An occupant personalized preference driven intelligent chassis control method, system, device and storage medium based on a vision-language-action model

Abstract

This invention discloses a method, system, device, and storage medium for intelligent chassis control driven by occupant personalized preferences based on a vision-language-action model, belonging to the field of intelligent vehicle control and intelligent chassis collaborative control technology. The method includes: collecting multi-source information; constructing a personalized dynamic experience preference profile of the occupant; performing scene-preference multimodal joint modeling to form a unified scene-preference semantic representation; inputting the vision-language-action model to generate personalized chassis style intent; generating dynamic performance constraints and controller parameter adjustment targets; generating specific chassis control targets; executing multi-actuator personalized collaborative control; and performing parameter recovery and preference learning updates. This method realizes vision-language-action joint reasoning between occupant personalized preferences, visual scene semantics, and chassis control actions, improving the overall performance of autonomous and assisted driving vehicles in terms of comfort, stability, safety, and personalized experience.

Description

Technical Field

[0001] This invention belongs to the field of intelligent vehicle control and intelligent chassis collaborative control technology, specifically relating to an intelligent chassis control method, system, device and storage medium driven by occupant personalized preferences based on a vision-language-action model. Background Technology

[0002] With the rapid development of autonomous driving technology, smart cockpit technology, and software-defined vehicle technology, vehicle control systems are gradually evolving from a traditional control mode centered on vehicle dynamics to a control mode that takes into account both vehicle performance and passenger experience. Modern intelligent chassis typically include multiple chassis systems such as steering system, braking system, drive system, active suspension system, and yaw stability control system. There are significant coupling relationships between these subsystems in terms of handling stability, ride comfort, safety, and ride smoothness.

[0003] Existing chassis control technologies typically adjust vehicle dynamics through fixed parameters or preset driving modes. For example, some vehicles offer preset modes such as Comfort, Sport, and Eco, altering vehicle dynamics by switching parameters like suspension damping, steering assist, and power response. However, these modes are limited in variety, have discrete parameters, and lack expressive power, making it difficult to meet the fine-grained, personalized needs of different occupants in various driving scenarios. On the other hand, with the development of in-vehicle voice interaction and natural language understanding technologies, vehicles can now control navigation, air conditioning, entertainment systems, and some functions via voice commands. However, most existing voice control systems remain at the functional command execution level, such as turning on the air conditioning or switching to Comfort mode. They cannot understand more complex dynamic experience requests, nor can they directly map such requests to vehicle chassis dynamic performance indicators and controller parameters.

[0004] In actual driving, occupants often express language preferences such as: "The road is slippery today, be more cautious"; "Go over speed bumps gently"; "Don't be too aggressive when changing lanes at high speeds"; "Someone is sleeping in the back, minimize body roll and head-down movement"; "Mountain roads have many curves, stability is paramount." This kind of language is not directly equivalent to low-level control commands, but rather a high-level semantic preference description of the vehicle's dynamic behavior. Existing systems typically struggle to integrate these natural language preferences with the current visual scene, road environment, and vehicle status, making it even more difficult to generate personalized chassis control strategies for steering, braking, drive, suspension, and stability control systems. Furthermore, relying solely on language for parameter mapping approximations is closer to a simple language model or a vision-language model, failing to reflect the joint reasoning between vision, language, and control actions. For a real vehicle, occupants' expressions of "be more cautious," "go more cautiously," and "don't be too aggressive" can only be translated into specific, reasonable, and executable chassis control actions when combined with the current road scene, vehicle status, and occupant status. Summary of the Invention

[0005] To address the issues mentioned in the background section regarding existing chassis control technologies, such as the lack of structured expression capabilities for personalized dynamic experience preferences of occupants, the lack of joint reasoning capabilities between occupant preferences and visual scene semantics, the inability to map personalized preferences to chassis dynamic performance indicators and controller parameters, and the lack of intelligent chassis control mechanisms oriented towards individual differences, this invention proposes an intelligent chassis control method, system, device, and storage medium based on a vision-language-action model driven by personalized occupant preferences, in order to achieve occupant preference modeling, scene-related style reasoning, and adaptive adjustment of the chassis controller.

[0006] Technical Solution: To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0007] A method for intelligent chassis control driven by occupant personalized preferences based on a vision-language-action model includes the following steps:

[0008] S1. Collect occupant language preference input, vehicle environment image, navigation semantic information, vehicle status information, chassis status information, and occupant status information;

[0009] S2. Construct or update a personalized dynamic experience preference profile of passengers based on their natural language preference input;

[0010] S3. Encode and fuse the personalized dynamic experience preference profile of the occupants, vehicle environment image, navigation semantic information, vehicle status information, chassis status information and occupant status information to obtain a unified scene-preference semantic representation.

[0011] S4. Input the unified scene-preference semantic representation into the vision-language-action model to generate scene-related personalized chassis style intent;

[0012] S5. Based on the personalized chassis style intent, vehicle status information, chassis status information, and current scene semantic information, generate chassis dynamic performance index constraints and controller parameter adjustment targets.

[0013] S6. Based on the chassis dynamic performance index constraints and controller parameter adjustment targets, perform parameter tuning, control constraint update, response characteristic switching, and control mode switching for each chassis system.

[0014] S7. Perform personalized active chassis control based on updated chassis parameters during vehicle operation;

[0015] S8. Based on passenger feedback, vehicle response, and scene changes, restore, correct, or update the passenger's personalized dynamic experience preference profile and personalized control template.

[0016] As a preferred option, the specific implementation details in S1 are as follows:

[0017] The multi-source information acquisition module is used to obtain passenger language preference input during vehicle operation. Environmental images Navigation semantic information Vehicle status information Chassis status information Crew status information ;

[0018] No. The multi-source information collected at different times is denoted as follows:

[0019] .

[0020] As a preferred option, the specific implementation details in S2 are as follows:

[0021] Personalized dynamic experience profile of passengers Defined as:

[0022] ;

[0023] in, Indicates the stability preference weight; Indicates the weight of comfort preference; Indicates the smoothness preference weight; Indicates the degree of radicalism in preference weights; Indicates the weight of safety preference; Indicates the passivity preference weight; Represents the transpose of a matrix;

[0024] Preference profiles are updated recursively. The current language input is first converted into an instantaneous preference vector by a language semantic parser. Then, it is integrated and updated with historical preference profiles, specifically as follows:

[0025] ;

[0026] in, This represents the immediate preference vector obtained by parsing the current language input and feedback information. Indicates the preference update coefficient; This represents the personalized dynamic experience preference profile of the occupants at time k+1; This represents a personalized dynamic experience preference profile of passengers.

[0027] As a preferred option, the specific implementation details in S3 are as follows:

[0028] The overall representation of the multimodal fusion coding process is as follows:

[0029] ;

[0030] in, This represents the multimodal fusion coding function. Indicates the first A unified scene-preference semantic representation at any given moment; This represents a personalized dynamic experience preference profile of passengers; Represents an environmental image; Represents navigation semantic information; Indicates vehicle status information; Indicates chassis status information; This indicates the occupant status information.

[0031] As a preferred option, the specific implementation details in S4 are as follows:

[0032] Using a visual-language-action personalized style reasoning module to represent the semantics of a unified scene and preference Perform semantic reasoning to output personalized chassis style intent, the process of which is represented as follows:

[0033] ;

[0034] in, Represents the visual-language-action model. This indicates an intention to personalize the chassis style; Indicates the first A unified scene-preference semantic representation for each moment.

[0035] As a preferred option, the specific implementation details in S5 are as follows:

[0036] S501: First, define the chassis dynamic performance index constraint vector. ;

[0037] S502: Style intent mapping of chassis dynamic performance index constraint vectors;

[0038] S503: Define the controller parameter adjustment target and provide a unified representation of the generation process of the chassis dynamic performance index constraint vector and the controller parameter adjustment target.

[0039] A vision-language-action model-based occupant personalized preference-driven intelligent chassis control system, implementing the vision-language-action model-based occupant personalized preference-driven intelligent chassis control method described above, the system comprising:

[0040] The multi-source information acquisition module is used to acquire occupant natural language preference input, vehicle environment images, navigation semantic information, vehicle status information, chassis status information, and occupant status information;

[0041] The passenger preference profile management module is used to build or update personalized dynamic experience preference profiles of passengers based on their natural language preference input.

[0042] The multimodal fusion coding module is used to fuse and encode the personalized dynamic experience preference profile of the occupants, vehicle environment image, navigation semantic information, vehicle status information, chassis status information and occupant status information to generate a unified scene-preference semantic representation.

[0043] The visual-language-action personalized style reasoning module is used to generate scene-related personalized chassis style intents based on a unified scene-preference semantic representation.

[0044] The dynamic performance index and controller parameter generation module is used to generate chassis dynamic performance index constraints and controller parameter adjustment targets based on personalized chassis style intent, vehicle operating status information, chassis status information and current scene semantic information.

[0045] A personalized control template library and a control target generation module are used to generate personalized chassis control targets based on personalized chassis style intentions.

[0046] The chassis personalization controller module is used to perform parameter tuning, control constraint update, response characteristic switching or control mode switching for the steering system, braking system, drive system, active suspension system and / or yaw stability control system according to the chassis dynamic performance index constraints and controller parameter adjustment targets, and to execute personalized chassis active control.

[0047] The feedback learning update module is used to restore, correct, or update the personalized dynamic experience preference profile and personalized control template of passengers based on passenger feedback, vehicle response, and changes in the scenario.

[0048] An electronic device includes a processor, a memory, and a computer program stored in the memory, wherein the processor executes the computer program to implement the method described in any of the preceding items.

[0049] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any of the preceding claims.

[0050] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0051] (1) This invention focuses on the personalized dynamic experience preferences of passengers, rather than on road events or fixed driving modes, and can significantly enhance the ability of chassis control to express individual differences.

[0052] (2) The present invention utilizes a vision-language-action model to achieve joint reasoning of occupant preferences, visual scenes and vehicle status, and can provide different and reasonable control interpretations for the same preference in different scenarios.

[0053] (3) The present invention maps personalized preferences to chassis dynamic performance index constraints and controller parameters, so that passenger language preferences can be truly incorporated into the chassis controller structure layer.

[0054] (4) This invention supports multi-actuator collaborative personalized adjustment of steering, braking, driving, suspension and stability control systems, which improves the system-level control effect.

[0055] (5) This invention achieves long-term user style memory and adaptive optimization of control strategies through preference profiling, personalized template library and feedback learning mechanism, and has good product application value. Attached Figure Description

[0056] Figure 1 This is a diagram illustrating the overall architecture of the method of the present invention;

[0057] Figure 2 This is a flowchart of the method of the present invention;

[0058] Figure 3 This is a schematic diagram of chassis pre-tuning based on VLA. Detailed Implementation

[0059] The present invention will be further illustrated below with reference to specific embodiments. These embodiments are implemented based on the technical solutions of the present invention, and it should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention.

[0060] Example 1

[0061] This embodiment provides a personalized intelligent chassis control method driven by occupant preferences based on a vision-language-action model. Specifically, it involves a personalized intelligent chassis control technology that performs multimodal joint semantic reasoning on occupant natural language preferences, vehicle external visual scenes, navigation semantics, vehicle operating status, chassis status, and occupant status to generate a structured description of occupant dynamic experience preferences and chassis control intentions, and maps them to chassis dynamic performance index constraints, controller parameters, and control modes.

[0062] like Figure 1 , Figure 2 As shown, the method flow provided in this embodiment is as follows.

[0063] S1. Collect occupant language preference input, vehicle environment image, navigation semantic information, vehicle status information, chassis status information, and occupant status information;

[0064] The multi-source information acquisition module is used to obtain passenger language preference input during vehicle operation. Environmental images Navigation semantic information Vehicle status information Chassis status information Crew status information .

[0065] In this embodiment, the first The multi-source information collected at different times is denoted as follows:

[0066]

[0067] Passenger language preference input This is used to characterize the subjective needs of occupants regarding the vehicle's dynamic experience, and can be collected through in-vehicle voice recognition units, central control interaction units, or mobile terminal interaction units. Typical inputs include, but are not limited to: "The road is slippery today, so be careful," "Go over speed bumps gently," "Don't be too aggressive when changing lanes on the highway," "Someone is sleeping in the back, so minimize body roll and head-down motion," "Many curves, so prioritize stability," "Smoother ride at low speeds in the city," etc.

[0068] Vehicle environment images It can be provided by forward-looking cameras, surround-view cameras, depth cameras, or their fusion perception results, and is used to characterize the current road environment, surrounding obstacles, road geometry, road surface undulations, and traffic flow status.

[0069] Navigation semantic information Used to characterize the current road task and context, it may include road type, lane change information, curved road sections, ramp information, speed bump areas, construction areas, speed limit information, and task priority.

[0070] Vehicle status information The current vehicle dynamics state can be represented as:

[0071]

[0072] in, For longitudinal velocity, For lateral velocity, The yaw rate is angular velocity. and These are longitudinal and lateral accelerations, respectively. Side slip angle, The steering angle of the front wheels; This represents the transpose of a matrix.

[0073] Chassis status information It can be represented as:

[0074]

[0075] in, These are the suspension damping parameters. These are the suspension stiffness parameters. For braking response gain, For steering assist parameters, To drive the response smoothing coefficient, To stabilize the control threshold parameter; This represents the transpose of a matrix.

[0076] Crew status information It can be represented as:

[0077]

[0078] in, Indicates the sleeping status of rear passengers. Indicates the occupant's posture or state. Indicates occupant dynamic sensitivity label. Labels indicating a preference for comfort; This represents the transpose of a matrix.

[0079] In this embodiment, the occupant status information includes at least one of the following: rear occupant sleep status, occupant posture status, occupant dynamic sensitivity label, and comfort priority requirement label.

[0080] S2. Construct or update a personalized dynamic experience preference profile of passengers based on their natural language preference input;

[0081] The passenger preference profile management module is used to build or update personalized dynamic experience preference profiles of passengers based on current language input and historical preference information. These profiles are used to characterize the long-term preference characteristics of different passengers in terms of stability, comfort, smoothness, aggressiveness, safety, and passability.

[0082] The personalized dynamic experience preference profile of occupants includes at least one of the following: stability preference weight, comfort preference weight, smoothness preference weight, aggression preference weight, safety preference weight, and passability preference weight.

[0083] In this embodiment, a personalized dynamic experience preference profile of the passenger is generated. Defined as:

[0084]

[0085] in, Indicates the stability preference weight; Indicates the weight of comfort preference; Indicates the smoothness preference weight; Indicates the degree of radicalism in preference weights; Indicates the weight of safety preference; Indicates the passivity preference weight; This represents the transpose of a matrix.

[0086] The personalized dynamic experience preference profile of passengers consists of an immediate preference component and a long-term preference component. The immediate preference component is obtained by parsing the current passenger's natural language preference input, while the long-term preference component is formed by at least one of the following: historical interaction records, historical control selection records, passenger feedback records, and scenario usage habits.

[0087] In this embodiment, the preference weights are located within the interval [0,1] and satisfy the normalization constraint, specifically:

[0088]

[0089] in, express The vector element in the vector represents any one of the following: stability preference, comfort preference, smoothness preference, aggressiveness preference, safety preference, and passability preference.

[0090] Preference profiles are updated recursively. The current language input is first converted into an instantaneous preference vector by a language semantic parser. Then, it is integrated and updated with historical preference profiles, specifically as follows:

[0091]

[0092] in, This represents the immediate preference vector obtained by parsing the current language input and feedback information. , Indicates the preference update coefficient; when A larger value indicates that the system places more emphasis on the current language preference; when... A lower value indicates that the system places greater emphasis on long-term preference memory. For example, if an occupant has a long-standing preference for comfort and smoothness, and the current input is "The road is slippery today, so be careful," the system will not completely discard the original comfort and smoothness preference. Instead, it will moderately enhance the stability preference based on the current scenario, so that personalized control reflects both temporary needs and long-term style characteristics. A personalized dynamic experience preference profile of the occupants at time k+1; This represents a personalized dynamic experience preference profile of passengers.

[0093] S3. Encode and fuse the personalized dynamic experience preference profile of the occupants, vehicle environment image, navigation semantic information, vehicle status information, chassis status information and occupant status information to obtain a unified scene-preference semantic representation, that is, form a scene-preference joint semantic representation for VLA reasoning.

[0094] The overall representation of the multimodal fusion coding process is as follows:

[0095]

[0096] in, This represents the multimodal fusion coding function. Indicates the first A unified scene-preference semantic representation at any given moment; This represents a personalized dynamic experience preference profile of passengers; Represents an environmental image; Represents navigation semantic information; Indicates vehicle status information; Indicates chassis status information; This indicates the occupant status information.

[0097] The key technical point of this step lies in the fact that occupant preferences are not interpreted in isolation, but rather in conjunction with information such as "the current scenario," "the current dynamic state of the vehicle," "whether the chassis currently has sufficient control margin," and "whether the occupants are currently in a sleep or comfort-sensitive state." In this way, the same linguistic preference can derive different control semantics in different scenarios. For example, "be more stable" in a low-adhesion cornering scenario can be interpreted as reducing the upper limit of yaw rate and enhancing stability control intervention; "be more stable" in a high-speed lane change scenario can be interpreted as limiting lateral acceleration and reducing steering angular velocity; "be more gentle" in a speed bump scenario can be interpreted as reducing pitch and vertical impact; and "be more gentle" in a city following scenario can be interpreted as reducing longitudinal jerk and braking pressure change rate. Therefore, this step provides a contextual basis for the subsequent generation of personalized chassis style intentions.

[0098] Furthermore, this embodiment also provides a specific implementation encoding method. A multimodal fusion encoding module is used to jointly encode occupant preference profiles, environmental images, navigation semantics, vehicle status, chassis status, and occupant status to form a unified scene-preference semantic representation. Specifically, the following encoding results can be constructed respectively:

[0099]

[0100] in, , , , , and These represent the preference encoder, visual encoder, navigation semantic encoder, vehicle status encoder, chassis status encoder, and occupant status encoder, respectively. This indicates a personalized dynamic experience preference profile of passengers. Preferred encoder The preference feature representation obtained after encoding; Represents vehicle environment images Vision encoder The visual scene feature representation obtained after encoding; Indicates navigation semantic information via navigation semantic encoder The semantic feature representation of the navigation task obtained after encoding; Indicates vehicle status information via vehicle status encoder The encoded representation of the vehicle's motion state characteristics; This indicates the chassis status information. chassis status encoder The encoded representation of the chassis system state characteristics; Indicates passenger status information Passenger status encoder The occupant state feature representation obtained after encoding.

[0101] Furthermore, a unified scene-preference semantic representation is formed through a multimodal fusion network, specifically represented as follows:

[0102]

[0103] in, This represents the multimodal fusion coding function. This can be achieved through methods such as splicing mapping, cross-attention fusion, gating fusion, or hierarchical fusion. Indicates the first A unified scene-preference semantic representation for each moment.

[0104] The technical role of the multimodal fusion coding module is to express not only "what the occupants want," but also "the current road and task context," "whether the vehicle is currently suitable for this style," and "what adjustable resources the chassis currently has." This is a key technical basis that distinguishes this invention from single voice mode switching schemes.

[0105] S4. Input the unified scene-preference semantic representation into the vision-language-action model to generate scene-related personalized chassis style intent;

[0106] Using a visual-language-action personalized style reasoning module to represent the semantics of a unified scene and preference Perform semantic reasoning to output a personalized chassis style intent. This process is represented as follows:

[0107]

[0108] in, Represents the visual-language-action model. This indicates an intention to personalize the chassis style. Indicates the first A unified scene-preference semantic representation for each moment.

[0109] Specifically: the personalized chassis style intent output by the vision-language-action model includes personalized control semantic template identifiers and parameterized adjustment vectors, namely:

[0110]

[0111] in, This indicates a personalized control semantic template identifier; This represents the parameterized adjustment vector.

[0112] Personalized control semantic template It includes at least one of the following: stability priority template; comfort priority template; smoothness priority template; sleep occupant relief template; high-speed lane change conservative template; low adhesion conservative template; and passability enhancement template.

[0113] The optimal definition of the parameterized adjustment vector is:

[0114]

[0115] in, This is the adjustment amount for the upper limit of lateral acceleration; This is the adjustment amount for the upper limit of yaw rate; This is the adjustment amount for the roll angle constraint; This is the pitch angle constraint adjustment amount; Adjustment amount for steering angular velocity limit; The adjustment amount is limited by the rate of change of braking pressure; This refers to the suspension damping adjustment amount; This refers to the suspension stiffness adjustment amount; To adjust the smoothness coefficient of the driving response; Adjust the weight of comfort control; The weight adjustment amount is used to control stability; This represents the transpose of a matrix.

[0116] The parameterized adjustment vector includes at least one of the following adjustment quantities: lateral acceleration upper limit adjustment quantity, yaw rate upper limit adjustment quantity, roll angle constraint adjustment quantity, pitch angle constraint adjustment quantity, steering angle rate limit adjustment quantity, brake pressure change rate limit adjustment quantity, suspension damping adjustment quantity, suspension stiffness adjustment quantity, drive response smoothness coefficient adjustment quantity, comfort control weight adjustment quantity, and stability control weight adjustment quantity.

[0117] The key point of the visual-language-action personalized style reasoning module is that the same preference profile will be interpreted as different style intentions in different scenarios. For example, "playing it safe" will emphasize yaw and side slip control on curved roads, while it will emphasize pitch and vertical impact suppression on speed bumps.

[0118] With the above settings, VLA outputs not low-level actuator commands, but high-level style intents for the chassis controller. This distinguishes the present invention from simple language control schemes, as well as from visual-language model schemes that only have scene understanding but no action semantics.

[0119] In this embodiment, the personalized chassis style intent is generated through mapping using a personalized control template library. This library stores chassis control templates corresponding to various combinations of occupant personalized preferences and scenarios. Each template in the library includes at least one of the following: dynamic performance index constraint rules, controller weight update rules, suspension adjustment rules, steering adjustment rules, braking adjustment rules, drive adjustment rules, stability control threshold adjustment rules, parameter recovery rules, and preference feedback update rules. The personalized control template library stores chassis control templates corresponding to various combinations of occupant preferences and scenarios and supports online updates based on feedback.

[0120] The personalized chassis style intent output by the vision-language-action model is not directly equal to the underlying actuator control quantity, but rather outputs a high-level style template and parameterized adjustment vector for the chassis controller. When generating personalized chassis style intent, the vision-language-action model also incorporates at least one of the following for modification: navigation task priority, prior road information, occupant status, and safety constraint information.

[0121] The personalized chassis control process includes the preference identification stage, the scenario joint interpretation stage, the parameter mapping stage, the personalized control execution stage, and the feedback update stage.

[0122] S5. Based on the personalized chassis style intent, vehicle status information, chassis status information, and current scene semantic information, generate chassis dynamic performance index constraints and controller parameter adjustment targets.

[0123] The dynamic performance index and controller parameter generation module generates dynamic performance constraints and controller parameter adjustment targets based on personalized chassis style intent, vehicle status, chassis status, and current scene semantics.

[0124] First, define the chassis dynamic performance index constraint vector. for:

[0125]

[0126] in, Indicates the upper limit of lateral acceleration; Indicates the upper limit of yaw rate; Indicates the upper limit of the roll angle; Indicates the upper limit of the pitch angle; Indicates the steering angular velocity limit; Indicates the limit of the rate of change of braking pressure; This indicates the limit on the rate of change of longitudinal acceleration.

[0127] In this embodiment, the above constraints can be obtained by style intent mapping, for example:

[0128]

[0129] in, This represents the nominal value of the upper limit of lateral acceleration, which is the baseline upper limit of lateral acceleration allowed by the vehicle when no personalized style adjustments are made; This is the adjustment amount for the upper limit of lateral acceleration; This represents the nominal value of the upper limit of yaw rate, which is the baseline upper limit of yaw rate allowed by the vehicle when no personalized style adjustment is made; This is the adjustment amount for the upper limit of yaw rate; This represents the nominal value of the upper limit of the roll angle, which is the baseline upper limit of the roll angle allowed by the vehicle when no personalized style adjustment is made; This is the adjustment amount for the roll angle constraint; This represents the nominal value of the brake pressure change rate limit, which is the upper limit of the baseline brake pressure change rate allowed by the vehicle when no personalized style adjustment is performed; The adjustment amount is limited by the rate of change of braking pressure.

[0130] Secondly, define the controller parameter adjustment target. for:

[0131]

[0132] in, Represents the controller state weight matrix; This represents the controller input weight matrix; This indicates the threshold parameter for stable control intervention; Indicates the suspension damping parameters; Indicates the suspension stiffness parameters; Indicates the steering assist characteristic parameters; This represents the braking response gain parameter; This represents the smoothness coefficient parameter of the drive response; This represents the transpose of a matrix.

[0133] Therefore, the process of generating constraints and parameters can be uniformly represented as:

[0134]

[0135] in, Represents the semantic information of the current scene. A function that represents the mapping from style intent to dynamic constraints and controller parameters. Indicates vehicle status information; This indicates the chassis status information. This indicates an intention to personalize the chassis style.

[0136] Furthermore, when the chassis controller employs a model predictive controller, its cost function... It can be written as:

[0137]

[0138] in, Indicates the length of the prediction time domain; Represents the predicted state vector transpose; Represents the controller state weight matrix; This represents the vehicle state vector at the i-th prediction step, starting from time k. Represents the predictive control input vector transpose; This indicates the penalty weight for different control inputs or actuator inputs; This represents the control input vector at the i-th prediction step, starting from time k.

[0139] In this embodiment, and The system adapts to the individual chassis style intent rather than remaining fixed. For example, under the stability-first template, the weights of yaw and sideslip-related states are increased; under the comfort-first template, the weights of pitch, roll, and jerk-related states are increased; and under the smoothness-first template, the penalty weight for the rate of change of control inputs is increased.

[0140] When stability is prioritized, increase the state weights corresponding to yaw rate, sideslip angle, and lateral error; when comfort is prioritized, increase the state weights corresponding to roll, pitch, and jerk; when smoothness is prioritized, increase the penalty weight for the rate of change of control input.

[0141] The chassis dynamic performance constraints include at least one of the following: upper limit of lateral acceleration, upper limit of yaw rate, upper limit of roll angle, upper limit of pitch angle, limit of steering rate, limit of brake pressure change rate, and limit of longitudinal acceleration change rate.

[0142] The controller parameter adjustment targets include at least one of the following: model predictive controller state weight matrix adjustment target, model predictive controller input weight matrix adjustment target, stability control intervention threshold adjustment target, suspension damping control parameter adjustment target, suspension stiffness control parameter adjustment target, steering assist characteristic adjustment target, braking response gain adjustment target, and drive torque smoothing control parameter adjustment target.

[0143] S6. Based on the chassis dynamic performance index constraints and controller parameter adjustment targets, perform parameter tuning, control constraint update, response characteristic switching, and control mode switching for each chassis system.

[0144] Call the personalized control template library to generate specific chassis control targets.

[0145] After generating dynamic performance constraints and controller parameters, this application further performs personalized coordinated control on the steering system, braking system, drive system, active suspension system, and yaw stability control system.

[0146] In this application, the multi-actuator personalized collaborative control includes at least one of the following operations:

[0147] 1. Personalized suspension adjustment;

[0148] The suspension damping and stiffness are adjusted according to preferences such as comfort priority, sleep-friendly, or gentle passage over speed bumps to reduce vertical impact, pitch, and roll.

[0149] 2. Personalized brake adjustment;

[0150] Adjust the brake build-up rate, brake response gain, and brake pressure change rate limit according to the preference for smoothness, comfort, or safety.

[0151] 3. Shift towards personalized adjustment;

[0152] Adjust the steering assist characteristics, steering sensitivity, and steering angular velocity limit according to whether the preference is for stability, high-speed lane changing, or comfort.

[0153] 4. Drive personalized adjustments;

[0154] Adjust the drive torque change rate, torque ramp rate, and drive response smoothness coefficient according to whether the preference is for low adhesion conservatism, comfort priority, or ride comfort.

[0155] 5. Stable control and personalized adjustment;

[0156] Adjust the yaw stability control intervention threshold, yaw rate error weight, and additional yaw torque trigger threshold based on stability preference, safety preference, and scenario risk level.

[0157] To avoid sudden changes in chassis and controller parameters, a gradual update approach is adopted:

[0158]

[0159] in, Indicates the current parameter set; Represents the target parameter set; This indicates the gradual adjustment factor for chassis parameters.

[0160] The update factor can be determined based on preference intensity, vehicle speed, road surface adhesion, and occupant sensitivity, for example:

[0161]

[0162] in, Indicates the level of preference intensity. For longitudinal velocity, Indicates the adhesion state parameters, This indicates the occupant status information. This represents a mapping function.

[0163] In this embodiment, a personalized chassis control target is generated by using a personalized control template library and a control target generation module, based on the control semantic template identifier, parameterized adjustment vector, and the current state of the vehicle and chassis.

[0164] Set up a personalized control template library for:

[0165]

[0166] Each template It includes at least: dynamic performance constraint rules; controller weight update rules; suspension adjustment rules; steering adjustment rules; braking adjustment rules; drive adjustment rules; stability control threshold adjustment rules; parameter recovery rules; and preference feedback update rules.

[0167] Furthermore, control target It can be represented as:

[0168]

[0169] in, This represents the mapping function from the template and adjustment vector to the specific control target. This indicates a personalized control semantic template identifier; This represents the parameterized adjustment vector. Indicates vehicle status information; Indicates chassis status information; This indicates the occupant status information. It represents the semantic information of the current scene.

[0170] In this embodiment, the current scene semantic information includes at least one of the following: road attachment state, road geometric features, road undulation features, lane change features, traffic flow state, road functional area semantics, and target driving task semantics.

[0171] In this embodiment, parameter tuning, control constraint updates, response characteristic switching, or control mode switching are implemented using a progressive smooth update method. The progressive smooth update method determines the update rate based on at least one of preference intensity, vehicle speed, road surface adhesion state, and occupant state.

[0172] S7. Perform personalized active chassis control based on updated chassis parameters during vehicle operation;

[0173] Perform personalized collaborative control of multiple actuators; when performing personalized active chassis control during vehicle operation, at least two actuators from the steering system, braking system, drive system, active suspension system, and yaw stability control system shall perform collaborative control.

[0174] During this phase, the controller continuously outputs control commands that align with the personalized chassis style intent based on updated dynamic performance constraints and controller parameters. For example, in high-speed lane change scenarios, if a conservative preference is adopted, the controller limits lateral acceleration and steering angular velocity to enhance lateral stability control; in speed bump scenarios, if a gentler preference is adopted, the controller limits pitch change rate, reduces abrupt changes in suspension damping, and smooths braking response; in low-adhesion cornering scenarios, if a stability-first preference is adopted, the controller enhances yaw and sideslip constraints and suppresses rapid changes in drive torque; in rear passenger sleeping scenarios, if a more relaxed preference is adopted, the controller focuses on constraining roll angle, pitch angle, and longitudinal jerk.

[0175] Therefore, the personalized chassis active control in this embodiment is not a one-time mode switch, but a dynamic control that is continuously performed according to preferences and scenarios throughout the entire operation.

[0176] By utilizing the chassis-specific controller, personalized collaborative control of multiple actuators is performed according to the control objectives.

[0177] In this embodiment, the chassis parameter set Defined as:

[0178]

[0179] in, Represents the controller state weight matrix; This represents the controller input weight matrix; This indicates the threshold parameter for stable control intervention; Indicates the suspension damping parameters; Indicates the suspension stiffness parameters; Indicates the steering assist characteristic parameters; This represents the braking response gain parameter; This represents the smoothness coefficient parameter of the drive response; This represents the transpose of a matrix.

[0180] Then the adjusted target parameter set for:

[0181]

[0182] in, This represents the chassis parameter set; This represents the chassis control target generated from the personalized control template library and personalized chassis style intent; This indicates the parameter update mapping function.

[0183] To avoid sudden parameter changes, incremental updates are adopted, specifically:

[0184]

[0185] in, This is the gradual adjustment factor for chassis parameters. This represents the chassis parameter set updated at time k+1. This represents the target chassis parameter set generated at time k based on the chassis control target; This represents the chassis parameter set.

[0186] In this embodiment, It can be determined based on preference intensity, vehicle speed, adhesion status, and occupant sensitivity.

[0187] The personalized controller performs at least one or more of the following operations: personalized suspension adjustment; personalized brake adjustment; personalized steering adjustment; personalized drive adjustment; personalized stability control adjustment; and unified controller mode switching.

[0188] S8. Based on passenger feedback, vehicle response, and scene changes, restore, correct, or update the passenger's personalized dynamic experience preference profile and personalized control template.

[0189] To achieve truly personalized control, this invention further incorporates a preference feedback learning and template update mechanism.

[0190] The system utilizes a feedback learning update module to modify and update occupant preference profiles and template libraries based on occupant feedback, vehicle response, and scenario changes. On one hand, when preferences are removed, scenarios change, or the current task ends, the system can gradually restore chassis parameters to the normal parameter set; on the other hand, the system modifies occupant preference profiles and personalized control templates based on occupant feedback, actual vehicle dynamic response, and scenario changes.

[0191] The parameter recovery process can be represented as:

[0192]

[0193] in, Represents the set of regular parameters. This indicates the recovery factor. This represents the chassis parameter set updated at time k+1. This represents the chassis parameter set.

[0194] Meanwhile, the learning and updating of preference profiles can still be represented as:

[0195]

[0196] in, This represents the personalized dynamic experience preference profile of the occupants at time k+1; Indicates the preference update coefficient; This represents a personalized dynamic experience preference profile of passengers. This represents the instantaneous preference vector obtained by parsing the current language input and feedback information.

[0197] The template parameters in the template library can be updated based on comfort evaluation indicators, yaw stability evaluation indicators, vehicle posture response indicators, occupant feedback scores, and control input smoothness indicators.

[0198] Through the aforementioned feedback learning mechanism, this invention can not only adjust the chassis behavior in real time based on the current occupant's expression, but also gradually form a control style that better suits individual preferences during long-term use.

[0199] Feedback information can come from: occupant voice correction; occupant rating; motion sickness or comfort monitoring results; vehicle posture and dynamic response evaluation indicators; and control input smoothness evaluation results.

[0200] like Figure 3 As shown, the implementation process of this invention includes occupant preference input, VLA joint reasoning, personalized chassis style intent generation, chassis collaborative control, and feedback learning update.

[0201] First, the system acquires information such as occupant natural language preferences, historical preference profiles, environmental images, navigation semantics, and vehicle status. For example, when an occupant inputs "Someone is sleeping in the back seat, minimize body roll and head nodding," this input represents the occupant's personalized needs for comfort, smoothness, and vehicle posture stability. Environmental images, navigation semantics, and vehicle status are then used to represent the current road scene and vehicle operating status.

[0202] Subsequently, the system inputs the aforementioned multi-source information into the vision-language-action model, which jointly understands occupant preferences and the current scenario, and generates personalized chassis style intents. These style intents are not low-level actuator instructions, but rather high-level control semantics oriented towards the chassis controller, such as stability priority, comfort priority, or sleep-friendly.

[0203] Furthermore, the system generates chassis dynamic performance constraints and controller parameter adjustment targets based on the personalized chassis style intent, and applies them to chassis systems such as steering, braking, drive, and suspension. For Figure 3 In the scenario shown, the system can reduce the upper limit of yaw rate, reduce roll and pitch response, smooth the braking process, and make the suspension response more gentle, thereby improving the ride comfort of rear passengers in sleeping scenarios.

[0204] Finally, vehicle response, occupant feedback, and scenario change information are fed back to the historical preference profile and control template library through the feedback learning update module. This is used to continuously correct the personalized preference model and chassis control strategy, achieving long-term personalized optimization.

[0205] Example 2

[0206] This embodiment provides an intelligent chassis control system driven by occupant personalized preferences based on a vision-language-action model, including: a multi-source information acquisition module, an occupant preference profile management module, a multimodal fusion coding module, a vision-language-action personalized style reasoning module, a dynamic performance index and controller parameter generation module, a personalized control template library and control target generation module, a chassis personalized controller, and a feedback learning and update module.

[0207] The multi-source information acquisition module is used to acquire multi-source information related to occupant preferences and vehicle status, including: images of the environment in front of the vehicle, navigation semantic information, map semantic information, vehicle operating status information, chassis status information, and occupant status information.

[0208] The passenger preference profile management module is used to model passengers' personalized and dynamic experience preferences. This module constructs long-term and short-term preference profiles by semantically parsing and recording passengers' natural language input (such as "The road is slippery today, be careful" or "Someone is sleeping in the back, try to minimize tilting").

[0209] The multimodal fusion coding module jointly encodes information from different sources to generate a unified scene-preference semantic representation, including: preference representation, visual information representation, and scene semantic representation.

[0210] The visual-language-action personalized style reasoning module generates personalized chassis style intents based on a unified scene-preference semantic representation. Using a visual-language-action model, this module infers from multimodal input information and outputs personalized control style intents.

[0211] The dynamic performance index and controller parameter generation module is used to generate chassis dynamic performance indexes and controller parameter adjustment targets based on visual-linguistic-motor style intent.

[0212] A personalized control template library and control target generation module are used to generate corresponding chassis control targets based on specific driving scenarios and occupant preferences. This module selects appropriate control templates from the template library and maps style intent parameterized adjustment values ​​to specific control targets.

[0213] The chassis personalization controller module is used to make personalized adjustments to the steering system, braking system, drive system, suspension system, stability control system, etc., according to the generated control objectives.

[0214] The feedback learning update module is used to continuously update the preference profile, control template library, and parameters based on passenger feedback, scenario characteristics, and vehicle responses during vehicle operation.

[0215] Specifically, the above modules can be embedded into a computer processing system. The computer calls the above modules to complete the control tasks according to the control method provided above. The above modules can perform operations according to the specific steps given by the control method above.

[0216] It should be noted that the division of the various modules in the above system is merely a division of logical functions. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. These modules can be implemented entirely in software through processing element calls; they can be fully implemented in hardware; or some modules can be implemented by processing element calls to software, while others are implemented in hardware. For example, the model building module can be a separate processing element, or it can be integrated into a chip in the aforementioned device. Alternatively, it can be stored as program code in the memory of the aforementioned device, and its signal processing module functions can be called and executed by a processing element of the device. The implementation of other modules is similar. Furthermore, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element mentioned here can be an integrated circuit with signal processing capabilities. During implementation, each step of the above method or each of the above modules can be completed through integrated logic circuits in the hardware of the processor element or through software instructions.

[0217] For example, these modules can be one or more integrated circuits configured to implement the above methods, such as one or more Application Specific Integrated Circuits (ASICs), one or more Digital Signal Processors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs). As another example, when a module is implemented using processing element scheduler code, the processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together to form a system-on-a-chip (SOC).

[0218] In this embodiment, a typical scenario implementation includes:

[0219] 1. Low adhesion and stable performance are preferred;

[0220] When an occupant inputs "The road is slippery today, let's play it safe," the environmental image identifies a wet and slippery road surface, and the navigation semantics indicate a curved section ahead. After constructing a preference profile, the system uses VLA inference to obtain a stability priority template and corresponding parameterized adjustment vectors, including: reducing the upper limit of lateral acceleration, reducing the upper limit of yaw rate, increasing the sensitivity of stability control intervention, and limiting the rate of change of drive torque. Based on this, the system updates the stability control threshold, MPC state weights, and drive response parameters, thereby improving lateral stability under low-adhesion conditions.

[0221] 2. Preference for comfortable passage over speed bumps;

[0222] When an occupant inputs "be gentler over speed bumps," the environmental image identifies a series of speed bumps ahead, indicating the presence of a comfort-sensitive occupant in the rear seats. The VLA outputs a comfort-priority template and parameterized adjustment vectors, including: reducing the rate of change of braking pressure, lowering suspension damping, limiting steering angular velocity, and increasing the weighting of pitch suppression. Based on this, the system implements gentler chassis control when entering the speed bump area.

[0223] 3. Conservative preference for high-speed lane changes;

[0224] The occupant inputs "Don't be too aggressive when changing lanes at high speed," and the vehicle is currently traveling on a high-speed straight road. The VLA outputs a conservative high-speed lane-changing template, including: limiting the upper limit of yaw rate, reducing steering sensitivity, increasing the weight of lateral stability, and limiting the rate of increase of lateral acceleration. Based on this, the system reduces the intensity of lane-change maneuvers.

[0225] 4. Sleep occupant comfort preference;

[0226] When an occupant inputs "Someone is sleeping in the back, minimize roll and head nodding," the occupant status indicates that the rear occupant is asleep. The VLA outputs a sleep occupant comfort template, including: limiting the upper limit of roll angle, limiting the upper limit of pitch angle, reducing the rate of change of braking and driving transients, and adjusting suspension damping to enhance comfort. Based on this, the system generates a chassis control strategy that is more conducive to a better sleep experience.

[0227] Example 3

[0228] The present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. The memory stores the computer program capable of running on the processor, and the processor loads and executes the computer program using the above-mentioned control method.

[0229] It should be noted that the terminal device can be a computer device such as a desktop computer, a laptop computer, or a cloud server, and the terminal device includes, but is not limited to, a processor and a memory. For example, the terminal device may also include input / output devices, network access devices, and buses.

[0230] Furthermore, the processor can be a central processing unit (CPU). Of course, depending on the actual use, other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. can also be used. The general-purpose processor can be a microprocessor or any conventional processor, etc., and this application does not limit it in this regard.

[0231] Example 4

[0232] This embodiment provides a computer-readable storage medium for storing a computer program that enables a computer to execute the above-described control method.

[0233] This invention proposes a novel chassis control scheme that enables the system to jointly reason about occupant natural language preferences, current road visual scene, vehicle state, and chassis state using a vision-language-action model. This not only identifies occupant dynamic experience preferences but also outputs chassis control intentions that correspond to these preferences and the scene. Furthermore, it generates dynamic performance constraints, control parameters, and control modes, thereby achieving personalized, scenario-based, and explainable control of the intelligent chassis.

[0234] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for occupant personalized preference driven intelligent chassis control based on vision-language-action model, characterized in that, Includes the following steps: S1. Collect occupant language preference input, vehicle environment image, navigation semantic information, vehicle status information, chassis status information, and occupant status information; S2. Construct or update a personalized dynamic experience preference profile of passengers based on their natural language preference input; S3. Encode and fuse the personalized dynamic experience preference profile of the occupants, vehicle environment image, navigation semantic information, vehicle status information, chassis status information and occupant status information to obtain a unified scene-preference semantic representation. S4. Input the unified scene-preference semantic representation into the vision-language-action model to generate scene-related personalized chassis style intent; S5. Based on the personalized chassis style intent, vehicle status information, chassis status information, and current scene semantic information, generate chassis dynamic performance index constraints and controller parameter adjustment targets. S6. Based on the chassis dynamic performance index constraints and controller parameter adjustment targets, perform parameter tuning, control constraint update, response characteristic switching, and control mode switching for each chassis system. S7. Perform personalized active chassis control based on updated chassis parameters during vehicle operation; S8. Based on passenger feedback, vehicle response, and scene changes, restore, correct, or update the passenger's personalized dynamic experience preference profile and personalized control template.

2. The occupant personalized preference-driven intelligent chassis control method based on a vision-language-action model according to claim 1, characterized in that, In S1, the specific implementation details are as follows: The multi-source information acquisition module is used to obtain passenger language preference input during vehicle operation. Environmental images Navigation semantic information Vehicle status information Chassis status information Crew status information ; No. The multi-source information collected at different times is denoted as follows: 。 3. The occupant personalized preference-driven intelligent chassis control method based on a vision-language-action model according to claim 1, characterized in that, In S2, the specific implementation details are as follows: Personalized dynamic experience profile of passengers Defined as: ； in, Indicates the stability preference weight; Indicates the weight of comfort preference; Indicates the smoothness preference weight; Indicates the degree of radicalism in preference weights; Indicates the weight of safety preference; Indicates the passivity preference weight; Represents the transpose of a matrix; Preference profiles are updated recursively. The current language input is first converted into an instantaneous preference vector by a language semantic parser. Then, it is integrated and updated with historical preference profiles, specifically as follows: ； in, This represents the immediate preference vector obtained by parsing the current language input and feedback information. Indicates the preference update coefficient; This represents the personalized dynamic experience preference profile of the occupants at time k+1; This represents a personalized dynamic experience preference profile of passengers.

4. The occupant personalized preference driven intelligent chassis control method based on visual-linguistic-action model of claim 1, wherein, In S3, the specific implementation details are as follows: The overall representation of the multimodal fusion coding process is as follows: ； in, This represents the multimodal fusion coding function. Indicates the first A unified scene-preference semantic representation at any given moment; This represents a personalized dynamic experience preference profile of passengers; Represents an environmental image; Represents navigation semantic information; Indicates vehicle status information; Indicates chassis status information; This indicates the occupant status information.

5. The occupant personalized preference-driven intelligent chassis control method based on a vision-language-action model according to claim 1, characterized in that, In S4, the specific implementation details are as follows: Using a visual-language-action personalized style reasoning module to represent the semantics of a unified scene and preference Perform semantic reasoning to output personalized chassis style intent, the process of which is represented as follows: ； in, Represents the visual-language-action model. This indicates an intention to personalize the chassis style; Indicates the first A unified scene-preference semantic representation for each moment.

6. The occupant personalized preference-driven intelligent chassis control method based on a vision-language-action model according to claim 1, characterized in that, In S5, the specific implementation details are as follows: S501: First, define the chassis dynamic performance index constraint vector. ; S502: Style intent mapping of chassis dynamic performance index constraint vectors; S503: Define the controller parameter adjustment target and provide a unified representation of the generation process of the chassis dynamic performance index constraint vector and the controller parameter adjustment target.

7. A occupant personalized preference-driven intelligent chassis control system based on a vision-language-action model, implementing the occupant personalized preference-driven intelligent chassis control method based on any one of claims 1 to 6, characterized in that, The system includes: The multi-source information acquisition module is used to acquire occupant natural language preference input, vehicle environment images, navigation semantic information, vehicle status information, chassis status information, and occupant status information; The passenger preference profile management module is used to build or update personalized dynamic experience preference profiles of passengers based on their natural language preference input. The multimodal fusion coding module is used to fuse and encode the personalized dynamic experience preference profile of the occupants, vehicle environment image, navigation semantic information, vehicle status information, chassis status information and occupant status information to generate a unified scene-preference semantic representation. The visual-language-action personalized style reasoning module is used to generate scene-related personalized chassis style intents based on a unified scene-preference semantic representation. The dynamic performance index and controller parameter generation module is used to generate chassis dynamic performance index constraints and controller parameter adjustment targets based on personalized chassis style intent, vehicle operating status information, chassis status information and current scene semantic information. A personalized control template library and a control target generation module are used to generate personalized chassis control targets based on personalized chassis style intentions. The chassis personalization controller module is used to perform parameter tuning, control constraint update, response characteristic switching or control mode switching for the steering system, braking system, drive system, active suspension system and / or yaw stability control system according to the chassis dynamic performance index constraints and controller parameter adjustment targets, and to execute personalized chassis active control. The feedback learning update module is used to restore, correct, or update the personalized dynamic experience preference profile and personalized control template of passengers based on passenger feedback, vehicle response, and changes in the scenario.

8. An electronic device, comprising: The method includes a processor, a memory, and a computer program stored in the memory, wherein the processor executes the computer program to implement the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 6.

Images