existing technologies, the intelligent driving real-time task scheduling method based on multi-source information aggregation has the following advantages:

[0026] 1. Improve decision-making accuracy through efficient aggregation of multi-source information

[0027] By integrating vehicle sensor data (such as chassis information and HMI interaction), planning control feedback (such as ACC / LCC / NOA status) and environmental perception information (such as parking space detection and map data), and using a message center for unified management, combined with debounced buttons and state machine (FSM) technology, the probability of false triggering is reduced, the accuracy of task scheduling decisions is increased by more than 30%, and the risk of incorrect operation is effectively reduced.

[0028] 2. Real-time task scheduling optimization to reduce system latency

[0029] Multi-threaded task management (TaskThreads) is used to decouple task scheduling (such as ACC activation and parking mode switching) from message publishing and subscription (EkaSubscriber / TaskCommandPublisher), and periodic status detection is implemented through timers. The average task response time is within 20ms, meeting the high real-time requirements of autonomous driving.

[0030] 3. Dynamic priority adjustment to enhance system robustness

[0031] Based on the finite state machine (FSM) and conditional trigger mechanism (TagConditions), task priorities are dynamically adjusted (for example, emergency braking takes precedence over cruise control), ensuring that critical tasks (such as abnormal parking exit) can be preempted for execution. This improves the system's fault tolerance by 40% and reduces the risk of failure due to task conflicts.

[0032] 4. Configuration management to improve scalability

[0033] Through unified configuration management (Config), dynamic parameter adjustment (such as brake pedal threshold and following distance level) is supported, which can adapt to different vehicle models or driving scenarios without recompilation, and the efficiency of development and actual vehicle testing is improved by more than 50%.

[0034] 5. Optimize resource usage and reduce computing overhead

[0035] It uses a lightweight message protocol (ProtoBuf), shared locks (shared_mutex), and smart pointers (shared_ptr, unique_ptr) to reduce data copying and lock contention, lowering CPU usage by 20% and memory consumption by 15%, making it suitable for embedded vehicle computing platforms.

[0036] 6. Scientific Verification

[0037] In real-vehicle testing, this method achieved a false trigger rate of less than 0.1% across more than 1,000 task switching cycles, outperforming traditional event-driven solutions (5% to 10%). In urban road and parking scenarios, the task scheduling success rate was ≥99.9%, with an average system latency of <10ms, meeting ISO 26262 ASIL-B functional safety requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a dynamic flow diagram from data input to task execution in an intelligent driving real-time task scheduling method based on multi-source information aggregation in the present invention. DETAILED DESCRIPTION

[0039] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0040] This embodiment of the present invention discloses a real-time task scheduling method for intelligent driving based on multi-source information aggregation. It is applicable to autonomous driving systems, such as ACC / LCC / NOA, automatic parking, and fingertip parking. It adopts a layered architecture design, primarily comprising an information receiving layer, a data processing layer, and a decision-making and scheduling layer. Through multi-source data aggregation, dynamic state machine scheduling, and priority management, it enables efficient and secure task decision-making.

[0041] The information receiving layer is primarily responsible for aggregating data from the vehicle's intelligent cockpit, MCU, planning, and perception, enabling real-time reception of multi-source information. Multi-source data, including vehicle sensors (CAN bus), HMI interaction (Ethernet), environmental perception (DDS), and planning and control feedback (DDS), is centrally managed through the Message Center. This data is stored in the Message Center via shared memory and a read-write lock (shared_mutex) to ensure thread safety. For example, when the HMI sends an ACC activation command, the Message Center updates the acc_active_button state and triggers the task scheduler's detection conditions (such as the T1 condition).

[0042] The data processing layer consists of two parts: data aggregation and feature extraction. The data aggregation module achieves spatiotemporal alignment of multi-source information, while the feature extraction module is responsible for extracting key driving scenario features and establishing a mapping relationship between T1-T53 trigger conditions and state transitions. For example, trigger conditions include driver behavior conditions (such as steering wheel torque and pedal position), vehicle status conditions (such as speed and gear position), environmental perception conditions (such as ODD satisfaction), and system health status (fault codes of various modules).

[0043] The decision-making scheduling layer, centered around a finite state machine (FSM) task scheduling engine, implements intelligent driving mode switching. The task scheduler (ScheduleTask) reads data from the MessageCenter and outputs task instructions based on detected conditions, such as activating ACC or switching to parking mode. The state machine engine (FSM Stack) receives these instructions and drives state transitions, such as switching from OFF to ACC standby to LCC activation.

[0044] In the specific implementation, multi-source information aggregation is first performed. All data is stored in the MessageCenter via shared memory and read-write locks to ensure thread safety. The message subscriber (EkaSubscriber) acquires external data, such as chassis information and HMI commands, in real time and writes it to the MessageCenter. The message publisher (TaskCommandPublisher) sends task commands to executors, such as ESP and EPS, via DDS.

[0045] The present invention also implements configuration management. Through the configuration manager (Config), thresholds are loaded from the configuration file (prototxt), such as triggering exit when the brake pedal opening is ≥15%. It supports dynamic parameter adjustment and can adapt to different vehicle models or driving scenarios without recompilation, thereby improving the scalability of the system.

[0046] The dynamic process from data input to task execution is a closed-loop process. The specific steps are as follows:

[0047] Step S1: Multi-source data input

[0048] This includes sensor data (such as vehicle sensor CAN bus data), human-machine interface (HMI) instructions, and intelligent driving cockpit data. These data sources are diverse and rich in information, providing basic information for subsequent task scheduling.

[0049] Step S2: MessageCenter aggregates and verifies the input multi-source data

[0050] As the core module for data management, MessageCenter is responsible for unified aggregation of sensor data, HMI instructions, and other information, ensuring data integrity and accuracy. During this step, data is stored in MessageCenter through shared memory and a read-write lock mechanism to ensure thread safety.

[0051] Step S3: ScheduleTask module detects conditions, such as detecting trigger conditions such as T1 to T53

[0052] ScheduleTask reads data from MessageCenter and makes decisions based on pre-set conditional logic. These conditions include driver behavior, vehicle status, environmental perception, and other factors to determine whether to trigger task scheduling.

[0053] Step S4: FSM state migration

[0054] When ScheduleTask detects that certain conditions are met, the FSM state machine drives state transitions. For example, transitioning from the ACC standby state to the LCC active state. State transitions are the core of task scheduling and determine how the autonomous driving system switches between different driving modes.

[0055] Step S5: TaskCommandPublisher issues control instructions

[0056] After the state migration is completed, TaskCommandPublisher sends task instructions to the actuator (such as ESP, EPS) through the data distribution service (DDS), so as to perform the corresponding driving tasks, such as ACC activation or parking mode switching.

[0057] Through the above steps, the present invention realizes a dynamic process from data input to task execution, constructs a closed-loop process of multi-source data aggregation, dynamic condition detection, state machine drive, and real-time instruction issuance, ensuring high reliability and low latency of intelligent driving task scheduling.

[0058] The specific steps of real-time task scheduling are as follows:

[0059] Step 1: Condition detection

[0060] The scheduler checks the conditions (dynamic priority) through GetT1Condition(), such as T1 condition:

[0061] bool T1_condition = (brake pedal not pressed) && (D gear) && (ACC ODD normal) && (user presses the ACC button), that is, the brake pedal is not pressed, the gear is D, the ACC_ODD is normal, and the user presses the ACC button.

[0062] Step 2: State Migration

[0063] If the T1 condition is met, the FSM engine drives the state transition:

[0064] stateDiagram

[0065] [*]-->OFF

[0066] OFF-->ACC_Standby: T1 condition is met

[0067] ACC_Standby-->LCC_Standby: T3 condition (turn signal off + LCC ODD normal)

[0068] Step 3: Send the command

[0069] Publish the ACC activation command (DDS Topic: GPURL_APPLICATION_TASK_COMMAND) through TaskCommandPublisher.

[0070] Regarding exception handling, when the brake pedal opening exceeds a threshold (e.g., 15%), the T36 condition is triggered, forcing the system to switch to the OFF state and resetting all buttons. The average system response time is less than 100ms, and the misoperation rate is reduced to less than 0.1%. This handling mechanism improves system performance and ensures timely exit in abnormal situations, avoiding potential risks.

[0071] The following are examples of key codes:

[0072]

[0073] The present invention involves multiple core abbreviations, which are as follows: for example, FSM, which stands for Finite State Machine, is a model for mutual transfer between different states for task scheduling; DDS, which stands for Data Distribution Service, is a commonly used communication middleware in ROS2 and the field of autonomous driving, which can realize data interaction and transmission between different modules; ODD, which stands for Operational Design Domain, is used to limit the specific conditions and scenarios for the effectiveness of autonomous driving functions to ensure that the functions operate safely within the design range; and HMI, which stands for Human-Machine Interface, is the interface for the driver to interact with the vehicle's autonomous driving system. Instructions such as the driver operating the ACC activation button are issued through this interface.

[0074] In practical applications, the present invention has demonstrated significant results. In terms of real-time performance, the task scheduling cycle is strictly controlled within 10 milliseconds, achieved through a high-precision timer, thus meeting the stringent real-time requirements of Automotive Safety Integrity Level D (ASIL-D). In terms of reliability, the use of debounced button technology effectively filters out noise interference, significantly reducing the false trigger rate of human-machine interface (HMI) commands by 90%. In terms of scalability, the system supports the dynamic addition of new tasks, such as the NOA navigation function, by simply expanding the state and condition detection logic of the finite state machine (FSM).

[0075] This invention successfully implements highly reliable and low-latency intelligent driving task scheduling by building a closed-loop process that aggregates multi-source data, detects dynamic conditions, drives state machines, and issues real-time commands. This innovative solution not only offers significant technical advantages but also has broad applicability across a wide range of scenarios, including L2 to L4 autonomous driving systems.

[0076] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A real-time task scheduling method for intelligent driving based on multi-source information aggregation, characterized in that: The following steps are involved: Multi-source data input; Message center data aggregation and verification; Task scheduler detection conditions; Finite state machine state transition; The message publisher issues control instructions.

2. The method for real-time task scheduling of intelligent driving based on multi-source information aggregation according to claim 1, characterized in that: Multi-source data includes vehicle sensor data, intelligent driving cabin data, MCU data, planning data and perception data.

3. The method for real-time task scheduling of intelligent driving based on multi-source information aggregation according to claim 2 is characterized in that: Message center data aggregation and verification achieve thread safety through shared memory and read-write lock mechanisms.

4. The method for real-time task scheduling of intelligent driving based on multi-source information aggregation according to claim 3 is characterized in that: The task scheduler detection conditions include driver behavior conditions, vehicle status conditions, environmental perception conditions and system health status.

5. The method for real-time task scheduling of intelligent driving based on multi-source information aggregation according to claim 4 is characterized in that: The finite state machine state transition includes transitioning from the ACC standby state to the LCC active state.

6. The method for real-time task scheduling of intelligent driving based on multi-source information aggregation according to claim 5, characterized in that: The message publisher issues control instructions, including sending task instructions to the executor through the data distribution service.

7. The method for real-time task scheduling of intelligent driving based on multi-source information aggregation according to any one of claims 1 to 6, characterized in that: It also includes configuration management, which loads thresholds from configuration files through the configuration manager and supports dynamic adjustment of parameters.

8. The method for real-time task scheduling of intelligent driving based on multi-source information aggregation according to any one of claims 1 to 6, characterized in that: It also includes an exception handling mechanism that forces a switch to a safe state and resets related states when an abnormal condition is detected.

9. The method for real-time task scheduling of intelligent driving based on multi-source information aggregation according to any one of claims 1 to 6, characterized in that: The task scheduling method meets the ASIL-D real-time requirements, and the task scheduling period is ≤10ms.

10. The method for real-time task scheduling of intelligent driving based on multi-source information aggregation according to any one of claims 1 to 6, characterized in that: Message de-jitter technology is used to reduce the false trigger rate of HMI commands and improve system reliability.

Images