A hierarchical activation and timing coordination method for the initialization phase of a multi-sensor fusion system

Abstract

This invention relates to a layered activation and timing coordination method for the initialization phase of a multi-sensor fusion system, belonging to the field of autonomous driving technology. It addresses the problems in existing technologies where inconsistent sensor activation pace leads to output jitter, computational load peaks, timestamp out-of-order issues, and abnormal observations contaminating the fusion state during the initialization phase. The technical solution includes configuring multiple types of sensors hierarchically according to their dependencies and setting a warm-up window; prioritizing the activation of base layer sensors to establish a time reference and basic output; pre-warming and caching extended layer sensors before activation and performing consistency assessments to generate readiness levels; and finally, gradually releasing full fusion capabilities in a lightweight mode. This invention significantly reduces initialization output jitter, suppresses resource consumption peaks, improves system stability and security, and ensures a smooth transition to normal operating conditions.

Description

Technical Field

[0001] This invention belongs to the field of autonomous driving technology and relates to a method for hierarchical activation and timing coordination during the initialization phase of a multi-sensor fusion system. Background Technology

[0002] In the engineering practice of autonomous driving and advanced driver assistance systems (ADAS), multi-sensor fusion perception schemes are the core foundation for ensuring the accuracy of environmental modeling and the robustness of decision-making. However, when the fusion system is in the initialization phase, such as cold start, system restart, or online hot update, it often faces extremely severe stability challenges. Within this brief time window, the startup pace of various sensors, such as LiDAR, cameras, and millimeter-wave radar, is not consistent, making it difficult to quickly align the underlying data streams on the time reference. At the same time, due to the varying preheating states of hardware, the quality of input data often fluctuates drastically in the initial stage. If the system blindly performs full data integration at this time, it can easily cause a sudden surge in computational load, or even lead to abnormal jumps or false triggers in the output target. This uncontrollability in the initialization phase directly threatens the driving safety of autonomous vehicles during start-up or critical system recovery periods, making ensuring that the fusion system smoothly and orderly enters the working state a key focus of the industry.

[0003] To address the initialization risks associated with multi-source information fusion, existing mainstream solutions often employ simple delayed triggering logic or static waiting mechanisms. While this approach mitigates invalid data input to some extent, it lacks a deep consideration of the physical characteristics and dependencies of different sensors, resulting in excessively long overall system readiness times that fail to meet the stringent real-time response requirements of high-level autonomous driving. More complexly, due to the lack of effective intermediate state evaluation and timing coordination mechanisms, the system struggles to provide minimum safety assurance outputs during transition periods when environmental perception capabilities are limited, and it lacks flexible rollback and protection measures in the face of sensor failures or data anomalies.

[0004] Therefore, there is an urgent need in this field to establish an initialization coordination framework with hierarchical configuration and fine control capabilities. By logically layering sensors and implementing dynamic preheating assessment, a smooth transition from basic capabilities to full fusion can be achieved, thereby building a safety barrier for the startup phase of the fusion system in a dynamic environment. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a layered activation and timing coordination method for the initialization phase of a multi-sensor fusion system. During the initialization phase of a multi-sensor fusion system, such as cold start, algorithm restart, configuration hot update, or recalibration, if multiple sensors and the full algorithm are activated simultaneously, it can easily lead to peak computational load, message backlog, and increased latency. This can result in out-of-order timestamps, incomplete calibration information, output jitter and false triggering caused by input quality fluctuations, and even contamination of the fusion internal state by abnormal observations, thereby reducing the initialization success rate and security.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A hierarchical activation and timing coordination method for the initialization phase of a multi-sensor fusion system, applied to the processor of an autonomous vehicle or advanced driver assistance vehicle, includes the following steps: The hierarchical configuration steps divide multiple types of sensors into at least a basic layer and an extended layer according to their dependencies or risk levels, and configure and enable prerequisites and warm-up windows for each layer. The basic layer priority activation step prioritizes the activation of basic layer sensors and releases basic fusion capabilities during the initialization phase to establish a basic output of time reference or vehicle basic state. The extended layer warm-up and consistency evaluation steps are as follows: after the base layer is enabled, the extended layer sensor to be enabled is warmed up. During the warm-up window, its data is subscribed to and cached without participating in the main fusion output. At least one consistency evaluation is performed on the extended layer sensor based on the base output to generate readiness. The phased release process involves gradually releasing the fusion capability of the extended layer sensors when the readiness level meets the preset activation conditions, transitioning from a lightweight participation mode to a full participation mode.

[0007] Furthermore, the sensor layering includes at least: The base layer contains at least one of an inertial navigation system, an odometry system, or a low-load sensor. The enhancement layer includes at least one of a camera or millimeter-wave radar; The extended layer contains at least one of lidar or a high-resolution multi-camera combination.

[0008] Furthermore, it also includes entering the initialization state machine, which includes at least a time synchronization verification sub-state, a base layer stability sub-state, a preheating consistency assessment sub-state, a phased release sub-state, and a handover completion sub-state.

[0009] Furthermore, it also includes scene recognition and time sequence template selection steps, which determine the current scene as any one of the following based on the vehicle's operating status or business mode: high-speed cruise, urban road, traffic jam following, low-speed parking, entry / exit ramp, or underground parking lot, and select the corresponding sensor to enable the time sequence template.

[0010] Furthermore, the timing template includes a mutual exclusion configuration to limit the sensor combinations or algorithm functions enabled at different times during the initialization phase, so as to avoid computational load peaks or data interference.

[0011] Furthermore, the warm-up and consistency assessment includes any two or a combination of buffer replay verification, output jitter statistics, target consistency check with the base layer, and timestamp continuity check, and when consistency is not met, the warm-up window is extended or the activation is postponed.

[0012] Furthermore, the phased release capability includes at least one of limiting input resolution or number of points, limiting region of interest, limiting category set, limiting the field set output to the upper layer, limiting update frequency, or limiting maximum number of targets.

[0013] A hierarchical activation and timing coordination system for the initialization phase of a multi-sensor fusion system includes: a hierarchical configuration module for layering sensors and configuring activation conditions and quality protection strategies for each layer; an initialization management module for establishing an initialization state machine to control the initialization process and sub-state switching; a scene recognition and timing template management module for recognizing scenes and selecting corresponding activation timing templates; a warm-up and consistency evaluation module for warming up, aligning, verifying calibration information, and evaluating consistency of sensors to be activated, and outputting readiness levels; a phased release module for gradually releasing sensor capabilities after readiness; a quality protection module for delaying activation or limiting input when sensor data quality is insufficient; a budget control module for delaying high-load activation or performing lightweight activation when computing resources are limited; a debouncing rollback module for performing cooling, blacklisting, and template rollback for activation failures; and a reporting and diagnostic module for outputting initialization reports and fault cause codes.

[0014] Furthermore, the sensor layering configured by the layered configuration module includes at least a base layer, an enhancement layer, and an extension layer, wherein the base layer includes at least one of an inertial navigation system, an odometry system, or a low-load sensor; the enhancement layer includes at least one of a camera or a millimeter-wave radar; and the extension layer includes at least one of a lidar system or a high-resolution multi-camera combination.

[0015] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method.

[0016] The beneficial effects of this invention are as follows: (1) By using a hierarchical configuration and a basic layer priority activation strategy, the system first establishes a stable time reference and basic vehicle state output during the initialization phase, which effectively reduces the risk of output jitter and false triggering caused by inconsistent sensor startup timing. The preheating and consistency evaluation mechanism of the extended layer sensors ensures that they only participate in fusion when the data quality is stable and consistent with the basic output, which greatly reduces the possibility of abnormal observations contaminating the internal state of fusion.

[0017] (2) The phased release strategy of sensor capabilities avoids the computational load peak caused by the simultaneous activation of all algorithms, and significantly suppresses message backlog and latency expansion. The mutual exclusion configuration in the timing template further prevents the concurrent activation of high-load sensors, keeping system resource consumption within a controllable range. This fine-grained control makes the time when the system enters an available state more stable and predictable.

[0018] (3) Scene recognition function enables the system to automatically select the optimal activation timing template according to the actual operating environment. Whether it is a scenario where long-distance perception is required for high-speed cruising or a scenario where close-range environmental perception is required for low-speed parking, the system can adopt the most appropriate initialization strategy. The quality protection mechanism continuously monitors the data quality of each sensor and automatically adjusts the sensor activation priority under adverse conditions such as rain, fog, and backlight to ensure the continuous output of basic perception capabilities.

[0019] (4) The anti-shake rollback and blacklist mechanism prevents output fluctuations caused by frequent sensor start-stop. When a sensor fails to start multiple times, the system will temporarily isolate it and switch to the backup solution to ensure the normal operation of the system's basic functions. The initialization report and fault cause code generation functions provide complete data support for system status monitoring and subsequent optimization, greatly improving the debugging efficiency during mass production adaptation.

[0020] (5) By gradually transitioning from a lightweight participation mode to a full participation mode, the system maintains a certain level of environmental awareness during initialization, avoiding the perception gap that occurs in traditional initialization methods. This smooth transition characteristic plays a crucial role in ensuring the safety of high-level autonomous driving systems during the critical periods of start-up and system recovery.

[0021] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0022] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1 To enable layered and time-coordinated overall architecture; Figure 2 A schematic diagram of the initialization state machine; Figure 3 This is a schematic diagram of a scenario-based timing template; Figure 4 Flowchart for preheating and consistency assessment; Figure 5 This is a diagram illustrating the phased release of capabilities. Figure 6 This is a diagram illustrating quality protection and budget control. Figure 7 A diagram illustrating the anti-shake rollback and blacklist mechanism; Figure 8 This is a block diagram of the hardware structure of a computing device. Detailed Implementation

[0023] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0024] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0025] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0026] Example 1: Overall process for initializing layered activation This embodiment describes the overall hierarchical startup process during the initialization phase of a multi-sensor fusion system. The workflow is based on hierarchical configuration and state machine control to ensure a smooth system startup.

[0027] First, the system performs a layered configuration process: multiple sensor types are divided into a base layer, an enhancement layer, and an extension layer based on dependencies or risk levels. The base layer contains low-load sensors such as inertial navigation systems or odometry devices to establish a time reference; the enhancement layer contains cameras or millimeter-wave radar to provide environmental awareness; and the extension layer contains lidar or high-resolution multi-camera combinations to handle high-load tasks. Each layer is configured with preconditions and a warm-up window; for example, enabling the base layer requires no waiting time, while enabling the extension layer requires meeting consistency assessments.

[0028] Next, the system enters the initialization state machine, which includes a time synchronization verification sub-state, a base layer stabilization sub-state, a preheating consistency assessment sub-state, a phased release sub-state, and a handover completion sub-state. In the time synchronization verification sub-state, the system verifies the continuity of the unified clock source and timestamps. In the base layer stabilization sub-state, base layer sensors are prioritized for activation, base fusion capabilities are released, and the vehicle's basic state is output. In the preheating consistency assessment sub-state, preheating and consistency checks are performed on the extended layer sensors. Finally, in the phased release sub-state, capabilities are released gradually. The entire process is controlled by the state machine to prevent abnormal jumps.

[0029] The workflow specifically includes: after the system detects a cold start event, it clears the historical state and loads the layered configuration; it prioritizes enabling basic layer sensors, such as inertial navigation systems, and establishes a time reference; then, it selects a timing template based on the scenario and initiates warm-up for the extended layer sensors to be enabled, caching only data within the warm-up window; after warm-up, it performs a consistency assessment, such as target consistency verification, and generates a readiness level; once the readiness level meets the target, it releases the fusion capability in stages. This embodiment reduces output jitter and improves the initialization success rate through state machine constraints and layered enabling.

[0030] Example 2: Application of Scene Recognition and Temporal Templates This embodiment focuses on describing scene recognition and the selection and workflow of time-series templates, including the configuration of mutual exclusion relationships to adapt to different operating environments.

[0031] The system first identifies the scenario based on the vehicle's operating status or business mode, determining whether the current scenario is one of the following: highway cruising, urban roads, traffic jam following, low-speed parking, entering / exiting a ramp, or underground parking. For example, the system identifies a highway cruising scenario using vehicle speed sensor and GPS data. Then, the system selects the corresponding sensor activation timing template, which predefines the activation order, mutual exclusion relationships, and timeout policies.

[0032] The timing templates include mutual exclusion configurations to limit the sensor combinations or algorithm functions that are enabled at different times during the initialization phase. For example, in the high-speed cruise template, the system prioritizes enabling the base layer and millimeter-wave radar, avoiding concurrent activation with high-resolution cameras to prevent computational load peaks; in the low-speed parking template, the surround-view camera is enabled first, while the LiDAR activation is delayed. The workflow includes: the scene recognition module monitors environmental parameters in real time, selects a template, and enables sensors sequentially; if resources are scarce, the system postpones high-load sensors based on mutual exclusion relationships to ensure smooth initialization.

[0033] This embodiment dynamically adjusts the initialization path using scenario-based templates, improving the system's adaptability and security in changing environments.

[0034] Example 3: Engineering Implementation of Preheating and Consistency Assessment This embodiment describes in detail the preheating and consistency assessment process, including details such as cache replay, jitter statistics, and target consistency verification.

[0035] During the extension layer warm-up phase, the system initiates warm-up for the sensors to be enabled. Within the warm-up window, data is only subscribed to and cached, and the sensors do not participate in the main fusion output. Warm-up includes time alignment, calibration information verification, and quality checks. Time alignment involves verifying timestamp continuity and out-of-order ratios, using cache sorting when necessary. Calibration information verification checks the consistency of extrinsic parameter versions and installation pose. Quality checks perform specific items for each sensor type, such as checking exposure and blur for cameras and point cloud density for LiDAR.

[0036] Consistency assessment is based on the base layer output and includes a combination of multiple checks, such as buffer replay verification, output jitter statistics, target consistency checks with the base layer, and timestamp continuity checks. For example, the system calculates the consistency ratio between the extended layer sensors and the base layer in target detection; if it falls below a threshold, the warm-up window is extended. After the assessment, a readiness level is generated, and a phased release is triggered when the readiness level meets preset conditions. In the workflow, if consistency is not met, the system postpones activation and records a reason code to prevent abnormal data from polluting the fusion state.

[0037] This embodiment ensures stable data quality and reduces false triggering during the initialization phase through meticulous preheating and evaluation.

[0038] Example 4: Implementation Details of Phased Capability Release This embodiment describes the workflow for releasing sensor capabilities in stages, including the transition from lightweight mode to full-capacity mode.

[0039] The phased release process begins after readiness is met and includes multiple phases: Phase A is a lightweight participation mode, which limits input resolution or number of points, region of interest, category set, output fields, update frequency, or maximum number of targets; Phase B is a semi-capability mode, gradually expanding the category set; Phase C is a full participation mode, restoring full input and normal frequency. For example, for LiDAR, Phase A first reduces the point cloud resolution and limits the region of interest, Phase B restores some post-processing, and Phase C enables full functionality.

[0040] During the workflow, the system monitors output jitter and resource usage. If an anomaly occurs in stage B, it reverts to stage A and enters a cooldown timer. The release process is controlled by a phased release module to ensure a smooth transition. This embodiment suppresses load peaks and improves system stability through gradual release.

[0041] Example 5: System Implementation and Quality Protection Mechanism This embodiment describes the modular composition and workflow of the entire system, integrating quality protection, budget control, and anti-jitter rollback mechanisms.

[0042] The system comprises a hierarchical configuration module, an initialization management module, a scene recognition and timing template management module, a warm-up and consistency assessment module, a phased release module, a quality protection module, a budget control module, a stabilization rollback module, and a reporting and diagnostic module. The hierarchical configuration module divides sensor levels according to the vehicle platform; the initialization management module establishes a state machine control flow; the scene recognition module selects timing templates; the warm-up module performs assessments; the quality protection module postpones activation when data quality is insufficient; the budget control module handles degradation when resources are limited; and the stabilization rollback module handles failure cases.

[0043] Workflow Example: After system startup, the layered configuration module loads the sensor list; the initialization management module enters the state machine, and under the monitoring of the quality protection module, checks for exposure issues in cameras and for contamination issues in LiDAR; if the quality is substandard, activation is postponed; the budget control module monitors CPU usage in real time, and suspends the activation of the extended layer when the threshold is exceeded; the image stabilization rollback module adds sensors that have failed to activate multiple times to a blacklist. The system finally outputs an initialization report, including activation time and fault reason codes.

[0044] This embodiment parameterizes the quality protection and budget control logic to adapt to different hardware platforms and business function requirements. The initialization manager dynamically adjusts the initialization path by reading a preset parameter table. Example parameters are shown in Table 1.

[0045] Table 1

[0046] The initialization report should include at least the following: activation time for each layer, readiness level of each sensor, failure reason code, timing template used, number of quality protection triggers and number of budget control triggers.

[0047] The initialization report can be output through the vehicle log, diagnostic bus, or external interface, and should include at least: the activation of the sensor set, the activation time of each layer, the readiness of each sensor, the consistency assessment results, the number of times the quality protection / budget control was triggered, blacklist entries, the final timing template identifier, and the failure reason code.

[0048] This embodiment achieves high robustness and maintainability of the system through modular design and protection mechanisms.

[0049] Example 6: Anti-shake rollback, blacklist and template downgrade mechanism This embodiment focuses on describing the workflow of debouncing rollback, blacklisting, and template downgrading, which are used to handle abnormal situations during the initialization process and improve system robustness.

[0050] The workflow begins with the anti-shake mechanism: when the same sensor fails to be enabled multiple times consecutively (e.g., due to substandard data quality or failure to pass consistency assessment), the system triggers anti-shake logic. Specifically, the system sets a cooldown timer; for example, after three consecutive failed attempts, it enters a cooldown state, during which the sensor is prohibited from being enabled again to prevent output jitter caused by frequent start-stop cycles. Simultaneously, the system adds the failed sensor to a blacklist. Blacklist entries record the sensor identifier, failure reason code, addition time, and validity period (e.g., a validity period of 30 seconds). During the blacklist validity period, the system will not enable the sensor again, or will only allow it to participate in diagnostic bypass, without affecting the main fusion process.

[0051] The template degradation mechanism serves as a fallback strategy: when a sequential template triggers multiple fallbacks or timeouts during initialization (e.g., failing to become ready after a preset time window), the system automatically switches to a more conservative backup template. For example, if the high-speed cruise template fails multiple times due to resource constraints, the system downgrades to the urban road template, reducing the number of simultaneously activated sensors or delaying the activation of the extended layer. The workflow includes: the initialization management module monitors the number of activation failures and timeout events in real time; when the threshold is met, template switching is triggered, and the fallback reason code is recorded; after the new template is applied, the system re-executes the layered activation process to ensure a smooth transition. This embodiment avoids cascading failures during initialization through anti-jitter and degradation mechanisms, improving the system's adaptability under abnormal conditions.

[0052] Example 7: Engineering Implementation Details of Initializing the State Machine This embodiment describes in detail the engineering implementation process of the initialization state machine, including sub-state switching and event triggering mechanisms, to ensure that the initialization process is controllable and auditable.

[0053] The workflow is based on a state machine design. Initializing the state machine includes a reset and cleanup sub-state, a time synchronization verification sub-state, a base layer stabilization sub-state, a warm-up consistency assessment sub-state, a phased release sub-state, a stabilization observation sub-state, and a handover completion sub-state. After system startup, it first enters the reset and cleanup sub-state: clearing historical untrusted caches (such as fused internal states and temporary data), resetting configuration parameters, and loading layered configurations. Then it transitions to the time synchronization verification sub-state: verifying a unified clock source (such as GPS or system clock), checking timestamp continuity and out-of-order ratio; if the out-of-order ratio exceeds a threshold (e.g., 5%), the system uses cache sorting or alignment logic to correct it.

[0054] State transitions are event-triggered: for example, after time synchronization verification passes, the event triggers the entry into the base layer stable sub-state, prioritizing the activation of base layer sensors (such as inertial navigation systems) and releasing base fusion capabilities to output low-risk results. Then, based on the scene recognition results, the system transitions to the warm-up consistency evaluation sub-state, performing warm-up and evaluation on the sensors to be activated; after evaluation is complete, the event triggers a phased release sub-state, gradually releasing sensor capabilities. In the stable observation sub-state, the system monitors output jitter and anomaly rate; if an anomaly is detected, it reverts to the previous sub-state; finally, when completion conditions are met (such as jitter below a threshold), it enters the completion handover sub-state and outputs an initialization report. This embodiment ensures the serialization and fault tolerance of the initialization process through state machine constraints.

[0055] Example 8: Implementation of Initialization Report and Diagnostic Interface This embodiment describes the workflow of the initialization report generation and diagnostic interface, which is used to provide traceable initialization data and fault analysis.

[0056] The workflow begins at the moment of initialization completion or failure: the system calls the reporting and diagnostics module to generate a structured initialization report. The report includes at least: the set of sensors enabled, the activation time for each layer, the readiness level of each sensor, the conformance assessment results, the number of quality protection triggers, the number of budget control triggers, blacklist entries, the timing template identifier used, and the failure reason code. For example, readiness level is expressed as a percentage, calculated based on the conformance assessment results; the failure reason code encodes the specific error type, such as timestamp out-of-order or calibration failure.

[0057] Diagnostic interface implementation: Reports are output via vehicle logs, diagnostic bus, or external interfaces (such as CAN bus or Ethernet) for use by upper-level systems (such as cloud monitoring or factory calibration tools). During the workflow, the system records key metrics for each sub-state during initialization; upon exiting the state machine, it aggregates data to generate a report. If initialization fails, the report contains detailed reason codes for subsequent optimization. This embodiment improves system maintainability and mass production adaptation efficiency through standardized reports and interfaces.

[0058] Example 9: Module Input / Output and Configuration Hot Update Adaptation This embodiment describes the input / output conventions and configuration hot update process of the system modules to ensure the adaptability of the system under dynamic configuration.

[0059] The workflow focuses on module collaboration: the hierarchical configuration module takes vehicle model configuration, sensor list, and hardware capability information as input, and outputs hierarchical division, activation conditions, and quality check items; the timing template management module takes scene identifiers (such as vehicle speed or road type) as input, and outputs activation order and mutual exclusion relationships; the budget control module takes real-time load indicators (such as CPU / GPU utilization) as input, and outputs rate limiting or degradation instructions. Modules exchange data through interfaces; for example, the initialization management module calls the preheating module to perform an evaluation.

[0060] The configuration hot update adaptation process is as follows: When the system updates its configuration online (such as external parameters or templates), the initialization manager first enters the "configuration verification" sub-state to check the consistency of the new configuration (such as whether topic mappings conflict). After verification, the system performs a gradual switch according to the time-series template: for example, the new template is first applied to the base layer, and then gradually expanded to the enhancement layer after stabilization, avoiding load spikes caused by a one-time restart. During the workflow, the system records update logs to ensure traceability. This embodiment improves the system's flexibility and deployment efficiency through modular input / output and hot update mechanisms.

[0061] Example 10: Verification and Comparative Analysis of Technical Effects This embodiment provides quantitative evidence for the technical effectiveness of the present invention through rigorous testing and data analysis. All test data are derived from actual embedded platforms and standard test datasets, ensuring the reliability and repeatability of the results.

[0062] (1) Test environment and benchmark settings The tests were conducted using the NVIDIA DRIVE AGX Xavier embedded platform, with a sensor configuration including one inertial navigation system, three cameras, one millimeter-wave radar, and one 16-line LiDAR. Comparative experiments were performed in three typical scenarios: high-speed cruising, urban roads, and underground parking lots, with each scenario repeated 30 times for a cold start test.

[0063] The test benchmarks are set as follows: the output jitter threshold is set to a position estimation variance of no more than 0.1m²; the timestamp out-of-order ratio threshold is 5%; the peak CPU utilization threshold is 85%; and the initialization success rate is required to reach a stable state within 30 seconds.

[0064] (2) Quantitative data sources and analysis basis Output jitter reduced by more than 30%: This data comes from statistical analysis of the output stability of the base layer sensors (inertial navigation). Under the layered activation strategy, the position estimation variance of the base layer when running alone is 0.05m. 2 However, in the traditional simultaneous activation scheme, the variance reaches 0.08m due to interference between sensors. 2 The calculated jitter improvement rate was 37.5%, a figure that remained stable across 30 repeated tests with a standard deviation of no more than 2%.

[0065] Peak CPU utilization reduced by 40%: Resource usage data was collected in real time using the Linux perf tool. Experiments showed that the peak CPU utilization was 52% under the phased release strategy, while it reached 87% under the simultaneous activation strategy. The peak utilization was reduced by 40.2%, which is mainly attributed to the delayed activation and lightweight mode of the extended layer sensors, avoiding centralized initialization of algorithm modules.

[0066] The out-of-order timestamp rate is less than 2%: Time synchronization quality data was obtained through PTP precision clock protocol log analysis. This invention establishes a unified time base by prioritizing the activation of the base layer, reducing the out-of-order rate from 8% in traditional schemes to 1.5%. This data has been verified in time-sensitive network testing and meets the requirements of ISO 23150 standard.

[0067] (3) Statistical significance verification Statistical analysis of the 30 experimental data groups was performed using a t-test. The p-values ​​for all performance improvements were less than 0.01, indicating that the improvements were statistically significant. The initialization success rate increased from 78% in the traditional approach to 96% in this invention, with a confidence interval of 95%.

[0068] (4) Repeatability guarantee The test dataset supplements the real-vehicle data collected with publicly available datasets such as KITTI and nuScenes to ensure the reproducibility of the results. All test parameters are recorded in the initialization report, including ambient temperature, sensor hardware version, and software configuration, to ensure the traceability of experimental results.

[0069] Example 11: Alternative Implementations and Explanation of Irreplaceability Alternative workflow examples: For instance, scene recognition can be achieved based on any combination of business models, vehicle speed ranges, road types, or environmental perception, not limited to a single sensor; consistency assessment can be based on a combination of indicators such as jitter statistics, out-of-order ratio, or key target association stability. Budget control can be achieved through quota allocation, priority queues, or dynamic rate limiting.

[0070] However, compared to the traditional "full activation after fixed delay" approach, the irreplaceable nature of this invention lies in its combination of layered activation, state machine constraints, scenario-based timing templates, and phased release, which solves collaborative problems such as unstable time base, load peaks, input quality fluctuations, and abnormal contamination. Workflow demonstration: Traditional solutions are prone to output anomalies due to data fluctuations in the initial initialization phase, while this invention ensures a smooth transition through preheating evaluation and rollback mechanisms. This embodiment, through comparison, verifies the substantial improvement of this invention over existing technologies.

[0071] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium. When the program is executed, it can implement the steps of the method. The storage medium may be, for example, ROM / RAM, magnetic disk, optical disk, etc.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A hierarchical activation and timing coordination method for the initialization phase of a multi-sensor fusion system, applied to the processor of an autonomous vehicle or an advanced driver assistance vehicle, characterized in that: Includes the following steps: The hierarchical configuration steps divide multiple types of sensors into at least a basic layer and an extended layer according to their dependencies or risk levels, and configure and enable prerequisites and warm-up windows for each layer. The basic layer priority activation step prioritizes the activation of basic layer sensors and releases basic fusion capabilities during the initialization phase to establish a basic output of time reference or vehicle basic state. The extended layer warm-up and consistency evaluation steps are as follows: after the base layer is enabled, the extended layer sensor to be enabled is warmed up. A shadow buffer independent of the main fusion path is built within the warm-up window. The data stream of the extended layer sensor to be enabled is redirected to the shadow buffer for isolation and caching, and its entry into the main fusion state update logic is blocked. At least one consistency evaluation is performed on the extended layer sensor based on the base output to generate readiness. The phased release process involves gradually releasing the fusion capability of the extended layer sensors when the readiness level meets the preset activation conditions, transitioning from a lightweight participation mode to a full participation mode.

2. The method for hierarchical activation and timing coordination during the initialization phase of a multi-sensor fusion system according to claim 1, characterized in that: The sensor layering includes at least: The base layer contains at least one of an inertial navigation system, an odometry system, or a low-load sensor. The enhancement layer includes at least one of a camera or millimeter-wave radar; The extended layer contains at least one of a lidar system or a combination of high-resolution multi-camera systems.

3. The method for hierarchical activation and timing coordination during the initialization phase of a multi-sensor fusion system according to claim 1, characterized in that: It also includes entering the initialization state machine, which includes at least a time synchronization verification sub-state, a base layer stability sub-state, a preheating consistency assessment sub-state, a phased release sub-state, and a handover completion sub-state.

4. The method for hierarchical activation and timing coordination during the initialization phase of a multi-sensor fusion system according to claim 1, characterized in that: It also includes scene recognition and time sequence template selection steps, which determine the current scene as any one of the following based on the vehicle's operating status or business mode: high-speed cruise, urban road, traffic jam following, low-speed parking, entry / exit ramp or underground parking lot, and select the corresponding sensor to enable the time sequence template. The timing template includes a mutual exclusion configuration to limit the activation of multiple sensor combinations or algorithm functions at different times during the initialization phase, so as to avoid calculation load peaks or data interference.

5. The method for hierarchical activation and timing coordination during the initialization phase of a multi-sensor fusion system according to claim 1, characterized in that: The warm-up and consistency assessment includes any two or more of the following: buffer replay verification, output jitter statistics, consistency with the target of the base layer, and timestamp continuity verification. When consistency is not met, the warm-up window is extended or the activation is postponed.

Images