A personalized vehicle control sequence determination method, device and equipment

Abstract

This application discloses a method, apparatus, and device for determining personalized vehicle control sequences, relating to the field of vehicle control technology. The method includes: extracting and embedding time-series features from current driving behavior data and current vehicle state data based on a target style preference prediction model to obtain a personalized feature vector; combining multi-source driving data with time alignment and channel concatenation based on a diffusion model to obtain a multi-source conditional feature tensor, then performing conditional encoding and back-diffusion denoising to obtain the vehicle's personalized control latent variables, thereby determining the initial vehicle control sequence; and performing vehicle dynamics feasibility verification and correction processing according to vehicle dynamics limit constraint rules to obtain the target vehicle control sequence within a preset future time domain. This scheme, by determining the personalized feature vector through a target style preference prediction model and then determining the target vehicle control sequence based on a diffusion model, improves the personalization and comfort of vehicle control.

Description

Technical Field

[0001] This application relates to the field of vehicle control technology, specifically to a method, apparatus, and device for determining a personalized vehicle control sequence. Background Technology

[0002] With the rapid development of intelligent driving technology, vehicle control systems are evolving from standardization to personalization.

[0003] Existing technologies mostly focus on single-dimensional driving style classification or simple trajectory prediction, lacking a detailed model of continuous quantitative representation of driving style and multi-dimensional comfort preferences, and cannot adapt to individual differences among different drivers in terms of driving style, comfort preferences, etc. Summary of the Invention

[0004] This application provides a method, apparatus, device, and storage medium for determining personalized vehicle control sequences to improve the personalization and comfort of vehicle control.

[0005] According to one aspect of this application, a method for determining a personalized vehicle control sequence is provided, the method comprising: Based on the trained target style preference prediction model, temporal feature extraction and embedding encoding are performed on the current driving behavior data and the current vehicle state data to obtain a personalized feature vector; the personalized feature vector is used to represent driving style preference. Based on the trained diffusion model, the current environmental perception data, current navigation intent data, current vehicle state data and personalized feature vector are time-aligned and channel-separated to obtain a multi-source conditional feature tensor. The multi-source conditional feature tensor is then subjected to conditional encoding and back-diffusion denoising to obtain the vehicle personalized control latent variables. Based on the diffusion model, the hidden variables of the vehicle personalized control are decoded and mapped to obtain the initial vehicle control sequence in the future preset time domain. Based on the vehicle dynamics limit constraint rules, the initial vehicle control sequence is subjected to vehicle dynamics feasibility verification and correction processing to obtain the target vehicle control sequence in the future preset time domain.

[0006] According to another aspect of this application, a personalized vehicle control sequence determination device is provided, the device comprising: The personalized feature vector determination module is used to extract and embed the temporal features of the current driving behavior data and the current vehicle state data based on the trained target style preference prediction model to obtain a personalized feature vector; the personalized feature vector is used to represent driving style preference. The vehicle personalized control latent variable determination module is used to perform time alignment and channel concatenation on the current environment perception data, current navigation intent data, current vehicle state data and personalized feature vector based on the trained diffusion model to obtain a multi-source conditional feature tensor, and to perform conditional encoding and back diffusion denoising on the multi-source conditional feature tensor to obtain the vehicle personalized control latent variables. The initial vehicle control sequence determination module is used to decode and map the vehicle personalized control latent variables based on the diffusion model to obtain the initial vehicle control sequence in the future preset time domain. The target vehicle control sequence determination module is used to perform vehicle dynamics feasibility verification and correction processing on the initial vehicle control sequence according to the vehicle dynamics limit constraint rules, so as to obtain the target vehicle control sequence in the future preset time domain.

[0007] According to another aspect of this application, an electronic device is provided, the electronic device comprising: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement any of the personalized vehicle control sequence determination methods provided in the embodiments of this application.

[0008] According to another aspect of this application, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements any of the personalized vehicle control sequence determination methods provided in the embodiments of this application.

[0009] According to another aspect of this application, a computer program product is provided, including a computer program that, when executed by a processor, implements any of the personalized vehicle control sequence determination methods provided in the embodiments of this application.

[0010] This application uses a trained target style preference prediction model to extract and embed features from current driving behavior data and current vehicle state data to obtain personalized feature vectors. These personalized feature vectors represent driving style preferences. Based on a trained diffusion model, current environmental perception data, current navigation intent data, current vehicle state data, and personalized feature vectors are time-aligned and concatenated to obtain multi-source conditional feature tensors. These multi-source conditional feature tensors are then conditionally encoded and subjected to backdiffusion denoising to obtain vehicle personalized control latent variables. Based on the diffusion model, these vehicle personalized control latent variables are decoded and mapped to obtain an initial vehicle control sequence within a future preset time domain. According to vehicle dynamics limit constraint rules, the initial vehicle control sequence is subjected to vehicle dynamics feasibility verification and correction to obtain the target vehicle control sequence within the future preset time domain. The above scheme determines personalized feature vectors through a target style preference prediction model, transforming abstract driving style preferences into high-dimensional continuous mathematical feature vectors to capture the subtle operating habits of drivers in their subconscious. Then, based on a diffusion model combined with multi-source driving data, it determines the initial vehicle control sequence within a future preset time domain. Subsequently, according to vehicle dynamics limit constraint rules, it performs vehicle dynamics feasibility verification and correction processing to obtain the target vehicle control sequence within the future preset time domain. This achieves a driving experience upgrade from "one-size-fits-all" to "one-person-one-policy," while satisfying vehicle dynamics characteristics and road safety boundaries, thus improving the personalization, comfort, and safety of vehicle control. Attached Figure Description

[0011] Figure 1 This is a flowchart of a personalized vehicle control sequence determination method provided in Embodiment 1 of this application; Figure 2 This is a flowchart of a personalized vehicle control sequence determination method according to Embodiment 2 of this application; Figure 3 This is a schematic diagram of a personalized vehicle control sequence determination device according to Embodiment 3 of this application; Figure 4 This is a schematic diagram of the structure of an electronic device that implements the personalized vehicle control sequence determination method of Embodiment 4 of this application. Detailed Implementation

[0012] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0013] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0014] Furthermore, it should be noted that the information collected in the technical solution of this application is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data all comply with the relevant laws, regulations and standards of the relevant countries and regions, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding operation portals are provided for users to choose to authorize or refuse.

[0015] Example 1 Figure 1 This is a flowchart of a personalized vehicle control sequence determination method according to Embodiment 1 of this application. This embodiment is applicable to situations where a safe and highly customized vehicle control command sequence for the future time domain is generated based on the driver's personalized style preferences. It can be executed by a personalized vehicle control sequence determination device, which can be implemented in hardware and / or software and can be configured in a computer device, such as a server. Figure 1 As shown, the method includes: S110. Based on the trained target style preference prediction model, perform temporal feature extraction and embedding encoding on the current driving behavior data and the current vehicle state data to obtain a personalized feature vector; the personalized feature vector is used to represent driving style preference.

[0016] The target style preference prediction model is used to predict the driver's driving style and comfort preference. This model can employ a deep learning network architecture. Specifically, in this embodiment, the target style preference prediction model primarily determines personalized feature vectors, transforming subjective driving style and comfort preference into quantifiable and inputtable objective feature vectors. Current driving behavior data can be collected through vehicle sensors (such as steering wheel angle sensors, accelerator or brake pedal travel sensors, etc.) and may include current steering wheel angle, current accelerator opening, current brake pedal travel, current operating frequency, and current operating rhythm. Current vehicle state data can be collected using onboard bus, inertial measurement unit, and other devices. This data may include current vehicle speed, current yaw rate, current longitudinal acceleration, current lateral acceleration, current tire adhesion, and current vehicle posture. The personalized feature vector is a concatenated representation containing driving style sub-vectors and comfort preference sub-vectors, used as conditional input for the subsequent diffusion model. The dimension of the personalized feature vector can be determined based on the granularity of the driving style sub-vectors and the number of comfort preference dimensions.

[0017] Optionally, the training process of the target style preference prediction model includes: acquiring sample driving data; the sample driving data includes sample driving behavior data, sample vehicle state data, actual driving style, and actual comfort preference; performing temporal feature extraction and feature concatenation on the sample driving behavior data and the sample vehicle state data to obtain sample fusion temporal features; inputting the sample fusion temporal features into the shared coding layer in the initial style preference prediction model for nonlinear transformation and dimensionality compression to obtain the initial sample embedding vector; inputting the initial sample embedding vector into the driving style classification branch and comfort preference regression branch in the initial style preference prediction model respectively to obtain driving style prediction sub-vectors and comfort preference prediction sub-vectors; training the initial style preference prediction model based on the driving style prediction sub-vectors, the comfort preference prediction sub-vectors, the actual driving style, and the actual comfort preference to obtain the target style preference prediction model.

[0018] The sample driving behavior data can be collected from vehicle sensors (such as steering wheel angle sensors, accelerator or brake pedal travel sensors) under diverse road conditions (such as urban roads, highways, continuous curves, bumpy roads, etc.), including steering wheel angle, accelerator opening, brake pedal travel, operation frequency and rhythm, etc. Sample vehicle state data can be collected synchronously using onboard bus, inertial measurement unit, and other equipment to collect vehicle dynamic response data, including real-time vehicle speed, yaw rate, longitudinal / lateral acceleration, tire adhesion, vehicle attitude, etc. In addition, the collected raw data can be cleaned (outliers and noise removed), time synchronized (ensuring timestamp alignment of behavior, state, and environmental data), and normalized to form a standardized training dataset.

[0019] Specifically, based on the encoding layer and feature fusion layer in the initial style preference prediction model, sample fusion temporal features can be obtained. These features are then input into the shared encoding layer of the initial style preference prediction model. The shared encoding layer, composed of multiple fully connected sub-layers and nonlinear activation functions, performs nonlinear transformation and dimensionality compression on the input features, mapping the high-dimensional temporal features into low-dimensional, compact initial sample embedding vectors. These initial embedding vectors are then input into the parallel driving style classification branch and comfort preference regression branch, respectively, to obtain driving style prediction sub-vectors and comfort preference prediction sub-vectors. Based on these sub-vectors, the style classification loss and comfort regression loss are then determined. Finally, the initial style preference prediction model is trained using the weighted sum of the style classification loss and comfort regression loss to obtain the target style preference prediction model.

[0020] S120. Based on the trained diffusion model, the current environment perception data, current navigation intent data, current vehicle state data and personalized feature vector are time-aligned and channel-concatenated to obtain a multi-source conditional feature tensor. The multi-source conditional feature tensor is then subjected to conditional encoding and back-diffusion denoising to obtain the vehicle personalized control latent variables.

[0021] The trained diffusion model can be a conditional denoising diffusion probability model trained offline. It learns the data distribution of vehicle control command sequences during the forward denoising process and generates control command sequences conforming to the data distribution based on personalized conditional vectors during the backward denoising process. During generation, the diffusion model learns the temporal dependencies and dynamic constraints of driving actions, ensuring the smoothness and temporal coherence of the control sequence. Current environmental perception data can be acquired through devices such as high-precision maps, LiDAR, cameras, and millimeter-wave radar, and may include current road curvature, current heading angle, current lane position, current relative distance to obstacles, current relative speed of obstacles, current obstacle type, and current road surface smoothness. Current navigation intent data may include target lane information, remaining driving distance, lane change request indicators, turning request indicators, and speed limit information. Time alignment synchronizes data using a unified timestamp, eliminating time deviations caused by sensor acquisition delays, transmission delays, and processing delays. Channel stitching stacks the time-aligned multi-source data along the feature dimension to form a high-dimensional joint feature representation. Multi-source conditional feature tensors are high-dimensional, dense representations of heterogeneous information, serving as direct inputs to the conditional coding layer in the diffusion model. Vehicle personalized control latent variables refer to the structured, compressed representations of future vehicle control command sequences formed in the latent space after back-diffusion denoising processing using the diffusion model.

[0022] Optionally, conditional encoding and backdiffusion denoising are performed on the multi-source conditional feature tensor to obtain vehicle personalized control latent variables, including: inputting the multi-source conditional feature tensor into the conditional encoding layer in the diffusion model to obtain a personalized conditional vector; and performing condition-guided backdiffusion denoising on the initial noise vector sampled from the Gaussian distribution based on the personalized conditional vector to obtain vehicle personalized control latent variables.

[0023] The conditional coding layer is used to compress and map high-dimensional, heterogeneous, multi-source conditional feature tensors into low-dimensional, compact, personalized conditional vectors. These personalized conditional vectors guide the denoising direction during backdiffusion, ensuring the final generated control sequence conforms to specific scenarios and personalized requirements. The initial noise vector refers to the latent space noise tensor randomly sampled from a standard Gaussian distribution at the start of the backdiffusion process of the diffusion model; its dimension is consistent with the latent space representation dimension of the target vehicle control sequence.

[0024] S130. Based on the diffusion model, the hidden variables of the vehicle personalized control are decoded and mapped to obtain the initial vehicle control sequence in the future preset time domain.

[0025] Optionally, the step of decoding and mapping the vehicle personalized control latent variables based on the diffusion model to obtain the initial vehicle control sequence in the future preset time domain includes: inputting the vehicle personalized control latent variables into the decoding layer in the diffusion model for nonlinear mapping to obtain the initial vehicle control sequence in the future preset time domain; wherein, the initial vehicle control sequence includes at least a steering angle time sequence, a longitudinal force control sequence, a torque control sequence, a desired speed sequence, a desired acceleration sequence, and a matching relationship sequence between the vehicle yaw response and the steering wheel angle.

[0026] The preset time domain can be manually set according to actual conditions or experience. This application embodiment does not specifically limit this; for example, the preset time domain can be 3 to 5 seconds. The steering angle timing sequence refers to the discrete sampling sequence of the vehicle's front wheel steering angle changing over time within the preset time domain. The longitudinal force control sequence refers to the discrete sampling sequence of the control force acting on the vehicle's longitudinal direction changing over time within the preset time domain. The longitudinal force may include driving force and braking force, used to adjust the vehicle's acceleration and deceleration states. The torque control sequence refers to the discrete sampling sequence of the powertrain system's output torque changing over time within the preset time domain. The torque may include drive motor torque or engine output torque, used to achieve precise execution of longitudinal power requests. The desired speed sequence refers to the discrete sampling sequence of the vehicle's target driving speed changing over time within the preset time domain, representing the speed reference trajectory of the longitudinal motion plan, serving as the target tracking benchmark for longitudinal control. The desired acceleration sequence refers to the discrete sampling sequence of the vehicle's target acceleration changing over time within the preset time domain, representing the acceleration reference trajectory of the longitudinal motion plan, used for feedforward control and ride comfort optimization. The matching sequence of vehicle yaw response and steering wheel angle refers to the time-varying parameter sequence of the dynamic mapping relationship between vehicle yaw rate and steering wheel angle within a future preset time domain. It characterizes the personalized correlation between steering input and vehicle lateral dynamic response and is used to define the steering sensitivity and response hysteresis characteristics preferred by the driver.

[0027] S140. Based on the vehicle dynamics limit constraint rules, perform vehicle dynamics feasibility verification and correction on the initial vehicle control sequence to obtain the target vehicle control sequence in the future preset time domain.

[0028] The vehicle dynamics limit constraints may include tire slip angle limit constraints, longitudinal acceleration limit constraints, and yaw stability constraints. For example, based on the tire adhesion and road adhesion coefficient in the current vehicle state data, the tire slip angle limit value under the current operating condition can be calculated, and the slip angle feasibility can be verified on the steering angle time series, as well as the amplitude clipping or smooth transition correction can be performed.

[0029] In one optional implementation, actual performance data of vehicle actuators and user feedback operation data can also be collected in real time; wherein, the actual performance data includes at least actual steering angle data, actual yaw response data, actual longitudinal force data, actual torque data, actual speed data, and actual acceleration data, and the feedback operation data includes at least user-manually corrected steering wheel angle data and adjusted throttle opening data; based on the deviation between the actual performance data and the vehicle control sequence, and the deviation between the feedback operation data and the driving behavior data, the target style preference prediction model and the diffusion model are updated.

[0030] Specifically, during vehicle operation, the actual performance of the actuators (such as actual steering angle and actual yaw response) and the driver's feedback operations (such as the driver manually correcting the steering wheel, adjusting the throttle or brake) are collected in real time. This data is then fed back to the target style preference prediction model and the diffusion model for model updates, continuously improving the personalization accuracy of vehicle control and the driving experience.

[0031] This application uses a trained target style preference prediction model to extract and embed features from current driving behavior data and current vehicle state data to obtain personalized feature vectors. These personalized feature vectors represent driving style preferences. Based on a trained diffusion model, current environmental perception data, current navigation intent data, current vehicle state data, and personalized feature vectors are time-aligned and concatenated to obtain multi-source conditional feature tensors. These multi-source conditional feature tensors are then conditionally encoded and subjected to backdiffusion denoising to obtain vehicle personalized control latent variables. Based on the diffusion model, these vehicle personalized control latent variables are decoded and mapped to obtain an initial vehicle control sequence within a future preset time domain. According to vehicle dynamics limit constraint rules, the initial vehicle control sequence is subjected to vehicle dynamics feasibility verification and correction to obtain the target vehicle control sequence within the future preset time domain. The above scheme determines personalized feature vectors through a target style preference prediction model, transforming abstract driving style preferences into high-dimensional continuous mathematical feature vectors to capture the subtle operating habits of drivers in their subconscious. Then, based on a diffusion model combined with multi-source driving data, it determines the initial vehicle control sequence within a future preset time domain. Subsequently, according to vehicle dynamics limit constraint rules, it performs vehicle dynamics feasibility verification and correction processing to obtain the target vehicle control sequence within the future preset time domain. This achieves a driving experience upgrade from "one-size-fits-all" to "one-person-one-policy," while satisfying vehicle dynamics characteristics and road safety boundaries, thus improving the personalization, comfort, and safety of vehicle control.

[0032] Example 2 Figure 2This is a flowchart of a personalized vehicle control sequence determination method according to Embodiment 2 of this application. Based on the technical solutions of the above embodiments, this embodiment refines the process of "extracting and embedding temporal features from current driving behavior data and current vehicle state data based on a trained target style preference prediction model to obtain a personalized feature vector" into "extracting temporal features of operating habits from current driving behavior data to obtain a driving operation feature sequence; extracting vehicle dynamic response features from current vehicle state data to obtain a vehicle dynamics feature sequence; concatenating the driving operation feature sequence and the vehicle dynamics feature sequence along the time dimension to obtain a fused temporal feature; inputting the fused temporal feature into the shared encoding layer of the trained target style preference prediction model for nonlinear transformation and dimensionality compression to obtain an initial embedding vector; inputting the initial embedding vector into the driving style classification branch and comfort preference regression branch of the target style preference prediction model to obtain a driving style sub-vector and a comfort preference sub-vector; concatenating the driving style sub-vector and the comfort preference sub-vector to obtain a personalized feature vector." It should be noted that for parts not detailed in this embodiment, please refer to the relevant descriptions in other embodiments. Figure 2 As shown, the method includes: S210. Extract the temporal features of the current driving behavior data to obtain the driving operation feature sequence.

[0033] Specifically, the current driving behavior data, such as the current steering wheel angle, current throttle opening, current brake pedal travel, current operation frequency, and current operation rhythm, can be extracted and differentiated using sliding window statistical features. Then, the driving behavior encoding layer in the target style preference prediction model is used to obtain the driving operation feature sequence.

[0034] S220. Extract vehicle dynamic response features from the current vehicle status data to obtain the vehicle dynamic feature sequence.

[0035] Specifically, dynamic relationship modeling can be performed on current vehicle state data such as current vehicle speed, current yaw rate, current longitudinal acceleration, current lateral acceleration, current tire adhesion, and current vehicle posture. Then, through the vehicle state encoding layer in the target style preference prediction model, the vehicle dynamic feature sequence is obtained.

[0036] S230. The driving operation feature sequence and the vehicle dynamics feature sequence are concatenated along the time dimension to obtain fused temporal features.

[0037] Specifically, through the feature fusion layer in the target style preference prediction model, driving operation features and vehicle dynamics features are stacked and fused in the channel dimension, so that the fused temporal features simultaneously contain coupled information of the driver's operating intention and the vehicle's dynamic response.

[0038] S240. The fused temporal features are input into the shared coding layer of the trained target style preference prediction model for nonlinear transformation and dimensionality compression to obtain the initial embedding vector.

[0039] The shared coding layer can be composed of multiple fully connected sub-layers and nonlinear activation functions. Nonlinear transformation introduces nonlinear mapping capability through activation functions, and dimensionality compression maps high-dimensional fused temporal features into low-dimensional compact initial embedding vectors through fully connected layers.

[0040] S250. Input the initial embedding vector into the driving style classification branch and comfort preference regression branch in the target style preference prediction model to obtain driving style sub-vectors and comfort preference sub-vectors.

[0041] Optionally, the driving style sub-vector includes continuous quantified values ​​of aggressive, stable, and conservative driving styles; the comfort preference sub-vector includes continuous quantified values ​​of longitudinal comfort preference, lateral comfort preference, vertical comfort preference, operational interaction comfort preference, and environmental adaptation comfort preference.

[0042] Among them, the continuous quantitative value of aggressive driving style represents the probability that the driver tends to drive with rapid response, large-scale operation, and high dynamic limits. The higher the value, the more the driver prefers an aggressive driving mode. The continuous quantitative value of stable driving style represents the probability that the driver tends to drive with gradual operation, small adjustments, and moderate dynamic response. The higher the value, the more the driver prefers a balanced and stable driving mode. The continuous quantitative value of conservative driving style represents the probability that the driver tends to drive with anticipation, conservative operation, and low dynamic limits. The higher the value, the more the driver prefers a safe and conservative driving mode. The continuous quantitative value of longitudinal comfort preference represents the driver's sensitivity to the smoothness of vehicle acceleration and deceleration, reflecting the tolerance threshold for changes in longitudinal acceleration rate. The higher the value, the more the driver prefers a smooth and shock-free acceleration and deceleration experience. The continuous quantitative value of lateral comfort preference represents the driver's sensitivity to body roll and yaw response during vehicle steering, reflecting the tolerance threshold for changes in lateral acceleration and body roll angle. The higher the value, the more the driver prefers smooth cornering and soft steering. Vertical comfort preference continuous quantification values ​​characterize the driver's sensitivity to road bumps and vertical vibrations, reflecting the tolerance threshold for vertical acceleration amplitude and vibration frequency. Higher values ​​indicate a greater preference for a ride experience with smooth vibration filtering and road feel isolation. Operational interaction comfort preference continuous quantification values ​​characterize the driver's sensitivity to steering effort, pedal feedback effort, and response delay, reflecting a preference for the tactile feedback and immediate response of human-machine interaction. Higher values ​​indicate a greater preference for a light, effortless, and slightly delayed operational feel. Environmental adaptation comfort preference continuous quantification values ​​characterize the driver's preference for dynamic adjustments to comfort strategies under different driving scenarios (such as congestion, highways, curves, and inclement weather), reflecting the adaptive tendency of driving style to change with the environment. Higher values ​​indicate a greater expectation that the system will automatically optimize comfort strategies according to the scenario.

[0043] S260. The driving style sub-vector and the comfort preference sub-vector are concatenated to obtain a personalized feature vector; the personalized feature vector is used to characterize driving style preference.

[0044] S270. Based on the trained diffusion model, the current environmental perception data, current navigation intent data, current vehicle state data, and personalized feature vector are time-aligned and channel-concatenated to obtain a multi-source conditional feature tensor. The multi-source conditional feature tensor is then subjected to conditional encoding and back-diffusion denoising to obtain the vehicle personalized control latent variables.

[0045] S280. Based on the diffusion model, the hidden variables of the vehicle personalized control are decoded and mapped to obtain the initial vehicle control sequence in the future preset time domain.

[0046] S290. Based on the vehicle dynamics limit constraint rules, the initial vehicle control sequence is subjected to vehicle dynamics feasibility verification and correction processing to obtain the target vehicle control sequence in the future preset time domain.

[0047] This application uses a trained target style preference prediction model to extract and embed features from current driving behavior data and current vehicle state data to obtain personalized feature vectors. These personalized feature vectors represent driving style preferences. Based on a trained diffusion model, current environmental perception data, current navigation intent data, current vehicle state data, and personalized feature vectors are time-aligned and concatenated to obtain multi-source conditional feature tensors. These multi-source conditional feature tensors are then conditionally encoded and subjected to backdiffusion denoising to obtain vehicle personalized control latent variables. Based on the diffusion model, these vehicle personalized control latent variables are decoded and mapped to obtain an initial vehicle control sequence within a future preset time domain. According to vehicle dynamics limit constraint rules, the initial vehicle control sequence is subjected to vehicle dynamics feasibility verification and correction to obtain the target vehicle control sequence within the future preset time domain. The above scheme determines personalized feature vectors through a target style preference prediction model, transforming abstract driving style preferences into high-dimensional continuous mathematical feature vectors to capture the subtle operating habits of drivers in their subconscious. Then, based on a diffusion model combined with multi-source driving data, it determines the initial vehicle control sequence within a future preset time domain. Subsequently, according to vehicle dynamics limit constraint rules, it performs vehicle dynamics feasibility verification and correction processing to obtain the target vehicle control sequence within the future preset time domain, which satisfies vehicle dynamics characteristics and road safety boundaries, thereby improving the personalization, comfort, and safety of vehicle control.

[0048] Example 3 Figure 3 This is a schematic diagram of a personalized vehicle control sequence determination device according to Embodiment 3 of this application. This embodiment is applicable to situations where a safe and highly customized vehicle control command sequence is generated in the future time domain based on the driver's personalized style preferences. This personalized vehicle control sequence determination device can be implemented in hardware and / or software, and can be configured in a computer device, such as a server. Figure 3 As shown, the device includes: The personalized feature vector determination module 310 is used to extract and embed the time-series features of the current driving behavior data and the current vehicle state data based on the trained target style preference prediction model to obtain a personalized feature vector; the personalized feature vector is used to characterize driving style preference. The vehicle personalized control latent variable determination module 320 is used to perform time alignment and channel concatenation on the current environment perception data, current navigation intent data, current vehicle state data and personalized feature vector based on the trained diffusion model to obtain a multi-source conditional feature tensor, and to perform conditional encoding and back diffusion denoising on the multi-source conditional feature tensor to obtain the vehicle personalized control latent variables. The initial vehicle control sequence determination module 330 is used to decode and map the vehicle personalized control latent variables based on the diffusion model to obtain the initial vehicle control sequence in the future preset time domain. The target vehicle control sequence determination module 340 is used to perform vehicle dynamics feasibility verification and correction processing on the initial vehicle control sequence according to the vehicle dynamics limit constraint rules, so as to obtain the target vehicle control sequence in the future preset time domain.

[0049] Optional, the personalized feature vector determination module 310 includes: The driving operation feature sequence determination unit is used to extract the temporal features of operating habits from the current driving behavior data to obtain the driving operation feature sequence. The vehicle dynamics feature sequence determination unit is used to extract vehicle dynamic response features from the current vehicle state data to obtain the vehicle dynamics feature sequence. The fusion temporal feature determination unit is used to concatenate the driving operation feature sequence and the vehicle dynamics feature sequence along the time dimension to obtain fused temporal features; The initial embedding vector determination unit is used to input the fused temporal features into the shared coding layer of the trained target style preference prediction model to perform nonlinear transformation and dimensionality compression to obtain the initial embedding vector. The sub-vector determination unit is used to input the initial embedding vector into the driving style classification branch and comfort preference regression branch in the target style preference prediction model, respectively, to obtain driving style sub-vectors and comfort preference sub-vectors; The personalized feature vector determination unit is used to concatenate the driving style sub-vector and the comfort preference sub-vector to obtain a personalized feature vector.

[0050] Optionally, the driving style sub-vector includes continuous quantified values ​​of aggressive, stable, and conservative driving styles; the comfort preference sub-vector includes continuous quantified values ​​of longitudinal comfort preference, lateral comfort preference, vertical comfort preference, operational interaction comfort preference, and environmental adaptation comfort preference.

[0051] Optionally, the training process for the target style preference prediction model includes: Acquire sample driving data; the sample driving data includes sample driving behavior data, sample vehicle status data, actual driving style, and actual comfort preference; Temporal feature extraction and feature concatenation are performed on the sample driving behavior data and the sample vehicle state data to obtain sample fusion temporal features; The sample fusion temporal features are input into the shared coding layer of the initial style preference prediction model for nonlinear transformation and dimensionality compression to obtain the initial embedding vector of the sample. The initial embedding vector of the sample is input into the driving style classification branch and the comfort preference regression branch in the initial style preference prediction model, respectively, to obtain the driving style prediction sub-vector and the comfort preference prediction sub-vector; The initial style preference prediction model is trained based on the driving style prediction sub-vector, the comfort preference prediction sub-vector, the actual driving style, and the actual comfort preference to obtain the target style preference prediction model.

[0052] Optionally, the vehicle personalization control latent variable determination module 320 includes: A personalized condition vector determination unit is used to input the multi-source condition feature tensor into the condition coding layer in the diffusion model to obtain a personalized condition vector. The vehicle personalized control latent variable determination unit is used to perform condition-guided backdiffusion denoising processing on the initial noise vector sampled from the Gaussian distribution based on the personalized condition vector to obtain the vehicle personalized control latent variables.

[0053] Optionally, the initial vehicle control sequence determination module 330 is specifically used for: The hidden variables of the vehicle personalized control are input into the decoding layer of the diffusion model for nonlinear mapping to obtain the initial vehicle control sequence in the future preset time domain. The initial vehicle control sequence includes at least a steering angle timing sequence, a longitudinal force control sequence, a torque control sequence, a desired speed sequence, a desired acceleration sequence, and a matching relationship sequence between the vehicle yaw response and the steering wheel angle.

[0054] Optionally, the device further includes: The model update module is used for: The system collects real-time data on the actual performance of vehicle actuators and user feedback data. The actual performance data includes at least actual steering angle data, actual yaw response data, actual longitudinal force data, actual torque data, actual speed data, and actual acceleration data. The feedback data includes at least data on the user's manual correction of the steering wheel angle and the adjusted throttle opening. The target style preference prediction model and the diffusion model are updated based on the deviation between the actual execution effect data and the vehicle control sequence, and the deviation between the feedback operation data and the driving behavior data.

[0055] This application uses a trained target style preference prediction model to extract and embed features from current driving behavior data and current vehicle state data to obtain personalized feature vectors. These personalized feature vectors represent driving style preferences. Based on a trained diffusion model, current environmental perception data, current navigation intent data, current vehicle state data, and personalized feature vectors are time-aligned and concatenated to obtain multi-source conditional feature tensors. These multi-source conditional feature tensors are then conditionally encoded and subjected to backdiffusion denoising to obtain vehicle personalized control latent variables. Based on the diffusion model, these vehicle personalized control latent variables are decoded and mapped to obtain an initial vehicle control sequence within a future preset time domain. According to vehicle dynamics limit constraint rules, the initial vehicle control sequence is subjected to vehicle dynamics feasibility verification and correction to obtain the target vehicle control sequence within the future preset time domain. The above scheme determines personalized feature vectors through a target style preference prediction model, transforming abstract driving style preferences into high-dimensional continuous mathematical feature vectors to capture the subtle operating habits of drivers in their subconscious. Then, based on a diffusion model combined with multi-source driving data, it determines the initial vehicle control sequence within a future preset time domain. Subsequently, according to vehicle dynamics limit constraint rules, it performs vehicle dynamics feasibility verification and correction processing to obtain the target vehicle control sequence within the future preset time domain. This achieves a driving experience upgrade from "one-size-fits-all" to "one-person-one-policy," while satisfying vehicle dynamics characteristics and road safety boundaries, thus improving the personalization, comfort, and safety of vehicle control.

[0056] The personalized vehicle control sequence determination device provided in this application embodiment can execute the personalized vehicle control sequence determination method provided in any embodiment of this application, and has the corresponding functional modules and beneficial effects for executing each personalized vehicle control sequence determination method.

[0057] According to embodiments of this application, this application also provides an electronic device, a readable storage medium, and a computer program product.

[0058] Example 4 Figure 4This is a schematic diagram of the structure of an electronic device 410 implementing the personalized vehicle control sequence determination method of the embodiments of this application. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present application described and / or claimed herein.

[0059] like Figure 4 As shown, the electronic device 410 includes at least one processor 411 and a memory, such as a read-only memory 412 or a random access memory 413, communicatively connected to the at least one processor 411. The memory stores computer programs executable by the at least one processor. The processor 411 can perform various appropriate actions and processes based on the computer program stored in the read-only memory 412 or loaded from storage unit 418 into the random access memory 413. The random access memory 413 can also store various programs and data required for the operation of the electronic device 410. The processor 411, read-only memory 412, and random access memory 413 are interconnected via a bus 414. An input / output interface 415 is also connected to the bus 414.

[0060] Multiple components in electronic device 410 are connected to input / output interface 415, including: input unit 416, such as keyboard, mouse, etc.; output unit 417, such as various types of monitors, speakers, etc.; storage unit 418, such as disk, optical disk, etc.; and communication unit 419, such as network card, modem, wireless transceiver, etc. Communication unit 419 allows electronic device 410 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0061] Processor 411 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 411 include, but are not limited to, central processing units, graphics processing units, various special-purpose artificial intelligence computing chips, various processors running machine learning model algorithms, digital signal processors, and any suitable processor, controller, microcontroller, etc. Processor 411 performs the various methods and processes described above, such as personalized vehicle control sequence determination methods.

[0062] In some embodiments, the personalized vehicle control sequence determination method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 418. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 410 via read-only memory 412 and / or communication unit 419. When the computer program is loaded into random access memory 413 and executed by processor 411, one or more steps of the personalized vehicle control sequence determination method described above may be performed. Alternatively, in other embodiments, processor 411 may be configured as the personalized vehicle control sequence determination method by any other suitable means (e.g., by means of firmware).

[0063] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays, application-specific integrated circuits (ASICs), application-specific standard products (ASICs), systems-on-a-chip (SoCs), payload programmable logic devices, computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0064] Computer programs used to implement the methods of this application may be written in any combination of one or more programming languages. These computer programs may be provided to the processor of a general-purpose computer, a special-purpose computer, or other programmable personalized vehicle control sequence determination device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0065] In the context of this application, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium can be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory, optical fibers, portable compact disk read-only memory, optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0066] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a cathode ray tube or liquid crystal display monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0067] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0068] A computing system can include requesters and servers. Requesters and servers are generally geographically separated and typically interact via communication networks. The relationship between a requester and a server is established by computer programs running on the respective computers and having a requester-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a host product within the cloud computing service system to address the shortcomings of traditional physical hosts and virtual private servers, such as high management difficulty and weak business scalability.

[0069] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this application can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this application can be achieved, and this is not limited herein.

[0070] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for determining a personalized vehicle control sequence, characterized in that, include: Based on the trained target style preference prediction model, temporal feature extraction and embedding encoding are performed on current driving behavior data and current vehicle state data to obtain personalized feature vectors. The personalized feature vector is used to characterize driving style preferences; Based on the trained diffusion model, the current environmental perception data, current navigation intent data, current vehicle state data and personalized feature vector are time-aligned and channel-separated to obtain a multi-source conditional feature tensor. The multi-source conditional feature tensor is then subjected to conditional encoding and back-diffusion denoising to obtain the vehicle personalized control latent variables. Based on the diffusion model, the hidden variables of the vehicle personalized control are decoded and mapped to obtain the initial vehicle control sequence in the future preset time domain. Based on the vehicle dynamics limit constraint rules, the initial vehicle control sequence is subjected to vehicle dynamics feasibility verification and correction processing to obtain the target vehicle control sequence in the future preset time domain.

2. The method according to claim 1, characterized in that, The trained target style preference prediction model extracts and embeds temporal features from current driving behavior data and current vehicle state data to obtain personalized feature vectors, including: Extract the temporal features of driving habits from the current driving behavior data to obtain a driving operation feature sequence; Vehicle dynamic response features are extracted from the current vehicle status data to obtain a vehicle dynamic feature sequence; The driving operation feature sequence and the vehicle dynamics feature sequence are concatenated along the time dimension to obtain fused temporal features; The fused temporal features are input into the shared coding layer of the trained target style preference prediction model for nonlinear transformation and dimensionality compression to obtain the initial embedding vector. The initial embedding vector is input into the driving style classification branch and comfort preference regression branch of the target style preference prediction model, respectively, to obtain driving style sub-vectors and comfort preference sub-vectors; The driving style subvector and the comfort preference subvector are concatenated to obtain a personalized feature vector.

3. The method according to claim 2, characterized in that, The driving style sub-vector includes continuous quantitative values ​​of aggressive, stable, and conservative driving styles; the comfort preference sub-vector includes continuous quantitative values ​​of longitudinal comfort preference, lateral comfort preference, vertical comfort preference, operational interaction comfort preference, and environmental adaptation comfort preference.

4. The training process of the target style preference prediction model according to any one of claims 1-3 includes: Obtain sample driving data; The sample driving data includes sample driving behavior data, sample vehicle status data, actual driving style, and actual comfort preference; Temporal feature extraction and feature concatenation are performed on the sample driving behavior data and the sample vehicle state data to obtain sample fusion temporal features; The sample fusion temporal features are input into the shared coding layer of the initial style preference prediction model for nonlinear transformation and dimensionality compression to obtain the initial embedding vector of the sample. The initial embedding vector of the sample is input into the driving style classification branch and the comfort preference regression branch in the initial style preference prediction model, respectively, to obtain the driving style prediction sub-vector and the comfort preference prediction sub-vector; The initial style preference prediction model is trained based on the driving style prediction sub-vector, the comfort preference prediction sub-vector, the actual driving style, and the actual comfort preference to obtain the target style preference prediction model.

5. The method according to claim 1, characterized in that, Conditional encoding and backdiffusion denoising are performed on the multi-source conditional feature tensor to obtain the vehicle personalized control latent variables, including: The multi-source conditional feature tensor is input into the conditional encoding layer of the diffusion model to obtain a personalized conditional vector. Based on the personalized condition vector, the initial noise vector sampled from the Gaussian distribution is subjected to condition-guided backdiffusion denoising processing to obtain the hidden variables for personalized vehicle control.

6. The method according to claim 1, characterized in that, The step of decoding and mapping the latent variables of the vehicle personalized control based on the diffusion model to obtain the initial vehicle control sequence in the future preset time domain includes: The hidden variables of the vehicle personalized control are input into the decoding layer of the diffusion model for nonlinear mapping to obtain the initial vehicle control sequence in the future preset time domain. The initial vehicle control sequence includes at least a steering angle timing sequence, a longitudinal force control sequence, a torque control sequence, a desired speed sequence, a desired acceleration sequence, and a matching relationship sequence between the vehicle yaw response and the steering wheel angle.

7. The method according to claim 1, characterized in that, The method further includes: The system collects real-time data on the actual performance of vehicle actuators and user feedback data. The actual performance data includes at least actual steering angle data, actual yaw response data, actual longitudinal force data, actual torque data, actual speed data, and actual acceleration data. The feedback data includes at least data on the user's manual correction of the steering wheel angle and the adjusted throttle opening. The target style preference prediction model and the diffusion model are updated based on the deviation between the actual execution effect data and the vehicle control sequence, and the deviation between the feedback operation data and the driving behavior data.

8. A personalized vehicle control sequence determination device, characterized in that, include: The personalized feature vector determination module is used to extract and embed the temporal features of the current driving behavior data and the current vehicle state data based on the trained target style preference prediction model to obtain personalized feature vectors. The personalized feature vector is used to characterize driving style preferences; The vehicle personalized control latent variable determination module is used to perform time alignment and channel concatenation on the current environment perception data, current navigation intent data, current vehicle state data and personalized feature vector based on the trained diffusion model to obtain a multi-source conditional feature tensor, and to perform conditional encoding and back diffusion denoising on the multi-source conditional feature tensor to obtain the vehicle personalized control latent variables. The initial vehicle control sequence determination module is used to decode and map the vehicle personalized control latent variables based on the diffusion model to obtain the initial vehicle control sequence in the future preset time domain. The target vehicle control sequence determination module is used to perform vehicle dynamics feasibility verification and correction processing on the initial vehicle control sequence according to the vehicle dynamics limit constraint rules, so as to obtain the target vehicle control sequence in the future preset time domain.

9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the personalized vehicle control sequence determination method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the personalized vehicle control sequence determination method as described in any one of claims 1-7.

Images