Adaptive steering calibration method and device for electric power steering system

Abstract

This application relates to the field of intelligent vehicle control technology, and in particular to an adaptive steering calibration method and apparatus for an electric power steering system. The method includes: constructing a state space vector containing scene identifiers and inputting it into a flexible actor-critic deep reinforcement learning model to generate action increments for hand force parameters; updating the current hand force parameters in real time based on the action increments to change the actual steering hand force of the vehicle; acquiring vehicle feedback data and the current scene identifier when the actual steering hand force is applied to calculate the scene-adaptive reward function value; and storing the state space vector, action increments, reward function value, and vehicle feedback data as an empirical data set in an experience playback buffer for online fine-tuning of the flexible actor-critic deep reinforcement learning model. This solves the problem that existing manual calibration methods based on human experience cannot adapt to the diverse road types and actual drivers.

Description

Technical Field

[0001] This application relates to the field of intelligent vehicle control technology, and in particular to an adaptive steering calibration method and apparatus for an electric power steering (EPS) system. Background Technology

[0002] With the development of the automotive industry, Electric Power Steering (EPS) systems are becoming increasingly common in automobiles. This system provides steering assistance by receiving signals such as vehicle speed, steering wheel angle, angular velocity, and steering torque, optimizing the driver's experience. However, actual road conditions are complex, and EPS systems often cannot adjust in a timely manner according to these conditions. For example, when driving long distances in mountainous areas with numerous turns, the EPS system needs to increase its motor output power to provide more assistance. Similarly, when off-roading, the EPS system often needs to increase its damping coefficient to reduce the steering wheel's kicking sensation. Given the diverse driving needs on various roads, existing EPS systems often cannot adequately cover these situations.

[0003] Traditional calibration methods rely on manual calibration based on human experience, adjusting the power assist curve to achieve the effects of multiple modes while driving on specific roads. However, this method has the following drawbacks: (1) The actual power steering effect of the electric power steering system EPS will be determined by the style of the calibrator. If there is a large deviation between the calibrator and the actual driver's driving style, it may lead to a bad driving experience. (2) Due to the development cycle, manual force calibration is often performed only for a single road, resulting in fewer options for users; (3) Traditional manual parameter tuning methods based on human experience are time-consuming and labor-intensive. With the increase of vehicle platform and functional requirements, the calibration workload increases exponentially. Summary of the Invention

[0004] This application provides an adaptive steering calibration method and device for an electric power steering system to solve the problems of existing manual calibration methods based on human experience being unable to adapt to the ever-changing road types and actual drivers, resulting in poor portability, limited applicable scenarios, and large workload.

[0005] The first aspect of this application provides an adaptive steering calibration method for an electric power steering system, comprising the following steps: Construct a state space vector containing scene identifiers, wherein the state space vector includes real-time vehicle operating status information, driver input information, and current driving scene identifiers; The state space vector is input into a pre-trained flexible actor-critic deep reinforcement learning model to generate the action increment that adjusts the hand force parameters of the electric power steering system. The current hand force parameters are updated in real time based on the action increment, so as to change the actual steering hand force of the vehicle using the updated hand force parameters; Obtain vehicle feedback data and current scene identifier when the actual steering force is applied, in order to calculate the scene-adaptive reward function value; The state space vector, the action increment, the reward function value, and the vehicle feedback data are stored as an experience data set in the experience replay buffer to fine-tune the flexible actor-critic deep reinforcement learning model online.

[0006] Optionally, the construction of the state space vector containing the scene identifier includes: Obtain the current driving scenario identifier; Based on the current driving scenario identifier, the lateral acceleration, yaw rate, steering wheel angle, steering velocity, steering wheel torque, vehicle speed, road surface adhesion coefficient estimate, and road curvature are collected and normalized to construct the state space vector containing the scenario identifier.

[0007] Optionally, the current driving scenario identifier includes Light Mode, Sport Mode, Eco Mode, and Off-road Mode.

[0008] Optionally, the flexible actor-critic deep reinforcement learning model is obtained through offline multi-scene training. The training data includes interaction data under different driving scenarios, and the reward function dynamically adjusts the weights according to different scenario labels during the training process to guide the model to learn the optimal steering force strategy under different scenario labels.

[0009] Optionally, the action increment includes the boost gain increment, the damping coefficient increment, and the return torque coefficient increment.

[0010] Optionally, the reward function value includes an ideal hand force tracking function, an operational stability function, a safety penalty function, and an energy consumption reward function.

[0011] A second aspect of this application provides an adaptive steering calibration device for an electric power steering system, comprising: The construction module is used to construct a state space vector containing scene identifiers, wherein the state space vector includes real-time vehicle operating status information, driver input information, and current driving scene identifiers. The generation module is used to input the state space vector into a pre-trained flexible actor-critic deep reinforcement learning model to generate the action increment that adjusts the hand force parameters of the electric power steering system. The update module is used to update the current hand force parameters in real time according to the action increment, so as to change the actual steering hand force of the vehicle using the updated hand force parameters; The calculation module is used to obtain vehicle feedback data and current scene identifier when the actual steering force is applied, so as to calculate the scene-adaptive reward function value; The fine-tuning module is used to store the state space vector, the action increment, the reward function value, and the vehicle feedback data as an experience data set in the experience playback buffer to perform online fine-tuning of the flexible actor-critic deep reinforcement learning model.

[0012] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the adaptive steering calibration method for an electric power steering system as described in the above embodiments.

[0013] A fourth aspect of the present invention provides a computer program product that, when executed by a processor, implements the above-described adaptive steering calibration method for an electric power steering system.

[0014] A fifth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described adaptive steering calibration method for an electric power steering system.

[0015] The adaptive steering calibration method and apparatus for electric power steering (EPS) proposed in this application, by inputting the state space vector into a pre-trained Soft Actor-Critic (SAC) deep reinforcement learning model, can output the action increment for adjusting the hand force parameters of the EPS. When facing complex road conditions over extended periods, online adaptive adjustment of the EPS can be activated to update the assist curve of the EPS output according to the actual situation. The hand force parameters can be optimized and adjusted online according to the driver's preferences, resulting in a personalized steering feel. The SAC deep reinforcement learning model can learn sufficient parameter control strategies through offline simulation alone, and with a small amount of pre-training, it can be adapted to newly developed vehicle models. The online fine-tuning mechanism can also adjust the hand force parameters according to changes in wear and tear of parts, ensuring EPS assistance adaptation throughout the vehicle's lifecycle.

[0016] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0017] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of an adaptive steering calibration method for an electric power steering system according to an embodiment of this application; Figure 2 This is a schematic diagram illustrating the specific execution of an adaptive steering calibration method for an electric power steering system according to an embodiment of this application; Figure 3 This is a schematic diagram of the online fine-tuning process of a flexible actor-critic deep reinforcement learning model provided in an embodiment of this application; Figure 4 This is a block diagram of an adaptive steering calibration device for an electric power steering system according to an embodiment of this application; Figure 5 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application.

[0018] Explanation of reference numerals in the attached figures: 40-Adaptive steering calibration device for electric power steering system, 401-Building module, 402-Generation module, 403-Update module, 404-Calculation module, 405-Fine-tuning module, 501-Memory, 502-Processor and 503-Communication interface. Detailed Implementation

[0019] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0020] The adaptive steering calibration method and apparatus of the electric power steering system according to embodiments of this application are described below with reference to the accompanying drawings.

[0021] Figure 1 This is a flowchart of an adaptive steering calibration method for an electric power steering system provided according to an embodiment of this application.

[0022] like Figure 1 As shown, the adaptive steering calibration method for this electric power steering system includes the following steps: In step S101, a state space vector containing a scene identifier is constructed. The state space vector includes real-time vehicle operating status information, driver input information, and the current driving scene identifier.

[0023] In some embodiments, constructing a state space vector containing scene identifiers includes: Obtain the current driving scenario identifier; Based on the current driving scenario identifier, the lateral acceleration, yaw rate, steering wheel angle, steering velocity, steering wheel torque, vehicle speed, road surface adhesion coefficient estimate, and road curvature are collected and normalized to construct a state space vector containing the scenario identifier.

[0024] In actual implementation, such as Figure 2 As shown, the current driving scenario identifier is determined based on the driver's selection or by automatically recognizing the vehicle's current state information. Further, based on the current driving scenario identifier, corresponding lateral acceleration, yaw rate, steering wheel angle, angular velocity, steering wheel torque, vehicle speed, estimated road surface adhesion coefficient, and road curvature are collected from real-vehicle sensors. These data are then normalized to construct a state space vector containing the scenario identifier. The current driving scenario identifier includes Light Mode, Sport Mode, Eco Mode, and Off-Road Mode.

[0025] In step S102, the state space vector is input into a pre-trained flexible actor-critic deep reinforcement learning model to generate the action increment that adjusts the hand force parameters of the electric power steering system.

[0026] In some embodiments, the flexible actor-critic deep reinforcement learning model is obtained through offline multi-scenario training. The training data includes interaction data under different driving scenarios, and the reward function dynamically adjusts the weights according to different scenario identifiers during the training process to guide the model to learn the optimal steering force strategy under different scenario identifiers.

[0027] In some embodiments, the motion increment includes the boost gain increment, the damping coefficient increment, and the aligning torque coefficient increment.

[0028] In actual implementation, such as Figure 2 As shown in the embodiment of this application, the interaction data under different driving scenarios will be obtained in advance to conduct offline multi-scenario training of the flexible actor-critic deep reinforcement learning model. During the training process, different driving scenario identifiers need to be traversed, and the reward function of different driving scenario identifiers will be dynamically adjusted according to different scenario identifiers to guide the model to learn the optimal steering force strategy under different scenario identifiers until all driving scenario identifiers have been traversed.

[0029] Furthermore, the state space vector is input into a pre-trained flexible actor-critic deep reinforcement learning model, and the actor network forward calculates and outputs the action increment that adjusts the hand force parameters of the electric power steering system.

[0030] In step S103, the current hand force parameters are updated in real time according to the action increment, so as to change the actual steering hand force of the vehicle using the updated hand force parameters.

[0031] In actual implementation, such as Figure 2 As shown, the current hand force parameters are updated according to the action increment to obtain the updated hand force parameters, and the updated hand force parameters are sent to the electric power steering system EPS controller for execution to change the steering hand force of the vehicle.

[0032] In step S104, vehicle feedback data and current scene identifier are obtained when the actual steering force is applied, so as to calculate the scene-adaptive reward function value.

[0033] In actual implementation, such as Figure 2 As shown, based on the vehicle feedback data after applying actual steering force, and combined with the current scene identifier, the scene-adaptive reward function value is calculated. The reward function value includes the ideal steering force tracking function, the operation smoothness function, the safety penalty function, and the energy consumption reward function.

[0034] In step S105, the state space vector, action increment, reward function value, and vehicle feedback data are stored as an experience data in the experience playback buffer to fine-tune the flexible actor-critic deep reinforcement learning model online.

[0035] In actual implementation, such as Figure 2 As shown, the state space vector, action increment, reward function value, and vehicle feedback data are stored as an experience data in the experience replay buffer. It is determined whether the fine-tuning step has been reached. If not, steps S101-S105 are executed iteratively until the fine-tuning step is reached. The fine-tuning flexible actor-critic deep reinforcement learning model is sampled from the experience replay buffer.

[0036] The adaptive steering calibration method for the electric power steering system proposed in this application will be described in detail below through a specific embodiment.

[0037] This embodiment includes two steps: an offline multi-scenario training phase and an online deployment and fine-tuning phase.

[0038] Step 1: Offline Multi-Scenario Training Phase First, a dynamic model of the entire vehicle was established in the vehicle dynamics simulation software CarSim. Then, controller and driver models were built in the multi-domain visualization modeling and simulation environment Simulink. Finally, different virtual road scenarios were generated using the intelligent driving and ADAS multiphysics simulation platform Prescan, including highways, urban areas, mountain roads, and off-road terrain. The road surface adhesion coefficient was also differentiated by dry, wet, and icy / snowy conditions. The scenario list included four modes: lightweight, sport, economy, and off-road.

[0039] Define the basic elements of a reinforcement learning model; the state space is defined as a 12-dimensional vector, including vehicle speed. Steering wheel angle angular velocity Steering wheel torque lateral acceleration yaw rate In addition, scene identification using 4D one-hot encoding and road surface adhesion coefficient estimation. and road curvature All vector inputs are normalized before being fed into the network. The action space output has three increments: a boost gain increment in the range [-0.1, 0.1]. Damping coefficient increments in the range [-0.5, 0.5] The increment of the aligning torque coefficient in the range [-0.05, 0.05] Limiting the step size is mainly to prevent the vehicle from losing control due to rapid parameter changes. Regarding the reward function, to allow the model to distinguish the feel of the controls in different scenarios, we break the reward down into four parts: ideal hand force tracking... (Deviation from scene adaptive reference torque ratio), operational stability (The smaller the steering wheel angle acceleration, the better), safety penalties (Penalty for lateral acceleration exceeding 0.4g), Energy consumption reward (Motor current square term). The weights for different scenarios are determined according to the target feel requirements, defined in lightweight mode. The weight is 0.3. The weight is 0.5. The weight is 0.1. Weight is 0.1; in motion mode The weight is 0.5. The weight is 0.2. The weight is 0.2. Weight is 0.1; Economic model The weight is 0.3. The weight is 0.2. The weight is 0.1. Weight is 0.4; Off-road mode The weight is 0.3. The weight is 0.2. The weight is 0.4. The weight is 0.1.

[0040] like Figure 3As shown, a SAC deep reinforcement learning model is built. The Actor network inputs the state and scene labels together, passes them through three fully connected layers and one activation function (256→128→64, ReLU), and outputs the mean and log-standard deviation of the actions. The actions are obtained through reparameterized sampling, and finally... Compressed to [-1, 1]. Temperature parameters during training. It will be used to calculate the loss of the Actor network. The Critic network uses two identical structures. The network takes input state, scene identifier, and action as a concatenation, and passes it through three fully connected layers and one activation function layer (256→128→64, ReLU) to output the result. Value. Using two networks is primarily to reduce overestimation. The target network is assigned to each of the two Critic networks. A soft update of 0.005. Temperature parameter. It's a trainable scalar used to adjust the entropy regularization strength; the target entropy is set to -3. Regarding training parameter settings, the discount factor... =0.99, learning rate of each network is 3×10 4 The experience pool capacity is 106, and the batch size is 256.

[0041] like Figure 2 As shown, the SAC deep reinforcement learning model is pre-trained in multiple scenarios. During pre-training, four scenarios are traversed first. In each scenario, the SAC network (Actor, Critic, ...) is randomly initialized. This allows the model to interact with preset working conditions and collect data. Store the data in the experience pool, along with a scene label. During training, sample mini-batches from the experience pool and calculate the target value accordingly. Values, update two Critics, update Actor, update temperature parameters. Then, the target network is softly updated. This process continues until the model for this scene converges. Finally, the Actor network weights are saved, and the model is used to train the next scene. Once all four scenes are completed, the offline multi-scene training phase ends, resulting in a pre-trained flexible actor-critic deep reinforcement learning model.

[0042] Step Two: Online Deployment and Fine-tuning Phase When deploying the pre-trained Actor network, only the mean output is taken. Quantization and code generation are performed using the lightweight open-source edge deep learning framework TensorFlow Lite, and then it is integrated into the domain controller. The ideal hand force model for scene adaptation is also made into a lightweight neural network and deployed into the domain controller. The current driving scene is determined by the driver selecting "Light," "Sport," "Eco," or "Off-Road" on the screen or knob.

[0043] After the vehicle starts, the Actor model corresponding to the current scene identifier is loaded. Sensor data is received via CAN every 10ms: provided by the IMU. and Angle sensor provides and Torque sensor provides Wheel speed sensor provides GPS provides road curvature and type. After data preprocessing and normalization, a state vector containing scene identifiers is generated. The input is fed into the Actor network to obtain the action increment. Superimposed to form new parameters The system sends a command to the EPS actuator, records the current state, action, and next state, calculates the reward based on the current scenario weight, and incorporates the experience data. Stored in the circular buffer zone of the actual vehicle.

[0044] Security monitoring strategies involve real-time monitoring. and .once >0.45g or If the threshold is exceeded, immediately switch the electric power steering (EPS) system parameters back to safe default values, pause online learning, and only resume online learning when the vehicle is stationary. Online fine-tuning involves randomly sampling 32 data points every second from the real-vehicle buffer, adjusting the Actor, two Critics, and temperature parameters for the current scenario. Perform a small number of gradient updates, and simultaneously perform soft updates to the target network. This allows the model to gradually adapt to current driving habits and scenario preferences, and also compensates for the effects of component aging.

[0045] In summary, the adaptive steering calibration method for the electric power steering system proposed in the embodiments of this application has the following beneficial effects: (1) By inputting the state space vector into the pre-trained SAC deep reinforcement learning model, the action increment for adjusting the hand force parameters of the electric power steering system can be output. When facing long-term complex road conditions, online EPS adaptive adjustment can be turned on to update the power assist curve of the electric power steering system EPS output according to the actual situation. (2) The hand force parameters can be optimized and adjusted online according to the driver's preferences, thereby bringing a personalized steering feel; (3) The SAC model can learn enough parameter control strategies through offline simulation alone, and with a small amount of pre-training, it can be adapted to newly developed models; (4) The online fine-tuning mechanism can also adjust the manual parameters according to the wear and tear of the parts, so as to ensure the electric power steering system EPS power assistance is adapted throughout the vehicle's entire life cycle.

[0046] Next, with reference to the accompanying drawings, an adaptive steering calibration device for an electric power steering system according to an embodiment of this application is described.

[0047] Figure 5 This is a block diagram of an adaptive steering calibration device for an electric power steering system according to an embodiment of this application.

[0048] like Figure 4 As shown, the adaptive steering calibration device 40 of the electric power steering system includes: a construction module 401, a generation module 402, an update module 403, a calculation module 404, and a fine-tuning module 404.

[0049] The system comprises the following modules: Construction module 401 constructs a state space vector containing scene identifiers, including real-time vehicle operating status information, driver input information, and the current driving scene identifier. Generation module 402 inputs the state space vector into a pre-trained flexible actor-critic deep reinforcement learning model to generate action increments that adjust the hand force parameters of the electric power steering system. Update module 403 updates the current hand force parameters in real time based on the action increments, thereby changing the actual steering hand force of the vehicle. Calculation module 404 acquires vehicle feedback data and the current scene identifier when the actual steering hand force is applied to calculate the scene-adaptive reward function value. Fine-tuning module 404 stores the state space vector, action increments, reward function value, and vehicle feedback data as an empirical data set in an experience playback buffer for online fine-tuning of the flexible actor-critic deep reinforcement learning model.

[0050] In some embodiments, the construction module 401 includes: The acquisition unit is used to acquire the current driving scenario identifier; The construction unit is used to collect and normalize lateral acceleration, yaw rate, steering wheel angle, angular velocity, steering wheel torque, vehicle speed, road surface adhesion coefficient estimate, and road curvature based on the current driving scenario identifier, in order to construct a state space vector containing the scenario identifier.

[0051] In some embodiments, the current driving scenario identifier includes Light Mode, Sport Mode, Eco Mode, and Off-Road Mode.

[0052] In some embodiments, the flexible actor-critic deep reinforcement learning model is obtained through offline multi-scene training. The training data includes interaction data under different driving scenarios, and the reward function dynamically adjusts the weights according to different scenario labels during the training process to guide the model to learn the optimal steering force strategy under different scenario labels.

[0053] In some embodiments, the motion increment includes the boost gain increment, the damping coefficient increment, and the aligning torque coefficient increment.

[0054] In some embodiments, the reward function values ​​include an ideal hand force tracking function, an operational stability function, a safety penalty function, and an energy consumption reward function.

[0055] It should be noted that the foregoing explanation of the adaptive steering calibration method embodiment for the electric power steering system also applies to the adaptive steering calibration device of the electric power steering system in this embodiment, and will not be repeated here.

[0056] The adaptive steering calibration device for an electric power steering system proposed according to the embodiments of this application has the following beneficial effects: (1) By inputting the state space vector into the pre-trained SAC deep reinforcement learning model, the action increment for adjusting the hand force parameters of the electric power steering system can be output. When facing long-term complex road conditions, online EPS adaptive adjustment can be turned on to update the power assist curve of the electric power steering system EPS output according to the actual situation. (2) The hand force parameters can be optimized and adjusted online according to the driver's preferences, thereby bringing a personalized steering feel; (3) The SAC model can learn enough parameter control strategies through offline simulation alone, and with a small amount of pre-training, it can be adapted to newly developed models; (4) The online fine-tuning mechanism can also adjust the manual parameters according to the wear and tear of the parts, so as to ensure the electric power steering system EPS power assistance is adapted throughout the vehicle's entire life cycle.

[0057] Figure 5 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application.

[0058] The electronic device may include a memory 501, a processor 502, and a computer program stored in the memory 501 and executable on the processor 502. When the processor 502 executes the program, it implements the adaptive steering calibration method for the electric power steering system provided in the above embodiments.

[0059] Furthermore, electronic devices also include: Communication interface 503 is used for communication between memory 501 and processor 502.

[0060] The memory 501 is used to store computer programs that can run on the processor 502.

[0061] Memory 501 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0062] If the memory 501, processor 502, and communication interface 503 are implemented independently, then the communication interface 503, memory 501, and processor 502 can be interconnected via a bus to complete communication between them. 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 representation, Figure 5 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0063] Optionally, in a specific implementation, if the memory 501, processor 502, and communication interface 503 are integrated on a single chip, then the memory 501, processor 502, and communication interface 503 can communicate with each other through an internal interface.

[0064] Processor 502 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0065] This invention also provides a computer program product, which, when executed by a processor, implements the above-described adaptive steering calibration method for an electric power steering system.

[0066] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described adaptive steering calibration method for an electric power steering system.

[0067] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0068] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0069] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0070] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0071] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0072] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0073] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0074] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. An adaptive steering calibration method for an electric power steering system, characterized in that, Includes the following steps: Construct a state space vector containing scene identifiers, wherein the state space vector includes real-time vehicle operating status information, driver input information, and current driving scene identifiers; The state space vector is input into a pre-trained flexible actor-critic deep reinforcement learning model to generate the action increment that adjusts the hand force parameters of the electric power steering system. The current hand force parameters are updated in real time based on the action increment, so as to change the actual steering hand force of the vehicle using the updated hand force parameters; Obtain vehicle feedback data and current scene identifier when the actual steering force is applied, in order to calculate the scene-adaptive reward function value; The state space vector, the action increment, the reward function value, and the vehicle feedback data are stored as an experience data set in the experience replay buffer to fine-tune the flexible actor-critic deep reinforcement learning model online.

2. The adaptive steering calibration method for an electric power steering system according to claim 1, characterized in that, The construction of the state space vector containing scene identifiers includes: Obtain the current driving scenario identifier; Based on the current driving scenario identifier, the lateral acceleration, yaw rate, steering wheel angle, steering velocity, steering wheel torque, vehicle speed, road surface adhesion coefficient estimate, and road curvature are collected and normalized to construct the state space vector containing the scenario identifier.

3. The adaptive steering calibration method for an electric power steering system according to claim 2, characterized in that, The current driving scenario is indicated by Light Mode, Sport Mode, Eco Mode, and Off-Road Mode.

4. The adaptive steering calibration method for an electric power steering system according to claim 1, characterized in that, The flexible actor-critic deep reinforcement learning model was obtained through offline multi-scene training. The training data includes interaction data under different driving scenarios, and the reward function dynamically adjusts the weights according to different scenario labels during the training process to guide the model to learn the optimal steering force strategy under different scenario labels.

5. The adaptive steering calibration method for an electric power steering system according to claim 1, characterized in that, The motion increment includes the boost gain increment, the damping coefficient increment, and the return torque coefficient increment.

6. The adaptive steering calibration method for an electric power steering system according to claim 1, characterized in that, The reward function values ​​include the ideal hand force tracking function, the operational stability function, the safety penalty function, and the energy consumption reward function.

7. An adaptive steering calibration device for an electric power steering system, characterized in that, include: The construction module is used to construct a state space vector containing scene identifiers, wherein the state space vector includes real-time vehicle operating status information, driver input information, and current driving scene identifiers. The generation module is used to input the state space vector into a pre-trained flexible actor-critic deep reinforcement learning model to generate the action increment that adjusts the hand force parameters of the electric power steering system. The update module is used to update the current hand force parameters in real time according to the action increment, so as to change the actual steering hand force of the vehicle using the updated hand force parameters; The calculation module is used to obtain vehicle feedback data and current scene identifier when the actual steering force is applied, so as to calculate the scene-adaptive reward function value; The fine-tuning module is used to store the state space vector, the action increment, the reward function value, and the vehicle feedback data as an experience data set in the experience playback buffer to perform online fine-tuning of the flexible actor-critic deep reinforcement learning model.

8. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the adaptive steering calibration method for an electric power steering system as described in any one of claims 1-6.

9. A computer program product, characterized in that, When the computer program / instruction is executed by the processor, it implements the adaptive steering calibration method for the electric power steering system according to any one of claims 1-6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the adaptive steering calibration method for the electric power steering system as described in any one of claims 1-6.

Images