Method and device for determining articulated angle between trailer and tractor and trailer controller

Abstract

The embodiment of the invention provides a method and device for determining the hinging angle of a trailer and a tractor and a trailer controller. The method comprises the following steps: determining a trailer driving state parameter and a tractor speed based on a communication bus signal of a trailer; based on the trailer driving state parameters, the tractor speed and a pre-trained machine learning model, determining an estimated value of a hinging angle between the trailer and the tractor; wherein the machine learning model is obtained by performing supervised training based on multi-dimensional features constructed based on the trailer driving state parameters and the tractor speed in combination with a real measurement value of the hinge angle. The method is used for autonomously and accurately estimating the hinging angle of the trailer, and the hinging angle estimation precision and scene applicability are improved.

Description

Technical Field

[0001] This application relates to the field of vehicle technology, and in particular to a method, apparatus and trailer controller for determining the articulation angle between a trailer and a tractor. Background Technology

[0002] The articulation angle refers to the angle formed between the longitudinal axis of the tractor and the longitudinal axis of the trailer. It is the most core and critical physical quantity describing the relative position and motion state between the tractor and the trailer, and it plays an extremely important role in vehicle dynamics, stability control, and accident prevention and early warning.

[0003] Currently, existing methods for calculating articulation angles mainly involve collecting signals from the tractor side and using kinematic models for calculation.

[0004] However, this method of existing technology requires signals from the tractor side to estimate the articulation angle, which has poor applicability to various scenarios. Summary of the Invention

[0005] The method, apparatus, and trailer controller provided in this application for determining the articulation angle between a trailer and a tractor are used to enable the trailer to autonomously and accurately estimate the articulation angle, thereby improving the applicability of the articulation angle estimation in various scenarios.

[0006] In a first aspect, embodiments of this application provide a method for determining the articulation angle between a trailer and a tractor, comprising:

[0007] Based on the trailer's communication bus signals, determine the trailer's driving status parameters and the tractor's speed;

[0008] Based on the trailer's driving state parameters, the tractor's speed, and a pre-trained machine learning model, the estimated value of the articulation angle between the trailer and the tractor is determined.

[0009] The machine learning model is a multi-dimensional feature constructed based on the trailer's driving state parameters and the tractor's speed, and trained by combining the actual measured values ​​of the articulation angle. The multi-dimensional feature includes original features, statistical features, and combined features.

[0010] In this embodiment, after obtaining the trailer's driving status parameters and the tractor's speed through the trailer's own communication bus signal, the trailer can independently estimate the articulation angle through a machine learning model deployed on the trailer side. This allows the trailer to independently perform subsequent collision avoidance and motion planning control, further improving safety.

[0011] In one possible implementation, the trailer driving state parameters include the wheel speed difference of the trailer's rear axle wheels, the trailer speed, the lateral acceleration of the trailer's rear axle, and the trailer's braking state.

[0012] In one possible implementation, the method includes:

[0013] Based on the trailer driving status parameters and the tractor speed, the original features are constructed;

[0014] Statistical features are constructed based on the first statistical data of the trailer driving status parameters and the second statistical data of the tractor speed within the sliding window;

[0015] The original features and the statistical features are cross-combined to construct combined features;

[0016] The machine learning model is trained based on the original features, the statistical features, the combined features, and the actual measured values ​​of the hinge angle.

[0017] In this embodiment of the application, by utilizing original features, statistical features and combined features to construct a model training dataset, a machine learning model with high estimation accuracy and strong robustness can be trained, thereby improving the accuracy of the articulation angle estimation value and further ensuring the safety of the trailer when making autonomous motion decisions based on the articulation angle estimation value.

[0018] In one possible implementation, the construction of statistical features includes:

[0019] Based on the first statistical data, determine at least one of the following: mean trailer speed, trailer speed change rate, variance of wheel speed difference between trailer rear axle wheels, and continuous braking duration of trailer.

[0020] Based on the second statistical data, determine at least one of the mean tractor speed and the rate of change of tractor speed;

[0021] The statistical features are constructed based on at least one of the following: the mean trailer speed, the rate of change of trailer speed, the mean tractor speed, the rate of change of tractor speed, the variance of wheel speed difference, and the continuous braking duration of the trailer.

[0022] In this embodiment of the application, by using the first statistical data and the second statistical data, the mean, variance, rate of change, etc. of various parameters on the trailer side and the tractor side are determined, which can make the constructed statistical features reflect the dynamic features of the time series data, thereby improving the training effect of the machine learning model.

[0023] In one possible implementation, the construction combination feature includes:

[0024] Based on the trailer driving state parameters, determine the trailer speed, trailer rear axle lateral acceleration, trailer rear axle lateral acceleration after low-pass filtering, wheel speed difference of trailer rear axle wheels, and wheel speed difference of trailer rear axle wheels after low-pass filtering.

[0025] Obtain the measurement difference between the trailer speed measurement value and the tractor speed measurement value;

[0026] The speed difference is calculated based on the trailer speed and the tractor speed.

[0027] Obtain the first numerical difference between the calculated vehicle speed difference and the measured difference, the second numerical difference between the lateral acceleration of the trailer rear axle and the lateral acceleration of the trailer rear axle after low-pass filtering, and the third numerical difference between the wheel speed difference of the trailer rear axle wheels and the wheel speed difference of the trailer rear axle wheels after low-pass filtering.

[0028] Based on the trailer driving state parameters, the tractor speed and the preset kinematic physical model, the calculated value of the articulation angle is determined;

[0029] The combined feature is constructed based on the first numerical difference, the second numerical difference, the third numerical difference, and the calculated hinge angle.

[0030] In the embodiments of this application, by combining and crossing different original features and statistical features to generate more complex combined features, the machine learning model can capture the nonlinear relationship between the articulation angle and the trailer driving state parameters and the tractor speed, enhance the model's sensitivity to input error modeling and nonlinear relationship characterization, and improve estimation accuracy and generalization ability.

[0031] In one possible implementation, determining the calculated articulation angle based on the trailer's driving state parameters, the tractor's speed, and a preset kinematic physical model includes:

[0032] Obtain the initial articulation angle, the first wheelbase from the rear axle of the tractor to the articulation point between the tractor and the trailer, the trailer width, and the second wheelbase from the rear axle of the trailer to the articulation point;

[0033] Based on the tractor speed, the wheel speed difference of the trailer rear axle wheels, the initial hinge angle, the first wheelbase, the trailer width, and the second wheelbase, the calculated value of the hinge angle is determined using a first kinematic physical model.

[0034] In this embodiment of the application, the accuracy of the hinge angle calculation can be improved by using parameters such as the first wheelbase, the second wheelbase, the tractor speed, the wheel speed difference of the trailer rear axle wheels, and the trailer width.

[0035] In one possible implementation, determining the calculated articulation angle based on the trailer's driving state parameters, the tractor's speed, and a preset kinematic physical model includes:

[0036] Based on the tractor speed, the trailer speed, the lateral acceleration of the trailer rear axle, the initial hinge angle, the first wheelbase, and the second wheelbase, the calculated value of the hinge angle is determined using a second kinematic physical model.

[0037] In this embodiment, by utilizing parameters such as the lateral acceleration of the trailer rear axle, trailer speed, tractor speed, first wheelbase, and second wheelbase, the hinge angle can be calculated directly from the trailer driving status parameters contained in the collected CAN signal without needing to obtain the trailer width, thus improving the convenience of hinge angle calculation.

[0038] In one possible implementation, determining the hinge angle calculation value based on the trailer driving state parameters, the tractor speed, and a preset kinematic physical model includes: determining the hinge angle calculation value using a third kinematic physical model based on the wheel speed difference of the trailer's rear axle wheels, the trailer width, the tractor speed, and the second wheelbase.

[0039] In this embodiment, by ignoring the first wheelbase (i.e., treating the first wheelbase as zero), the hinge angle is calculated directly using the wheel speed difference, trailer width, second wheelbase, and tractor speed. This reduces the number of parameters that need to be collected and improves the efficiency of hinge angle calculation.

[0040] In one possible implementation, determining the hinge angle calculation value based on the trailer driving state parameters, the tractor speed, and a preset kinematic physical model includes: determining the hinge angle calculation value using a fourth kinematic physical model based on the trailer rear axle lateral acceleration, the trailer speed, the tractor speed, and the second wheelbase.

[0041] In this embodiment, by ignoring the first wheelbase (i.e., treating the first wheelbase as zero), and using only the tractor speed, the lateral acceleration of the trailer rear axle, the trailer speed, and the second wheelbase to determine the articulation angle calculation value, the additional acquisition of the trailer width parameter can be avoided. This allows for articulation angle calculation in scenarios where trailer width is lacking, thereby improving the adaptability of articulation angle calculation in different scenarios.

[0042] In one possible implementation, determining the hinge angle calculation value based on the trailer driving state parameters, the tractor speed, and a preset kinematic physical model includes: determining the hinge angle calculation value using a fifth kinematic physical model based on the trailer speed and the tractor speed.

[0043] In this embodiment of the application, the efficiency of articulation angle calculation can be improved by directly using the speed of the tractor and the speed of the trailer for inverse cosine calculation.

[0044] In one possible implementation, the method further includes:

[0045] Using the multi-dimensional features as training data, and based on a pre-configured machine learning algorithm, the mapping relationship between the multi-dimensional features and the actual measured value of the hinge angle is fitted to train the machine learning model.

[0046] The performance of the trained machine learning model is verified based on a pre-defined performance verification algorithm.

[0047] If the machine learning model passes the performance verification, then the machine learning model is compressed and optimized to obtain a compressed and optimized machine learning model.

[0048] The compressed and optimized machine learning model is converted into a target data format that matches the operating environment of the trailer controller;

[0049] The machine learning model, after being converted to the target data format, is deployed to the trailer controller.

[0050] In this embodiment, the machine learning model is trained, its performance is verified, compressed and optimized, and its format is converted so that the machine learning model can be deployed to the trailer controller. This enables the trailer to perform online autonomous articulation angle estimation, so that the trailer can make corresponding planning decisions based on the estimated articulation angle, thereby further improving the safety of the trailer.

[0051] In one possible implementation, the method further includes: generating target information based on the estimated articulation angle, the target information including at least one of the trailer's collision risk assessment result, the trailer's planned path, and the trailer's driving control command.

[0052] In this embodiment of the application, the collision risk of the trailer itself is assessed by using the estimated articulation angle obtained by reasoning, and a future driving path is planned for the trailer. The trailer is then controlled to drive according to the future driving path based on control commands. This enables the trailer to drive safely, smoothly and efficiently, thereby further improving the safety of the trailer.

[0053] Secondly, embodiments of this application provide a device for determining the articulation angle between a trailer and a tractor, comprising:

[0054] The determination module is used to determine the trailer's driving status parameters and the tractor's speed based on the trailer's communication bus signals;

[0055] The estimation module is used to determine the estimated value of the articulation angle between the trailer and the tractor based on the trailer driving state parameters, the tractor speed, and a pre-trained machine learning model.

[0056] The machine learning model is a multi-dimensional feature constructed based on the trailer's driving state parameters and the tractor's speed, and trained by combining the actual measured values ​​of the articulation angle. The multi-dimensional feature includes original features, statistical features, and combined features.

[0057] Thirdly, embodiments of this application provide a trailer controller, including: a memory and a processor; the memory stores computer execution instructions; the processor executes the computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0058] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0059] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect. Attached Figure Description

[0060] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0061] Figure 1 This is a schematic diagram of a semi-trailer tractor turning, provided in an embodiment of this application.

[0062] Figure 2 A flowchart for determining the articulation angle between the trailer and the tractor provided in this application;

[0063] Figure 3 A flowchart of machine learning model training provided in the embodiments of this application;

[0064] Figure 4 A flowchart illustrating the training and deployment process of a machine learning model provided in this application embodiment;

[0065] Figure 5 A flowchart illustrating the model training and deployment process provided in this application embodiment;

[0066] Figure 6 A schematic diagram of the device for determining the articulation angle between the trailer and the tractor provided in this application;

[0067] Figure 7 This is a schematic diagram of the trailer controller provided in this application.

[0068] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concepts of this application to those skilled in the art through reference to specific embodiments, and are not intended to limit the scope of protection of this application. Detailed Implementation

[0069] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, unless otherwise indicated, the same reference numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0070] A tractor unit, commonly known as a "truck cab," is a vehicle that provides driving power, steering, and braking control, and provides a driver's operating space. However, it does not carry cargo itself or has a very small cargo box and is specifically designed to tow trailers. Trailers are primarily used for loading goods. Traditional trailers are not powered and need to be towed by a tractor unit. As the carrier of cargo, especially in high-safety-requirement transportation scenarios such as hazardous materials transport, trailers need to possess higher safety and autonomy capabilities. Traditional trailers have a low level of intelligence and struggle to adapt to the precise control demands of complex working conditions, leading to the development of intelligent trailers.

[0071] Intelligent trailers can function as independent intelligent units, possessing the ability to autonomously perceive their environment and their own status. This includes situational awareness of the surrounding environment and real-time assessment of collision risks. They can independently plan and control their movements, achieving autonomous braking and other actions, thereby effectively avoiding collision risks and improving overall transportation safety and automation. Intelligent trailers are compatible with tractor units of varying intelligence levels, enhancing system flexibility and application scope, and ensuring the safety of people, vehicles, and goods.

[0072] Regardless of whether it's a traditional or intelligent trailer, there's always an articulation angle, also known as a folding angle, after the tractor and trailer are connected. This articulation angle is the angle formed between the longitudinal axis of the tractor and the longitudinal axis of the trailer, and it's a key parameter reflecting the vehicle's dynamic state. The articulation angle plays a central role in vehicle control, stability analysis, and autonomous driving systems, especially during cornering, where it's crucial for achieving high-precision vehicle dynamic control and safety warnings.

[0073] Traditional methods for estimating articulation angles mainly include the following: (1) Kinematic model-based calculation method, which calculates the articulation angle by combining data such as wheel speed and acceleration based on the vehicle's kinematics principles. This method relies on signals from the tractor side, which is difficult to meet the needs of independent trailer operation. Moreover, under low-speed slip conditions, wheel speed errors are significant, resulting in insufficient accuracy of kinematic model estimation. (2) Direct measurement method, which obtains angle information in real time by installing a dedicated angle sensor at the articulation mechanism. This method requires the installation of a dedicated sensor, which is diverse and costly, increasing system complexity and post-maintenance difficulty. Furthermore, the diversity of trailer platforms also leads to poor sensor compatibility. (3) Optical image recognition method, which uses an onboard camera to collect images and uses computer vision to identify the relative position of the vehicle to calculate the articulation angle. This method is greatly affected by ambient light and weather, resulting in poor stability.

[0074] To address the aforementioned issues, this application provides a method, apparatus, and trailer controller for determining the articulation angle between a trailer and a tractor. The method acquires the trailer's own driving state parameters and the tractor's speed via the trailer's own communication bus signal. These parameters are then used as input to a machine learning model deployed in the trailer controller. The machine learning model determines an estimated value for the articulation angle between the trailer and the tractor. This allows the trailer to autonomously estimate the articulation angle without relying on signals from the tractor side, enabling the trailer to independently perform subsequent collision avoidance and motion planning control, thereby improving safety.

[0075] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0076] Figure 1 This is a schematic diagram of a semi-trailer tractor turning, provided in an embodiment of this application. Figure 1 As shown, Indicates the hinge angle. Indicates the yaw rate of the tractor unit. This indicates the yaw rate of the trailer. Indicates the speed of the tractor unit. L1 represents the trailer speed, L2 represents the wheelbase from the rear axle of the tractor to the articulation point, and L3 represents the wheelbase from the rear axle of the trailer to the articulation point.

[0077] Traditional kinematic model-based articulation angle estimation methods rely on signals from the tractor side to obtain information such as the tractor's yaw rate before articulation angle estimation can be performed. However, in this application, based on international standards for digital information exchange between the tractor and trailer, the trailer can obtain limited information from the tractor side (such as the tractor's speed) and, combined with its own trailer-side state signals, independently perform articulation angle estimation using a machine learning model deployed in the trailer controller. This eliminates reliance on tractor-side signals and enables independent, autonomous articulation angle estimation by the trailer.

[0078] Among them, this application relates to the articulation angle autonomous estimation technology of intelligent trailers in transportation scenarios. Specific application scenarios include: (1) Complex dynamic working conditions of semi-trailer tractors: such as turning, low-speed driving, braking and other operations, the articulation angle between the tractor and the trailer is a key parameter reflecting the dynamic state of the vehicle. Accurate estimation of the articulation angle is crucial for trailer control, stability analysis and safety warning of the autonomous driving system. (2) Independent intelligent operation requirements of trailers: As an independent intelligent unit, the trailer has the ability to autonomously perceive the environment and its own state, covering collision risk assessment, motion planning and control decision-making (such as autonomous braking), so as to improve transportation safety and automation level. (3) High safety requirements scenarios such as hazardous chemical transportation: As a cargo carrier, especially when transporting flammable, explosive and other dangerous goods, the trailer needs to achieve more accurate risk control through high-precision articulation angle estimation to ensure the safety of people, vehicles and goods. (4) Compatibility with the flexibility requirements of different tractor units: The trailer needs to be compatible with multiple tractor units platforms. It should avoid relying on signals from the tractor unit side (such as steering wheel angle) and only use the trailer's own bus signal to estimate the articulation angle, so as to improve the system flexibility and application range.

[0079] Figure 2 This is a flowchart illustrating the process for determining the articulation angle between a trailer and a tractor unit, as provided in this application. The method is executed by a trailer controller, which is located on the trailer side, enabling the trailer to independently estimate its articulation angle without relying on signals from the tractor unit. Figure 2 As shown, the method includes:

[0080] Step 210: Determine the trailer's driving status parameters and the tractor's speed based on the trailer's communication bus signal.

[0081] Step 220: Based on the trailer driving state parameters, tractor speed and pre-trained machine learning model, determine the estimated value of the articulation angle between the trailer and the tractor.

[0082] The machine learning model consists of multi-dimensional features constructed based on trailer driving state parameters and tractor speed, and supervised training obtained by combining the actual measured values ​​of the articulation angle. The multi-dimensional features include original features, statistical features, and combined features.

[0083] Regarding step 210 above, without relying on the tractor-side signal, due to the international standard for digital information exchange between the tractor and trailer—for example, communication between the trailer and tractor via the ISO 11992-2 protocol—the trailer can still obtain limited information about the tractor (such as the tractor's speed) during communication with the tractor. However, it cannot directly obtain key control commands such as the tractor's steering wheel angle. This solution is based on this limited information, combined with the trailer's own side signal, to achieve independent and autonomous articulation angle estimation for the trailer.

[0084] In this embodiment, the trailer's driving status parameters can be acquired through the trailer's own communication bus signals. For example, sensors on the trailer are connected to different Electronic Control Units (ECUs) within the trailer via a Controller Area Network (CAN) bus. The relevant data acquired by the sensors is transmitted to the ECU via CAN messages. After processing by the ECU to obtain the CAN message signal (i.e., the communication bus signal mentioned in this application), the ECU transmits this communication bus signal to the trailer controller. The trailer controller can obtain the trailer's driving status parameters by reading the message data records in these communication bus signals.

[0085] For example, trailer driving status parameters may include the wheel speed difference of the trailer's rear axle wheels (i.e., the wheel speed difference between the left and right wheels of the trailer's rear axle), trailer speed, lateral acceleration of the trailer's rear axle, and the trailer's braking status.

[0086] It should be noted that the trailer driving status parameters may be obtained directly from the message, or they may be obtained by first reading the relevant physical quantities in the message and then further calculating them based on these physical quantities and related formulas.

[0087] For example, the "measured values" mentioned in this article all originate directly from the messages. The data recorded in these messages comes from the on-board sensors. The on-board sensors process the collected data through different electronic control units (ECUs) to form communication bus signals. For trailers, some messages originate from the ECU of the trailer's electronic braking system (EBS).

[0088] Furthermore, the “calculated value” mentioned in this article means that the “calculated value” is not directly obtained through sensor measurement (i.e., it is not directly derived from the message), but rather the relevant physical quantities in the message are first obtained, and then indirectly calculated through formulas.

[0089] Regarding step 220 above, the original features, statistical features, and combined features are constructed using trailer driving state parameters and tractor speed as the original data sources. For example, the original features are formed by directly performing preliminary processing on the trailer driving state parameters and tractor speed. The statistical features are constructed by statistically analyzing the trailer driving state parameters and tractor speed within a sliding window (e.g., a time window of 1 second), and by analyzing the dynamic changes of these parameters within the sliding window. Finally, the combined features are formed by cross-combining the original features and statistical features.

[0090] By using multi-dimensional features to train the machine learning model, the accuracy of the machine learning model in estimating the hinge angle can be improved.

[0091] The machine learning model is pre-deployed in the trailer controller so that after the trailer driving status parameters and the tractor speed are obtained, the trailer controller can directly call the machine learning model to achieve independent and autonomous estimation of the trailer's articulation angle.

[0092] In this embodiment, the actual measured value of the hinge angle can be obtained in advance by a high-precision sensor (such as a folding angle sensor). Alternatively, the actual measured value of the hinge angle can be calculated using a kinematic model based on signals from the tractor side. For example, high-precision tractor yaw rate, tractor steering wheel angle, and tractor front axle speed can be used to calculate the actual measured value of the angle in advance using a kinematic model.

[0093] In this embodiment, the trailer controller includes the following functional modules or units to perform reasoning on the estimated articulation angle:

[0094] (1) CAN signal acquisition module, which collects sensor data and vehicle speed information (including the vehicle speed of the trailer and the speed of the tractor) on the trailer's CAN bus in real time.

[0095] (2) Real-time feature construction module: Based on the predefined feature engineering scheme, the module processes the collected signals (i.e., trailer driving status parameters and tractor speed, etc.) in real time and constructs multi-dimensional feature inputs consistent with the model training stage.

[0096] (3) Feature input unit: The constructed multi-dimensional features are input into the model inference engine.

[0097] (4) Model inference engine: executes machine learning model and quickly outputs the estimated hinge angle at the current moment.

[0098] (5) Articulation angle estimation processing module: Based on the articulation angle estimation value, the trailer can make autonomous safety decisions.

[0099] For example, in some embodiments, target information can be generated based on the estimated hinge angle.

[0100] The target information includes at least one of the following: the trailer's collision risk assessment results, the trailer's planned route, and the trailer's driving control commands.

[0101] The collision risk assessment results include obstacles in the surrounding environment of the trailer, the collision risk level between the trailer and these obstacles (e.g., high risk, medium risk, and low risk), and the corresponding collision time. For example, the collision risk assessment result could be "A vehicle is rapidly approaching from the left lane and is expected to enter the high-risk area in 2 seconds; collision risk level: high."

[0102] The planned route is the driving path that the trailer should choose to reach its destination safely and efficiently. Based on real-time collision risk assessment results, a safe and drivable driving path can be planned for the trailer from its current location to the target location.

[0103] Among them, driving control commands can be used to precisely control the trailer to travel along the planned path, and can specifically include steering commands, driving commands, and braking commands.

[0104] In this embodiment, by using the estimated articulation angle obtained through reasoning, the collision risk of the trailer itself is assessed, and a future driving path is planned for the trailer. Based on control commands, the trailer is controlled to drive according to the future driving path. This enables the trailer to drive safely, smoothly, and efficiently autonomously, thereby further improving the trailer's safety.

[0105] The method for determining the articulation angle between the trailer and the tractor provided in this application obtains the trailer's driving status parameters and the tractor's speed through the trailer's own communication bus signal. Then, through a machine learning model deployed on the trailer side, the trailer can independently estimate the articulation angle, which allows the trailer to independently perform subsequent collision avoidance and motion planning control, thereby further improving safety.

[0106] Since the trailer only obtains limited information about the tractor (i.e., the tractor's speed), the traditional kinematic model cannot accurately estimate the articulation angle when the information is insufficient. In order to ensure that the trailer can estimate the articulation angle autonomously and independently, and to ensure the accuracy of the estimation, it is necessary to use multi-dimensional features to optimize and train the machine learning model, so as to use the machine learning model to estimate the articulation angle and ensure the accuracy of the estimation.

[0107] In some embodiments, Figure 3 The machine learning model training flowchart provided in the embodiments of this application is as follows: Figure 3 As shown, it includes:

[0108] Step 310: Construct the original features based on the trailer's driving status parameters and the tractor's speed;

[0109] Step 320: Construct statistical features based on the first statistical data of the trailer driving status parameters and the second statistical data of the tractor speed within the sliding window;

[0110] Step 330: Combine the original features and statistical features to construct combined features;

[0111] Step 340: Supervised training of the machine learning model based on the original features, statistical features, combined features, and the actual measured hinge angle.

[0112] Regarding step 310 above, the trailer driving status parameters and the tractor speed are obtained directly by reading the message data records in the CAN message signal, or indirectly calculated using relevant formulas based on the directly obtained data.

[0113] The trailer driving status parameters include the wheel speed difference of the trailer's rear axle wheels, the trailer speed, the lateral acceleration of the trailer's rear axle, and the trailer's braking status.

[0114] In this embodiment, the collected trailer driving status parameters and tractor speed can be pre-processed and then directly converted into original features through characteristic operating conditions.

[0115] For example, preliminary processing may include noise reduction, outlier removal, missing value imputation, and timestamp synchronization to ensure data consistency and reliability. During preliminary processing, online low-pass filters, such as first-order linear low-pass filters or Butterworth 2nd / 4th order filters, may also be used to ensure that the data and the input processing methods used when the machine learning model is deployed to the trailer controller are consistent.

[0116] Regarding step 320 above, the trailer driving status parameters and tractor speed at each time point are recorded with timestamps. At this time, the sliding window is similar to a time window that can slide on the timestamp (the duration of the time window can be 1 second). Each time the time window slides, the trailer driving status parameters and tractor speed at each time point within that duration can be collected.

[0117] The first statistical data may refer to the trailer speed at each point in time within the time period, and the second statistical data may refer to the tractor speed at each point in time within the time period.

[0118] In this embodiment, the average speed of the trailer can be calculated based on the first statistical data; in addition, the average speed of the tractor can also be calculated based on the second statistical data, thereby constructing statistical features.

[0119] Regarding step 330 above, by combining and cross-generating more complex combined features through different original features and statistical features, the nonlinear relationship between the hinge angle and the above features can be captured more accurately, thereby improving the generalization ability of the model.

[0120] When performing feature combination and cross-combination, two or more statistical features can be selected for cross-combination to obtain a combined feature. Alternatively, two or more original features can be selected for cross-combination to obtain a combined feature. Furthermore, at least one statistical feature and at least one original feature can be selected for cross-combination to obtain a combined feature.

[0121] For example, taking the first statistical feature as including the trailer speed and the second statistical feature as including the tractor speed, the speed difference between the trailer and the tractor can be obtained by combining these two statistical features. By using feature engineering to construct features from this speed difference, a combined feature is obtained.

[0122] In step 340 above, the three types of features together constitute the training dataset for training the machine learning model.

[0123] The model training method provided in this application constructs a model training dataset based on original features, statistical features, and combined features. This can train a machine learning model with high estimation accuracy and strong robustness, thereby improving the accuracy of the articulation angle estimation and further ensuring the safety of the trailer when making autonomous decisions based on the articulation angle estimation.

[0124] Furthermore, based on the above embodiments, in some embodiments, statistical features can be constructed through the following steps:

[0125] Step (11): Based on the first statistical data, determine at least one of the following: mean trailer speed, trailer speed change rate, variance of wheel speed difference between trailer rear axle wheels, and continuous braking duration of trailer;

[0126] Step (12): Based on the second statistical data, determine at least one of the mean tractor speed and the rate of change of tractor speed;

[0127] Step (13): Construct statistical features based on at least one of the following: mean trailer speed, trailer speed change rate, mean tractor speed, tractor speed change rate, wheel speed difference variance, and continuous braking duration of the trailer.

[0128] For steps (11) and (12) above, the sliding window can be selected to have a time window of 1 second. If the data sampling frequency is 50 Hz (i.e., 50 times per second), 50 frames of values ​​can be collected in 1 second. Each frame of value represents the trailer driving status parameters and tractor speed at a sampling time point. For example, the value of the 25th frame represents the trailer driving status parameters and tractor speed at 0.5 seconds.

[0129] Based on 50 frames of data, the average speed of the trailer, the average speed of the tractor, the rate of change of the trailer's speed in this second, the rate of change of the tractor's speed in this second, the variance of the wheel speed difference of the trailer's rear axle wheels, and the duration of continuous braking of the trailer in this second can be calculated.

[0130] Regarding step (13) above, when constructing statistical features, based on the trailer speed, the speed segment marker of the trailer speed and the first 20 and first 30 frame values ​​of the speed difference between the left and right rear axles of the trailer can be determined.

[0131] Among them, feature engineering can be used to transform the mean trailer speed, trailer speed change rate, mean tractor speed, tractor speed change rate, wheel speed difference variance, continuous braking duration of the trailer, speed segment marker of the trailer, and the first 20 and 30 frame values ​​of the left and right wheel speed difference of the trailer rear axle to obtain statistical features.

[0132] For example, statistical features include: (1) the average speed of the tractor, the average speed of the trailer, and the variance of the speed difference between the left and right rear axles of the trailer within a time window (e.g., 1 second); (2) the values ​​of the speed difference between the left and right rear axles of the trailer in the first 20 and 30 frames; (3) the speed segment marker of the trailer speed; (4) the speed of the tractor and the rate of change of the trailer speed; and (5) the length of time the trailer has been continuously braked.

[0133] The statistical feature construction method provided in this application determines the mean, variance, rate of change, etc. of various parameters on the trailer side and the tractor side by using statistical data calculated through a sliding window. This allows the constructed statistical features to reflect the dynamic characteristics of time series data, thereby improving the training effect of machine learning models.

[0134] Furthermore, based on the above embodiments, in some embodiments, the combined features can be constructed through the following steps:

[0135] Step (21): Based on the trailer driving state parameters, determine the trailer speed, trailer rear axle lateral acceleration, trailer rear axle lateral acceleration after low-pass filtering, and the wheel speed difference of the trailer rear axle wheels.

[0136] Step (22): Obtain the measurement difference between the trailer speed measurement value and the tractor speed measurement value;

[0137] Step (23): Determine the calculated value of the speed difference based on the trailer speed and the tractor speed;

[0138] Step (24): Obtain the first numerical difference between the calculated speed difference and the measured speed difference, the second numerical difference between the lateral acceleration of the trailer rear axle and the lateral acceleration of the trailer rear axle after low-pass filtering, and the third numerical difference between the wheel speed difference of the trailer rear axle wheels and the wheel speed difference of the trailer rear axle wheels after low-pass filtering.

[0139] Step (25): Determine the calculated value of the articulation angle based on the trailer driving state parameters, the tractor speed and the preset kinematic physical model;

[0140] Step (26): Construct combined features based on the first numerical difference, the second numerical difference, the third numerical difference and the hinge angle calculation value.

[0141] In this embodiment, the trailer's communication bus signal is data collected by on-board sensors and processed by the trailer's ECU. By reading the CAN messages in the communication bus signal, some parameters can be directly obtained, such as the trailer speed, the lateral acceleration of the trailer's rear axle, and the wheel speeds of the left and right wheels of the trailer's rear axle.

[0142] In addition, some parameters can be calculated further using formulas based on these original parameters. For example, the average trailer speed and wheel speed difference are indirectly calculated based on the original parameters.

[0143] In summary, the "measured values" mentioned in this article (such as trailer speed measurement values, tractor speed measurement values, etc.) are all directly derived from CAN messages, i.e., obtained directly from CAN messages. The CAN message signal (i.e., the communication bus signal) is obtained by processing data collected by onboard sensors through different ECUs. In this embodiment, the trailer's CAN message signal can originate from the trailer's EBS ECU.

[0144] The “calculated values” mentioned in this article (such as the calculated value of vehicle speed difference) are obtained indirectly by combining formulas based on the “measured values”.

[0145] The data involved in different CAN messages have different error characteristics, so it can be assumed that the errors they contain are also different.

[0146] Regarding step (21) above, the trailer driving status parameters acquired based on the trailer's communication bus signal have errors. For example, the wheel speed error is large when the trailer is in a low-speed scenario, especially if the trailer brakes, which will further aggravate the error. Thus, the parameters such as the wheel speed difference of the trailer's rear axle wheels obtained indirectly may also have errors.

[0147] Errors can be reduced by using a first-order linear low-pass filter or a Butterworth 2nd / 4th order filter for low-pass filtering.

[0148] Regarding step (22) above, the trailer speed measurement value and the tractor speed measurement value are obtained by sensor measurement, rather than by indirect calculation through formula, and their accuracy is higher.

[0149] For steps (23) and (24) above, by calculating the first numerical difference, the second numerical difference and the third numerical difference as a combined feature, the machine learning model can also learn the magnitude of the error in the parameters acquired through the communication bus signal, thereby improving the model performance.

[0150] For the above step (25), the trailer driving status parameters include the wheel speed difference of the trailer rear axle wheels, trailer speed, trailer rear axle lateral acceleration and trailer braking status.

[0151] The preset kinematic physical model is easily affected by the communication bus signal error, which can lead to a large error in the calculated articulation angle. For example, the wheel speed error is large when the trailer is at low speed, and this error is aggravated when the vehicle is braking.

[0152] In this embodiment, based on the trailer's driving state parameters, the tractor's speed, and a preset kinematic physical model, the calculated value of the articulation angle is determined, and a combined feature is constructed. This allows the machine learning model to learn the difference between the articulation angle calculated from the parameters collected by the communication bus signal and the actual measured value of the articulation angle, thereby improving the model's estimation accuracy.

[0153] For example, the combined features may include: (1) the speed difference between the tractor and the trailer (i.e., the calculated speed difference mentioned above); (2) the numerical difference between the speed difference between the tractor and the trailer and the measured difference, i.e., the first numerical difference; (3) the numerical difference between the lateral acceleration of the trailer rear axle and the lateral acceleration of the trailer rear axle after low-pass filtering, i.e., the second numerical difference; (4) the numerical difference between the wheel speed difference between the left and right wheels of the trailer rear axle and the wheel speed difference between the left and right wheels of the trailer rear axle after low-pass filtering, i.e., the third numerical difference; (5) the hinge angle calculated based on the kinematic physical model, i.e., the hinge angle calculated value.

[0154] When using machine learning models, using the above results as feature inputs allows the model to focus on learning the patterns of how input errors affect the results. Because the error characteristics are obvious, it can enhance the generalization ability.

[0155] The method for constructing combined features provided in this application combines and crosses different original features and statistical features to generate more complex combined features. This enables machine learning models to capture the nonlinear relationship between the articulation angle and the trailer driving state parameters and the tractor speed, enhancing the model's sensitivity to input errors and characterizing nonlinear relationships, thereby improving estimation accuracy and generalization ability.

[0156] The following examples describe in detail how to calculate the articulation angle mentioned above based on limited information obtained from the tractor side by the trailer (such as the tractor front axle speed or tractor speed), the trailer's own driving state parameters, and the kinematic physical model.

[0157] The kinematic model used in traditional hinge angle calculation is as follows:

[0158]

[0159] In the above formula, θ represents the calculated hinge angle. Indicates the initial hinge angle. Indicates the yaw rate of the tractor unit. This indicates the yaw rate of the trailer. L1 represents the speed of the tractor vehicle, L2 represents the wheelbase from the rear axle of the tractor vehicle to the articulation point, T represents the time window, and t represents time.

[0160] Traditional kinematic models calculate the articulation angle θ by integrating the difference in yaw rates between the tractor and trailer. However, in this application, the trailer side can only obtain limited information from the tractor side (such as the tractor's front axle speed or the tractor's speed). Therefore, it is necessary to construct a new kinematic physical model and calculate the articulation angle based on the limited information.

[0161] In some embodiments, the calculated hinge angle value can be determined in the following manner:

[0162] Step (31) Obtain the initial articulation angle, the first wheelbase from the rear axle of the tractor to the articulation point between the tractor and the trailer, the trailer width, and the second wheelbase from the rear axle of the trailer to the articulation point.

[0163] Step (32) Based on the tractor speed, the wheel speed difference of the trailer rear axle wheels, the initial articulation angle, the first wheelbase, the trailer width and the second wheelbase, the calculated value of the articulation angle is determined using the first kinematic physical model.

[0164] In this embodiment, the first kinematic physical model is as follows:

[0165]

[0166] In the above formula, θ represents the calculated hinge angle. Indicates the initial hinge angle. Indicates the speed of the tractor unit. This indicates the difference in wheel speed between the left and right wheels of the trailer's rear axle. L1 represents the trailer width (i.e., the difference in turning radius between the left and right wheels of the trailer's rear axle), L2 represents the first wheelbase from the rear axle of the tractor to the articulation point, L2 represents the second wheelbase from the rear axle of the trailer to the articulation point, T represents the time window, and t represents time.

[0167] In this embodiment, by utilizing parameters such as the first wheelbase, the second wheelbase, the tractor speed, the wheel speed difference of the trailer's rear axle wheels, and the trailer width, the accuracy of the calculated articulation angle can be improved.

[0168] In other embodiments, the calculated hinge angle value can be determined using a second kinematic physical model based on the tractor speed, trailer speed, lateral acceleration of the trailer rear axle, initial hinge angle, first wheelbase and second wheelbase.

[0169] In this embodiment, the second kinematic physical model is as follows:

[0170]

[0171] In the above formula, θ represents the calculated hinge angle. Indicates the initial hinge angle. Indicates the speed of the tractor unit. Indicates the trailer speed. L1 represents the lateral acceleration of the trailer's rear axle, L2 represents the first wheelbase from the tractor's rear axle to the articulation point, T represents the time window, and t represents time.

[0172] In this embodiment, by utilizing parameters such as the lateral acceleration of the trailer rear axle, trailer speed, tractor speed, first wheelbase, and second wheelbase, the hinge angle can be calculated directly from the parameters contained in the collected CAN message signal without needing to obtain the trailer width, thus improving the convenience of hinge angle calculation.

[0173] In other embodiments, the hinge angle can be calculated using a third kinematic physical model based on the wheel speed difference of the trailer's rear axle wheels, the trailer width, the tractor speed, and the second wheelbase.

[0174] In practical applications, the distance between the rear axle of the tractor and the hinge point saddle is very small (less than 10 cm). Relative to the overall length of the vehicle, the first wheelbase L1 can be ignored, meaning L1 is considered to be 0. Based on this, the third kinematic physical model in this embodiment is obtained as follows:

[0175]

[0176] In the above formula, θ represents the calculated hinge angle. This indicates the difference in wheel speed between the left and right wheels of the trailer's rear axle. L2 represents the trailer width, and L2 represents the second wheelbase from the rear axle of the trailer to the articulation point. This indicates the speed of the tractor unit.

[0177] In this embodiment, by ignoring the first wheelbase and directly using the wheel speed difference, trailer width, second wheelbase, and tractor speed, the articulation angle is calculated. This reduces the number of parameters that need to be collected and improves the efficiency of articulation angle calculation.

[0178] In other embodiments, the hinge angle can be calculated using a fourth kinematic physical model based on the lateral acceleration of the trailer rear axle, the trailer speed, the tractor speed, and the second wheelbase.

[0179] In practical applications, the distance between the rear axle of the tractor and the hinge point saddle is very small (less than 10 cm). Relative to the overall length of the vehicle, the first wheelbase L1 can be ignored, meaning L1 can be considered as 0. Based on this, the fourth kinematic physical model in this embodiment is obtained as follows:

[0180]

[0181] In the above formula, θ represents the calculated hinge angle. This indicates the lateral acceleration of the rear axle of the trailer. Indicates the speed of the tractor unit. L1 represents the trailer speed, and L2 represents the second wheelbase from the rear axle of the trailer to the articulation point.

[0182] In this embodiment, by ignoring the first wheelbase and using only the tractor speed, trailer rear axle lateral acceleration, trailer speed, and second wheelbase to determine the articulation angle calculation value, the additional acquisition of the trailer width parameter can be avoided. This allows for articulation angle calculation in scenarios where trailer width is lacking, thus improving the adaptability of articulation angle calculation in different scenarios.

[0183] In other embodiments, the hinge angle can be calculated using a fifth kinematic physical model based on the trailer speed and the tractor speed.

[0184] In this embodiment, based on the geometric relationship between the tractor and the trailer, inverse cosine calculation is performed using speed and the fifth kinematic physical model, which is as follows:

[0185]

[0186] In the above formula, θ is the calculated value of the hinge angle. Indicates the speed of the tractor unit. This indicates the trailer speed.

[0187] In this embodiment, the efficiency of articulation angle calculation can be improved by directly using the speed of the tractor and the speed of the trailer for inverse cosine calculation.

[0188] The articulation angle calculation method provided in the above embodiments calculates the articulation angle value using only limited information (tractor speed or tractor front axle speed) obtained from the trailer side from the tractor side, and without additional reliance on the tractor side signal. This calculated value can be used as training data for the machine learning model, further improving the training effect of the machine learning model. This allows the machine learning model to output a more accurate articulation angle estimate using the input trailer driving state parameters and the limited information obtained from the tractor side.

[0189] The model training and deployment processes are described in detail below through some examples. In some embodiments, Figure 4 The flowchart for training and deploying machine learning models provided in the embodiments of this application is as follows: Figure 4 As shown, it specifically includes the following steps:

[0190] Step 410: Using multi-dimensional features as training data, and based on a pre-configured machine learning algorithm, fit the mapping relationship between the multi-dimensional features and the actual measured values ​​of the hinge angle to train a machine learning model.

[0191] Step 420: Based on the preset performance verification algorithm, verify the performance of the trained machine learning model;

[0192] Step 430: If the machine learning model passes the performance verification, then compress and optimize the machine learning model to obtain the compressed and optimized machine learning model.

[0193] Step 440: Convert the compressed and optimized machine learning model into a target data format that matches the operating environment of the trailer controller;

[0194] Step 450: Deploy the machine learning model, after converting it to the target data format, to the trailer controller.

[0195] Regarding step 410 above, a suitable machine learning algorithm can be selected based on the task characteristics and data features of the trailer articulation angle estimation. For example, machine learning algorithms may include Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Multi-Layer Perceptron (MLP).

[0196] Taking XGBoost as an example, for the ensemble tree model, hyperparameters such as the number of trees, depth, and number of leaf nodes are set, and the ensemble learning algorithm is used to train the model to fit the mapping relationship between the multi-dimensional features constructed from data sampled by complex sensors and the true value of the hinge angle.

[0197] Regarding step 420 above, in order to ensure the estimation accuracy of the trained machine learning model, it is necessary to perform performance verification on the trained model. For example, performance verification algorithms include methods such as cross-validation and test set evaluation.

[0198] After performance verification is completed, a model evaluation report can be generated, and evaluation metrics can be set to assess whether the machine learning model meets the verification pass conditions. For example, evaluation metrics include estimation accuracy, generalization ability, and computational efficiency.

[0199] Regarding step 430 above, if the model evaluation report meets all the evaluation indicators, it means that the machine learning model can pass the performance verification.

[0200] In order to enable machine learning models to be deployed in trailer controllers and achieve embedded deployment, the machine learning models need to be compressed and optimized to reduce model size and computing resource consumption, while ensuring real-time inference performance.

[0201] For example, compression optimization operations may include parameter pruning and knowledge distillation.

[0202] For step 440 above, the format of the trailer controller's operating environment can include TensorRT / ONNX or a manufacturer-customized format. Format conversion and operator adaptation need to be performed to ensure compatibility and performance.

[0203] By converting the machine learning model into a target data format that matches the trailer controller's operating environment, the compatibility and high execution efficiency of the machine learning model after deployment to the trailer controller can be guaranteed.

[0204] For step 450 above, when deploying the model, it is necessary to integrate the final model and related runtime dependencies to generate a complete software package that can be directly deployed to the trailer controller.

[0205] The deployment package can be transmitted to the trailer controller wirelessly or via wired means to enable remote or on-site installation of the model.

[0206] In this embodiment, the machine learning model is trained, its performance is verified, compressed and optimized, and its format is converted so that the machine learning model can be deployed to the trailer controller. This enables the trailer to perform online autonomous articulation angle estimation, so that the trailer can make corresponding planning decisions based on the estimated articulation angle, thereby further improving the safety of the trailer.

[0207] Figure 5 The model training and deployment flowchart provided for the embodiments of this application is as follows: Figure 5 As shown, the model training and deployment process includes a data preparation module, a feature engineering module, and a model training module.

[0208] The data preparation module is used to ensure the quality and accuracy of the data required for model training, thus establishing a high-quality training foundation.

[0209] The raw signal data (from CAN message data records) includes, for example, the speed of the tractor's front axle or the tractor's speed, the wheel speeds of the left and right wheels of the trailer's rear axle, the trailer's speed, the lateral acceleration of the trailer's rear axle, and the trailer's braking status.

[0210] Among them, high-precision articulation angle labels are derived directly from sensors (such as folding angle sensors) or based on kinematic calculations of tractor information (such as using high-precision tractor yaw rate, tractor steering wheel angle, and tractor front axle speed).

[0211] The data cleaning module preprocesses the raw signal data and labels, including noise reduction, outlier removal, missing value imputation, and timestamp synchronization, ensuring data consistency and reliability. An online low-pass filter can be used for data preprocessing to maintain consistency with the input processing methods used during deployment.

[0212] The feature engineering module is used to extract and construct multi-level, multi-dimensional features from cleaned data to enhance the model's expressive power.

[0213] The model training module is used to train high-precision, robust machine learning models using training datasets.

[0214] The method for determining the articulation angle between a trailer and a tractor provided in this application significantly improves the accuracy of articulation angle estimation under complex dynamic conditions such as low-speed slippage, compared to traditional kinematic calculation methods. Furthermore, this method relies solely on the trailer's existing CAN message signals, eliminating the need for additional sensors (such as yaw rate sensors or optical sensors), effectively reducing system costs. It also boasts high accuracy in articulation angle estimation, thereby improving the precision of risk assessment, motion planning, and control for intelligent trailers. Additionally, by constructing multi-dimensional features, the adaptability of the machine learning model to complex operating conditions is enhanced.

[0215] Figure 6 A schematic diagram of the device for determining the articulation angle between the trailer and the tractor provided in this application is shown below. Figure 6 As shown, the device 60 for determining the articulation angle between the trailer and the tractor provided in this embodiment includes:

[0216] The determination module 610 is used to determine the trailer's driving status parameters and the tractor's speed based on the trailer's communication bus signals.

[0217] The estimation module 620 is used to determine the estimated value of the articulation angle between the trailer and the tractor based on the trailer's driving state parameters, the tractor's speed, and a pre-trained machine learning model.

[0218] The machine learning model consists of multi-dimensional features constructed based on trailer driving state parameters and tractor speed, and supervised training obtained by combining the actual measured values ​​of the articulation angle. The multi-dimensional features include original features, statistical features, and combined features.

[0219] In one possible implementation, the trailer driving state parameters include the wheel speed difference of the trailer's rear axle wheels, the trailer speed, the lateral acceleration of the trailer's rear axle, and the trailer's braking state.

[0220] In one possible implementation, a model training module is also included, used for:

[0221] Based on the trailer's driving status parameters and the tractor's speed, the original features are constructed.

[0222] Statistical features are constructed based on the first statistical data of trailer driving status parameters and the second statistical data of tractor speed within the sliding window;

[0223] Combined features are constructed by cross-combining original features and statistical features.

[0224] Supervised training of machine learning models is conducted based on original features, statistical features, combined features, and actual measurements of hinge angles.

[0225] In one possible implementation, the model training module is specifically used for:

[0226] Based on the first statistical data, determine at least one of the following: mean trailer speed, trailer speed change rate, variance of wheel speed difference between trailer rear axle wheels, and continuous braking duration of trailer.

[0227] Based on the second statistical data, determine at least one of the mean tractor speed and the rate of change of tractor speed;

[0228] Statistical features are constructed based on at least one of the following: mean trailer speed, trailer speed change rate, mean tractor speed, tractor speed change rate, wheel speed difference variance, and continuous braking duration of the trailer.

[0229] In one possible implementation, the model training module is specifically used for:

[0230] Based on the trailer's driving state parameters, determine the trailer speed, the lateral acceleration of the trailer's rear axle, the lateral acceleration of the trailer's rear axle after low-pass filtering, the wheel speed difference of the trailer's rear axle wheels, and the wheel speed difference of the trailer's rear axle wheels after low-pass filtering.

[0231] Obtain the measurement difference between the trailer speed measurement value and the tractor speed measurement value;

[0232] The speed difference is calculated based on the trailer speed and the tractor speed.

[0233] The first numerical difference between the calculated and measured speed difference, the second numerical difference between the lateral acceleration of the trailer rear axle and the lateral acceleration of the trailer rear axle after low-pass filtering, and the third numerical difference between the wheel speed difference of the trailer rear axle wheels and the wheel speed difference of the trailer rear axle wheels after low-pass filtering are obtained.

[0234] Based on the trailer's driving state parameters, the tractor's speed, and the preset kinematic physical model, the calculated value of the articulation angle is determined.

[0235] Based on the first numerical difference, the second numerical difference, the third numerical difference, and the calculated hinge angle, a combined feature is constructed.

[0236] In one possible implementation, the model training module is specifically used for:

[0237] Obtain the initial articulation angle, the first wheelbase from the rear axle of the tractor to the articulation point between the tractor and the trailer, the trailer width, and the second wheelbase from the rear axle of the trailer to the articulation point.

[0238] Based on the tractor speed, the wheel speed difference of the trailer's rear axle wheels, the initial articulation angle, the first wheelbase, the trailer width, and the second wheelbase, the calculated value of the articulation angle is determined using the first kinematic physical model.

[0239] In one possible implementation, the model training module is specifically used to: determine the calculated value of the hinge angle using a second kinematic physical model based on the tractor speed, trailer speed, lateral acceleration of the trailer rear axle, initial hinge angle, first wheelbase and second wheelbase.

[0240] In one possible implementation, the model training module is specifically used to: determine the calculated hinge angle value using a third kinematic physical model based on the wheel speed difference of the trailer's rear axle wheels, the trailer width, the tractor speed, and the second wheelbase.

[0241] In one possible implementation, the model training module is specifically used to: determine the calculated hinge angle value using a fourth kinematic physical model based on the lateral acceleration of the trailer rear axle, the trailer speed, the tractor speed, and the second wheelbase.

[0242] In one possible implementation, the model training module is specifically used to: determine the calculated hinge angle based on the trailer speed and the tractor speed using the fifth kinematic physical model.

[0243] In one possible implementation, a model deployment module is also included, used for:

[0244] Using multi-dimensional features as training data, and based on a pre-configured machine learning algorithm, the mapping relationship between the multi-dimensional features and the actual measured values ​​of the hinge angle is fitted to train a machine learning model.

[0245] The performance of the trained machine learning model is verified based on a pre-defined performance verification algorithm.

[0246] If the machine learning model passes the performance verification, then the machine learning model is compressed and optimized to obtain the compressed and optimized machine learning model.

[0247] The compressed and optimized machine learning model is converted into a target data format that matches the operating environment of the trailer controller;

[0248] The machine learning model, after being converted to the target data format, is deployed to the trailer controller.

[0249] In one possible implementation, a trailer control module is also included, which generates target information based on the estimated articulation angle. The target information includes at least one of the following: the trailer's collision risk assessment results, the trailer's planned path, and the trailer's driving control commands.

[0250] The device for determining the articulation angle between the trailer and the tractor provided in this embodiment can perform the method provided in the above-described method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0251] Figure 7 This is a schematic diagram of the trailer controller provided in this application. Figure 7 As shown, the trailer controller 70 provided in this embodiment includes at least one processor 701 and a memory 702. Optionally, the trailer controller 70 further includes a communication component 703. The processor 701, memory 702, and communication component 703 are connected via a bus.

[0252] In a specific implementation, at least one processor 701 executes computer execution instructions stored in memory 702, causing at least one processor 701 to perform the above-described method.

[0253] The specific implementation process of processor 701 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0254] In this embodiment, the trailer controller operates in the trailer's field environment and is responsible for data acquisition, feature calculation, model inference, and control command output.

[0255] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0256] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0257] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0258] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0259] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0260] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0261] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0262] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0263] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0264] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0265] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0266] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0267] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for determining the articulation angle between a trailer and a tractor, characterized in that, include: Based on the trailer's communication bus signals, determine the trailer's driving status parameters and the tractor's speed; Based on the trailer's driving state parameters, the tractor's speed, and a pre-trained machine learning model, the estimated value of the articulation angle between the trailer and the tractor is determined. The machine learning model is a multi-dimensional feature constructed based on the trailer driving state parameters and the tractor speed, and obtained through supervised training by combining the actual measured values ​​of the articulation angle. The multi-dimensional feature includes original features, statistical features and combined features.

2. The method according to claim 1, characterized in that, The trailer driving status parameters include the wheel speed difference of the trailer's rear axle wheels, the trailer speed, the lateral acceleration of the trailer's rear axle, and the trailer's braking status.

3. The method according to claim 1 or 2, characterized in that, The method includes: Based on the trailer driving status parameters and the tractor speed, the original features are constructed; Statistical features are constructed based on the first statistical data of the trailer driving status parameters and the second statistical data of the tractor speed within the sliding window; The original features and the statistical features are cross-combined to construct combined features; The machine learning model is trained under supervision based on the original features, the statistical features, the combined features, and the actual measured values ​​of the hinge angle.

4. The method according to claim 3, characterized in that, The constructed statistical features include: Based on the first statistical data, determine at least one of the following: mean trailer speed, trailer speed change rate, variance of wheel speed difference between trailer rear axle wheels, and continuous braking duration of trailer. Based on the second statistical data, determine at least one of the mean tractor speed and the rate of change of tractor speed; The statistical features are constructed based on at least one of the following: the mean trailer speed, the rate of change of trailer speed, the mean tractor speed, the rate of change of tractor speed, the variance of wheel speed difference, and the continuous braking duration of the trailer.

5. The method according to claim 4, characterized in that, The constructed combined features include: Based on the trailer driving state parameters, determine the trailer speed, trailer rear axle lateral acceleration, trailer rear axle lateral acceleration after low-pass filtering, wheel speed difference of trailer rear axle wheels, and wheel speed difference of trailer rear axle wheels after low-pass filtering. Obtain the measurement difference between the trailer speed measurement value and the tractor speed measurement value; The speed difference is calculated based on the trailer speed and the tractor speed. Obtain the first numerical difference between the calculated vehicle speed difference and the measured difference, the second numerical difference between the lateral acceleration of the trailer rear axle and the lateral acceleration of the trailer rear axle after low-pass filtering, and the third numerical difference between the wheel speed difference of the trailer rear axle wheels and the wheel speed difference of the trailer rear axle wheels after low-pass filtering. Based on the trailer driving state parameters, the tractor speed and the preset kinematic physical model, the calculated value of the articulation angle is determined; The combined feature is constructed based on the first numerical difference, the second numerical difference, the third numerical difference, and the calculated hinge angle.

6. The method according to claim 5, characterized in that, The determination of the articulation angle based on the trailer's driving state parameters, the tractor's speed, and a preset kinematic physical model includes: Obtain the initial articulation angle, the first wheelbase from the rear axle of the tractor to the articulation point between the tractor and the trailer, the trailer width, and the second wheelbase from the rear axle of the trailer to the articulation point; Based on the tractor speed, the wheel speed difference of the trailer rear axle wheels, the initial hinge angle, the first wheelbase, the trailer width, and the second wheelbase, the calculated value of the hinge angle is determined using a first kinematic physical model.

7. The method according to claim 6, characterized in that, The determination of the articulation angle based on the trailer's driving state parameters, the tractor's speed, and a preset kinematic physical model includes: Based on the tractor speed, the trailer speed, the lateral acceleration of the trailer rear axle, the initial hinge angle, the first wheelbase, and the second wheelbase, the calculated value of the hinge angle is determined using a second kinematic physical model.

8. The method according to claim 6, characterized in that, The determination of the articulation angle based on the trailer's driving state parameters, the tractor's speed, and a preset kinematic physical model includes: The hinge angle is calculated using a third kinematic physical model based on the wheel speed difference of the trailer's rear axle wheels, the trailer width, the tractor speed, and the second wheelbase.

9. The method according to claim 6, characterized in that, The determination of the articulation angle based on the trailer's driving state parameters, the tractor's speed, and a preset kinematic physical model includes: Based on the lateral acceleration of the trailer rear axle, the trailer speed, the tractor speed, and the second wheelbase, the calculated value of the hinge angle is determined using a fourth kinematic physical model.

10. The method according to claim 5, characterized in that, The determination of the articulation angle based on the trailer's driving state parameters, the tractor's speed, and a preset kinematic physical model includes: Based on the trailer speed and the tractor speed, the calculated value of the hinge angle is determined using the fifth kinematic physical model.

11. The method according to claim 1, characterized in that, The method further includes: Using the multi-dimensional features as training data, and based on a pre-configured machine learning algorithm, the mapping relationship between the multi-dimensional features and the actual measured value of the hinge angle is fitted to train the machine learning model. The performance of the trained machine learning model is verified based on a pre-defined performance verification algorithm. If the machine learning model passes the performance verification, then the machine learning model is compressed and optimized to obtain a compressed and optimized machine learning model. The compressed and optimized machine learning model is converted into a target data format that matches the operating environment of the trailer controller; The machine learning model, after being converted to the target data format, is deployed to the trailer controller.

12. The method according to any one of claims 1, 2, 4-11, characterized in that, The method further includes: Based on the estimated articulation angle, target information is generated, which includes at least one of the following: the collision risk assessment result of the trailer, the planned path of the trailer, and the driving control command of the trailer.

13. A device for determining the articulation angle between a trailer and a tractor, characterized in that, include: The determination module is used to determine the trailer's driving status parameters and the tractor's speed based on the trailer's communication bus signals; The estimation module is used to determine the estimated value of the articulation angle between the trailer and the tractor based on the trailer driving state parameters, the tractor speed, and a pre-trained machine learning model. The machine learning model is a multi-dimensional feature constructed based on the trailer driving state parameters and the tractor speed, and obtained through supervised training by combining the actual measured values ​​of the articulation angle. The multi-dimensional feature includes original features, statistical features and combined features.

14. A trailer controller, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-12.

15. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-12.

16. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method described in any one of claims 1-12.

Images