Methods and vehicles for adjusting comfort in autonomous driving

Abstract

This application relates to the field of autonomous driving technology, providing a method and vehicle for adjusting the comfort of autonomous driving. The method includes: real-time acquisition of image data of occupants within the vehicle cabin; performing posture recognition processing on the image data to determine the current posture category of the occupants; determining a control mode corresponding to the posture category based on the posture category; generating cross-domain cooperative control commands based on the control mode, and distributing the cross-domain cooperative control commands to at least two of the autonomous driving domain controller, chassis domain controller, and cabin domain controller for execution. This method enables dynamic adjustment of the vehicle's control strategy based on the real-time posture of the occupants, and improves ride comfort under different postures through cross-domain cooperative control.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automatic driving, in particular to an automatic driving comfort adjustment method and a vehicle. BACKGROUND

[0002] With the continuous development of automatic driving technology, L3 and above levels of automatic driving systems gradually enter the mass production application stage. In the automatic driving mode, the driver can be liberated from the continuous driving operation task, and can engage in reading, using electronic devices or resting and other non-driving activities during vehicle travel.

[0003] The existing automatic driving vehicle mainly formulates a control strategy according to external environment perception information (such as front vehicle distance, road curvature, traffic signs, etc.) in terms of path planning and vehicle control. This control method makes all passengers in the cabin bear the same acceleration and lateral acceleration change when the vehicle performs acceleration, braking or lane changing operations, regardless of their body posture. However, the sensitivity of passengers to vehicle dynamic response varies under different activity states. When the passengers are in the state of reading or resting, they are more sensitive to changes in vehicle motion. Driving according to a unified control strategy is likely to cause discomfort to passengers.

[0004] In addition, although the existing vehicle provides a variety of driving modes for the driver to choose from, these modes usually need to be manually switched by hand, and the definition of the modes is relatively fixed and cannot be adaptively adjusted according to the real-time state of the passengers in the cabin. At the same time, the adjustment of comfort not only involves the control of the driving behavior of the vehicle, but also relates to the suspension response of the chassis, the cabin environment and other dimensions. However, the existing technology still has deficiencies in coordinating these different control domains.

[0005] Therefore, how to dynamically adjust the control strategy of the automatic driving system according to the real-time state of the passengers in the vehicle and cooperatively adjust multiple control domains to improve the comfort of riding has become a technical problem to be solved in the field. SUMMARY

[0006] Therefore, the embodiments of the present application provide an automatic driving comfort adjustment method and a vehicle, which can dynamically adjust the control strategy of the vehicle based on the real-time posture of the passengers and improve the comfort of riding in different postures through cross-domain cooperative control.

[0007] The first aspect of the embodiments of the present application provides an automatic driving comfort adjustment method, comprising: real-time acquisition of image data of passengers in a vehicle cabin; performing posture recognition processing on the image data to determine the current posture category of the passengers; determining a control mode corresponding to the posture category according to the posture category; Based on the control mode, cross-domain collaborative control instructions are generated and distributed to at least two of the autonomous driving domain controller, chassis domain controller, and cockpit domain controller for execution.

[0008] A second aspect of this application provides a vehicle, including: Image acquisition devices deployed inside the cockpit are used to collect image data of the occupants in real time; A processor communicatively connected to the image acquisition device is configured to: perform attitude recognition processing on the image data to determine the current attitude category of the occupant, determine the corresponding control mode based on the attitude category, and generate cross-domain collaborative control instructions based on the control mode; At least two of the autonomous driving domain controller, chassis domain controller, and cockpit domain controller, which are respectively communicatively connected to the processor, are used to receive and execute the cross-domain cooperative control instructions.

[0009] The first aspect of the autonomous driving comfort adjustment method of this application collects image data of occupants in the vehicle cabin in real time, performs posture recognition processing on the image data to determine the current posture category of the occupants, determines the corresponding control mode according to the posture category, and generates cross-domain cooperative control commands according to the control mode and distributes them to at least two of the autonomous driving domain controller, chassis domain controller and cabin domain controller. By introducing the real-time posture information of the occupants into the control decision of the autonomous driving system, multi-domain cooperative comfort adjustment based on the occupant state is realized, so that occupants in different postures can obtain dynamic response and environmental experience that matches their current activities, thereby improving the riding comfort of autonomous vehicles.

[0010] It is understandable that the beneficial effects of the second aspect mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart illustrating the autonomous driving comfort adjustment method provided in an embodiment of this application; Figure 2 This is a timing diagram of multi-domain collaborative control provided in an embodiment of this application; Figure 3This is a schematic diagram of the posture recognition and classification process provided in the embodiments of this application; Figure 4 This is a schematic diagram of the key skeletal points of the occupant provided in the embodiments of this application; Figure 5 This is a schematic diagram of the architecture of the autonomous driving comfort adjustment system provided in the embodiments of this application. Detailed Implementation

[0013] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0014] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0015] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0016] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0017] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0018] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0019] The autonomous driving comfort adjustment method provided in this application can be applied to various vehicles equipped with Level 3 or higher autonomous driving systems, including but not limited to passenger cars, commercial vehicles, and buses. In autonomous driving mode, drivers and passengers may be in different postures, such as reading, using electronic devices, resting, or looking ahead. The sensitivity of occupants to vehicle dynamic responses varies depending on their posture. This application provides targeted comfort experiences for occupants in different postures by sensing their postures in real time and adjusting the vehicle control strategy accordingly.

[0020] like Figure 1 As shown, the autonomous driving comfort adjustment method provided in this application includes the following steps S101 to S104: Step S101: Real-time acquisition of image data of occupants inside the vehicle cabin.

[0021] In this application, cameras deployed within the vehicle cabin capture real-time images of occupants. These cameras may include a Driver Monitoring System (DMS) mounted facing the driver and an Occupant Monitoring System (OMS) mounted facing the passengers, providing comprehensive coverage of occupants in different positions within the cabin. To ensure accuracy and real-time performance in attitude recognition, the camera's frame rate should ideally be no less than 30fps, and its resolution no less than 720p. Furthermore, the cameras may be equipped with infrared illumination to ensure clear occupant image data is acquired even at night or in low-light conditions. The image data captured by the cameras is transmitted in real-time to the vehicle's processor via an in-vehicle communication interface for subsequent attitude recognition processing.

[0022] Step S102: Perform posture recognition processing on the image data to determine the current posture category of the occupant.

[0023] In this application, the posture recognition process utilizes computer vision and deep learning to analyze the occupant's body posture from acquired image data and categorize it into predefined posture types. Posture recognition analyzes the spatial position and motion state of various parts of the occupant's body to infer the type of activity the occupant is currently engaged in. For example, when the occupant's head is lowered and both hands are holding an object, it can be determined that they are in a reading or using an electronic device posture; when the occupant's eyelids are closed and their head is tilted back, it can be determined that they are in a resting posture. Different posture categories reflect the occupant's varying sensitivity to the vehicle's dynamic response, providing a basis for subsequent control mode determination. Specific posture recognition processing methods will be described in detail in subsequent embodiments.

[0024] Step S103: Determine the control mode corresponding to the posture category based on the posture category.

[0025] In application, the control mode refers to a predefined set of vehicle control parameters, with different control modes corresponding to different vehicle dynamic response strategies. The system has a built-in attitude-control mode mapping rule, which can be implemented based on a rule engine or a lightweight decision tree. The input is the occupant's attitude category, and the output is the corresponding control mode label. When the occupant is in an attitude highly sensitive to vehicle dynamic response (such as reading or resting), the system automatically maps to a smoother or quieter control mode to reduce the interference of vehicle motion changes on the occupant; when the occupant is in an attitude less sensitive to vehicle dynamic response (such as looking ahead), the system can maintain the standard control mode. This mapping process does not require manual operation by the occupant, realizing automated control strategy switching based on occupant status.

[0026] In application, when the system simultaneously identifies the postures of multiple occupants and the posture categories of each occupant are inconsistent, a strategy prioritizing the highest comfort requirement can be adopted for mode decision-making. Specifically, the system sorts the posture categories of each occupant from high to low sensitivity to vehicle dynamics (resting with eyes closed > reading and writing > looking straight ahead), and takes the posture category with the highest sensitivity as the final mapping input to ensure that the most sensitive occupant in the vehicle will not experience discomfort due to the vehicle's dynamic response.

[0027] Step S104: Based on the control mode, generate cross-domain collaborative control instructions and distribute the cross-domain collaborative control instructions to at least two of the autonomous driving domain controller, chassis domain controller and cockpit domain controller for execution.

[0028] In application, cross-domain cooperative control refers to simultaneously sending corresponding control commands to multiple control domains of the vehicle according to a determined control mode, enabling each control domain to coordinate and cooperate to create a riding environment that matches the current posture of the occupants. The autonomous driving domain controller is responsible for the vehicle's path planning and motion control, influencing acceleration, deceleration, and lane-changing behavior by adjusting trajectory planning parameters; the chassis domain controller is responsible for adjusting the suspension system, affecting the vehicle's filtering effect on road vibrations by adjusting the damping coefficient; and the cabin domain controller is responsible for adjusting the cabin environment, including multimedia volume, air conditioning mode, and window status. By synchronously sending control commands to at least two of the above domain controllers, coordinated adjustment of driving behavior, chassis response, and cabin environment is achieved.

[0029] like Figure 2 The multi-domain collaborative control timing diagram shown illustrates the collaborative control process in detail, using a resting, closed-eye posture as an example: The posture recognition module outputs a posture label (e.g., resting posture), which is then passed to the policy decision module. After mapping this label to a quiet mode, the policy decision module sends instructions in parallel to the three domain controllers, setting an impact limit of 0.5 m / s for the autonomous driving domain. 3 The system adjusts the suspension damping to the softest setting in the chassis domain and lowers the volume and switches the air conditioning to recirculation mode in the cabin domain. After execution, each domain controller returns a confirmation signal to the strategy decision module, including confirmation of trajectory optimization completion, damping adjustment completion, and environmental adjustment completion. Upon receiving confirmation from each domain, the system continues to monitor changes in occupant posture, forming a continuous closed-loop adjustment.

[0030] The autonomous driving comfort adjustment method provided in this embodiment introduces an in-cabin occupant posture perception link into the control link of the autonomous driving system, expanding the original single-dimensional control based solely on external environmental information into a dual-dimensional fusion control of the external environment and the in-vehicle occupant state. The introduction of occupant posture categories allows the control strategy to dynamically adapt to the occupant's current activity state, solving the problem in existing technologies where all occupants experience the same vehicle dynamic response indiscriminately. The design of cross-domain collaborative control commands enables simultaneous execution of driving behavior adjustment, chassis response adjustment, and cabin environment adjustment, avoiding the limited effect of single-dimensional adjustment and achieving multi-dimensional collaborative comfort enhancement. The entire adjustment process is completed automatically by the system, requiring no manual mode switching or operation by the occupant, achieving seamless intelligent comfort adjustment.

[0031] In one embodiment, performing pose recognition processing on the image data to determine the current pose category of the occupant includes: The image data is input into a pre-trained posture recognition model for processing to obtain the current posture category of the occupant; The pose recognition model includes a skeleton point recognition network module and a pose classifier module. The skeletal point recognition network module is used to identify the spatial coordinates of key skeletal points of the occupant from the image data. The key skeletal points include at least one of the following: eyes, neck, shoulders, elbows, wrists, and hips. The pose classifier module is used to output the pose category based on time series analysis and the spatial coordinates of the key skeleton points in multiple consecutive frames.

[0032] In application, the pose recognition model adopts a two-stage serial architecture, such as... Figure 3 The complete data flow for attitude recognition is shown from top to bottom: cockpit camera images serve as raw data and are fed into the model's 3D skeletal point recognition network (HRNet / OpenPose). This network extracts key skeletal point information of the occupant's body from the images and inputs it into the feature processing layer. The feature processing layer includes two cascaded processing steps: key point coordinate extraction and time series analysis (LSTM / GRU). It performs temporal modeling of the skeletal point coordinates of consecutive frames and outputs the final attitude label through the attitude classifier (based on temporal features).

[0033] Key skeletal points refer to joint positions in human anatomy that have significant kinematic importance, including the eyes (used to determine the direction of vision and eyelid status), neck (used to determine the head's orientation and tilt angle), shoulders (used to determine upper body posture), elbows (used to determine the arm's flexion), wrists (used to determine the hand's position and movement), and hips (used to determine sitting posture and the body's center of gravity). For example... Figure 4 The schematic diagram of the skeletal points shown illustrates the topological structure of key skeletal points in the human body. The skeletal point recognition network module can employ architectures such as HRNet or OpenPose, or the lightweight MobileNetV3 can be used as the backbone network to reduce computational resource consumption while ensuring recognition accuracy, thus meeting the computing power constraints of automotive embedded platforms. The skeletal point recognition network module outputs the coordinates of each key skeletal point in three-dimensional space.

[0034] The attitude classifier module, based on time series analysis methods (such as Long Short-Term Memory (LSTM) or Gated Recurrent Unit (GRU), models the spatial coordinate sequence of skeletal points across multiple consecutive frames. By analyzing the trajectories of skeletal points across these frames, transient interference can be effectively filtered out, accurately identifying the occupant's continuous posture state. The attitude classifier module ultimately outputs the occupant's posture category label and its corresponding confidence score. To ensure real-time attitude recognition, the attitude recognition model can be deployed on a processor integrated with a Neural Processing Unit (NPU). The NPU, through its parallel computing architecture, can efficiently perform forward inference operations on deep neural networks.

[0035] In application, when the highest confidence score output by the pose classifier is lower than a preset confidence threshold (e.g., 0.7), the system determines that the current pose recognition result is unreliable and will maintain the control mode already in effect in the previous cycle until the confidence score recovers to above the threshold in subsequent sampling cycles. This confidence filtering mechanism ensures the reliability of control mode switching and avoids unnecessary mode changes when recognition is uncertain.

[0036] This embodiment designs the pose recognition model as a cascaded structure of a skeletal point recognition network module and a pose classifier module, achieving end-to-end processing from the original image to the pose category. The skeletal point recognition network module transforms high-dimensional image data into low-dimensional skeletal point coordinate representations, effectively reducing the computational complexity of subsequent classification tasks. The pose classifier module models continuous frame data through time series analysis, improving the robustness and anti-interference ability of pose recognition and avoiding frequent control mode jumps caused by single-frame recognition errors. The collaborative work of the two modules enables pose recognition to achieve both high accuracy and high stability.

[0037] In one embodiment, the attitude category includes at least two of the following: The forward-looking posture is determined by the head facing the direction of vehicle travel, and the relative positional relationship between the key bone points of the neck and the key bone points of the eyes indicates that the face is facing forward. The reading and writing posture is determined by the following conditions: the head is lowered at an angle exceeding a preset angle threshold, the key bone point of the elbow is bent, and the key bone point of the wrist is located above the key bone point of the hip. The resting posture with eyes closed is determined by the following conditions: the key skeletal points of the eye show that the eyelids are closed, the head is tilted back or to the side, and the duration exceeds a preset duration threshold.

[0038] In practice, the definitions of the above three posture categories are based on ergonomic and kinesiology studies, and correspond to the three most common activity states of occupants in autonomous driving mode.

[0039] The forward-looking posture represents the occupant maintaining a normal sitting position with their gaze directed towards the vehicle's direction of travel. During the determination, the system analyzes the relative positional relationship between key cervical and ocular skeletal points. When the spatial coordinates of these points indicate that the occupant's face is directly facing forward, the posture is determined to be forward-looking. In this posture, the occupant's vestibular system can perceive vehicle movement through visual information, thus exhibiting relatively low sensitivity to acceleration, deceleration, and lateral acceleration changes, allowing the vehicle to operate according to standard control strategies.

[0040] The reading and writing posture refers to the state of an occupant when reading, writing, or using a handheld electronic device with their head down. Three conditions must be met simultaneously for determination: 1. Calculate the head drooping angle based on the relative positions of the cervical and scalp bones. When this angle exceeds a preset threshold, it indicates the head is in a drooping state; 2. The elbow flexion point is clearly defined, indicating a significant angle between the upper arm and forearm, suggesting the arm is in a holding position; 3. The spatial coordinates of the wrist flexion point are above the hip flexion point, further confirming the hand is held above the waist. Motion pathology studies have shown that when occupants look down to read, the sensory conflict between visual and vestibular perception intensifies, significantly increasing sensitivity to vehicle acceleration and deceleration, and making them more prone to motion sickness.

[0041] The resting, closed-eyes posture represents the occupant's state of rest or sleep with their eyes closed. Three conditions must be met simultaneously for determination: 1. Key skeletal features of the eye area show that the eyelids are closed; 2. The head is tilted back or to the side, indicating relaxed neck muscles; 3. The above state must last for more than a preset duration threshold (e.g., 30 seconds) to exclude non-resting behaviors such as brief blinking. During the confirmation waiting period (i.e., the period before the closed-eyes duration reaches the preset threshold), the system maintains the currently active control mode to avoid unnecessary mode switching due to instantaneous state changes. Once the closed-eyes duration reaches the preset threshold, the system confirms the current posture as a resting, closed-eyes posture and triggers a switch to the corresponding control mode. In this posture, the occupant completely loses the ability to visually predict vehicle movement and is extremely sensitive to any sudden changes in motion, thus requiring the smoothest and quietest control strategy.

[0042] In applications, in addition to the three posture categories mentioned above, the system can selectively extend support for other posture categories, such as conversational postures or eating postures. When no separate mapping rules are configured for the above optional posture categories, they are mapped to the standard mode by default to ensure that the system can still function normally in postures that are not explicitly identified.

[0043] This embodiment establishes a correspondence between posture judgment and ergonomic principles by precisely defining posture categories based on skeletal point features. The three posture categories—looking forward, reading / writing, and resting with eyes closed—correspond to vehicle dynamic sensitivity levels from low to high, providing a scientific basis for the subsequent differentiation of control modes. Simultaneously, the duration threshold design for the resting, eyes-closed posture effectively avoids misjudgments caused by transient behaviors such as brief blinks, ensuring the reliability of posture recognition results.

[0044] In one embodiment, the control mode includes at least one of a standard mode, a smooth mode, and a quiet mode; The step of determining the control mode corresponding to the posture category includes: When the posture category is a forward-looking posture, the control mode is determined to be the standard mode; When the posture category is reading and writing posture, the control mode is determined to be smooth mode; When the posture category is a resting posture with eyes closed, the control mode is determined to be the quiet mode.

[0045] In application, the above three control modes are control parameter schemes designed specifically for the differences in the sensitivity of occupants to the vehicle's dynamic response under different postures. Their correspondence follows the principle that the higher the sensitivity, the gentler the control.

[0046] The standard mode is the control scheme used when the occupants are looking straight ahead. In this mode, the vehicle operates according to conventional control parameters, with impact limits, lane change frequency, and suspension damping all maintained at normal levels, and the suspension damping set to medium stiffness. Because the occupants are looking straight ahead, their visual system can perceive and predict changes in vehicle movement in real time, and the sensory consistency between the vestibular and visual systems is good. Therefore, the vehicle's dynamic response in standard mode does not cause significant discomfort to the occupants.

[0047] The Smooth Mode is the control scheme employed when occupants are in a reading or writing posture. In this mode, route planning prioritizes straight-line travel, reducing unnecessary lane changes, and acceleration changes are smoother, using a trapezoidal acceleration curve instead of a step curve. Simultaneously, suspension damping is appropriately reduced to enhance the filtering effect on road vibrations. By suppressing abrupt changes in vehicle motion, Smooth Mode reduces the probability of motion sickness caused by visual-vestibular sensory conflict.

[0048] The Quiet Mode is a control scheme used when occupants are in a resting, closed-eye position. In this mode, the vehicle's motion control is at its gentlest level, the suspension damping is adjusted to the softest setting to filter road vibrations to the maximum extent, and the cabin environment is adjusted, including lowering the multimedia volume, changing the music type, closing the windows, switching the air conditioning to recirculation mode and lowering the fan speed, etc., to create a quiet and stable resting environment in all aspects.

[0049] This embodiment establishes a one-to-one mapping between posture categories and control modes, enabling the system to automatically select the most suitable control scheme based on the occupant's real-time posture. The three control modes progressively enhance comfort from standard mode to quiet mode, precisely matching the progressively increasing sensitivity of the occupant as they move from looking straight ahead to resting with their eyes closed. This achieves an adaptive adjustment logic where the more sensitive the perception, the gentler the control.

[0050] In one embodiment, the standard mode, the smooth mode, and the quiet mode each correspond to different upper limit constraints on impact. The impact limit of the smooth mode is lower than that of the standard mode, and the impact limit of the quiet mode is lower than that of the smooth mode. When in the smooth mode or the quiet mode, the weight of the impact constraint term is increased and the weight of the time efficiency term is decreased in the cost function of trajectory planning.

[0051] In applications, the impact (Jerk) is the derivative of acceleration with respect to time, i.e. Its physical meaning represents the rate at which a vehicle's acceleration changes. By limiting the upper limit of the impact force, the abrupt changes in vehicle motion can be suppressed, improving ride smoothness.

[0052] Specifically, the upper limit constraint of the impact intensity for the three control modes can be set as follows: Upper limit of impact J in standard mode max,std It can be set to 2.0 m / s 3 This value allows for normal acceleration variations during normal driving, meeting the efficiency requirements of daily driving; the upper limit of the impact J in smooth mode. max,smooth It can be set to 1.0 m / s 3 This reduces the abruptness of acceleration changes to half that of standard mode, making acceleration and braking processes more gradual; the maximum impact J in Quiet Mode... max,sleep It can be set to 0.5m / s 3 This further controls acceleration changes within an extremely gentle range, minimizing disturbance to resting occupants.

[0053] In trajectory planning algorithms, the cost function J cost The general form is: J cost =w1 × (Time Efficiency) + w2 × (Sum of Squared Impacts) + w3 × (Distance from Centerline). Where w1 is the weight of the time efficiency term, w2 is the weight of the impact constraint term, and w3 is the weight of the distance from centerline term. When the system is in smooth or quiet mode, the weight w2 of the impact constraint term in the cost function is increased, while the weight w1 of the time efficiency term is decreased. The trajectory planning algorithm, in its optimization solution, will focus more on reducing the drastic changes in acceleration, while appropriately relaxing the pursuit of travel time efficiency. Through this weight adjustment, the generated travel trajectory will exhibit a smoother speed and acceleration change curve, prioritizing comfort over efficiency.

[0054] This embodiment introduces the impact factor as a control variable into the parameter definitions of different control modes. By setting differentiated upper limit constraints on the impact factor and dynamically adjusting the weights of the cost function, it achieves refined control of the vehicle's dynamic response from the source of trajectory planning. The impact factor constraint directly acts on the rate of change of acceleration, rather than just constraining the acceleration itself, thus suppressing the jerking sensation that causes discomfort to passengers more deeply. The dynamic adjustment of the cost function weights allows the trajectory planning algorithm to adaptively balance comfort and efficiency according to the control mode, without needing to design a separate planning algorithm for each mode, reducing the system's implementation complexity.

[0055] In one embodiment, before generating cross-domain collaborative control instructions according to the control mode, the method further includes: The system acquires planning data from an autonomous driving system and extracts a sequence of dynamic parameters for a future preset time domain from the planning data; wherein the sequence of dynamic parameters includes at least two of the following: target velocity, target acceleration, and target impact. Simultaneously acquire external environmental information; wherein, the external environmental information includes at least one of the following: road curvature ahead, target vehicle distance, traffic light status, and speed limit information.

[0056] In application, before generating cross-domain cooperative control commands, the system needs to comprehensively acquire planning data from the autonomous driving system and external environmental information to ensure that comfort adjustment commands are coordinated with the current driving state and external environment.

[0057] For acquiring planning data, the system reads the planning data output by the autonomous driving domain controller in real time via the vehicle bus (such as CAN bus or FlexRay bus) and extracts a sequence of dynamic parameters for a future preset time domain (e.g., the next 5 seconds). In this dynamic parameter sequence, the target speed v(t) represents the vehicle's expected speed at each future moment, the target acceleration a(t) represents the vehicle's expected acceleration at each future moment, and the target impact... It characterizes the drasticness of acceleration changes. By previewing the dynamic parameter sequence in the future time domain, the system can anticipate the vehicle's upcoming motion behavior and prepare corresponding comfort adjustments before the motion changes occur.

[0058] For the synchronous acquisition of external environmental information, the system obtains information such as the curvature of the road ahead, the distance to the target vehicle, the status of traffic lights, and speed limits from the perception and mapping modules of the autonomous driving system. This information serves as a safety constraint in the final decision of the control mode: when the external environmental information indicates that the current driving scenario is high-risk (e.g., the curvature of the road ahead exceeds a preset curvature threshold, the distance to the target vehicle is less than a preset safe distance threshold, or the traffic light status is about to change), regardless of the occupant's current posture, the system will force the control mode to be constrained to the standard mode or let the autonomous driving safety strategy take over, to ensure that driving safety takes precedence over comfort adjustment.

[0059] This embodiment provides a complete decision-making basis for the generation of cross-domain collaborative control commands by synchronously acquiring planning data and external environmental information before generating control commands. The pre-reading of planning data enables the system to proactively adjust comfort settings, while the introduction of external environmental information sets safety constraints for comfort adjustments, ensuring that driving safety always takes precedence over comfort goals under any circumstances.

[0060] In one embodiment, distributing the cross-domain cooperative control command to at least two of the autonomous driving domain controller, chassis domain controller, and cockpit domain controller includes at least two of the following: The impact constraint parameters are sent to the autonomous driving domain controller so that the autonomous driving domain controller can re-optimize the driving trajectory based on the impact constraint parameters; Send a suspension damping adjustment command to the chassis domain controller to adjust the damping coefficient of the suspension system according to the control mode; Send an environmental adjustment command to the cockpit domain controller to adjust at least one environmental parameter in the cockpit; the environmental parameter includes at least one of multimedia volume, music type, window status, air conditioning circulation mode, and fan speed.

[0061] In application, the specific execution content of cross-domain collaborative control commands is as follows: For the autonomous driving domain controller, the system incorporates the impact constraint parameters corresponding to the current control mode into the cost function of the trajectory planning algorithm, thereby re-optimizing the driving trajectory. Changes in the impact constraint parameters directly affect the shape of the velocity and acceleration curves generated by the planning algorithm, enabling it to seek the optimal trajectory while satisfying the upper limit constraint of the impact. The re-optimized driving trajectory will exhibit dynamic characteristics that match the current control mode.

[0062] For the chassis domain controller, the system sends damping adjustment commands to the CDC (Continuous Damping Control) suspension or air suspension system via CAN signals, adjusting the damping coefficient of the suspension system according to the current control mode. Different control modes correspond to different damping coefficient settings. For example, in standard mode, the damping coefficient K=0.7 provides relatively solid chassis support; in smooth mode, the damping coefficient K=0.5 strikes a balance between support and comfort; in quiet mode, the damping coefficient K=0.3 uses the softest damping setting to filter road vibrations to the occupants to the greatest extent. Optionally, in quiet mode and at vehicle speeds exceeding 80 km / h, the vehicle height can be further lowered by 10 mm to improve high-speed stability. When the vehicle height is lowered, the suspension damping coefficient increases accordingly from K=0.3 to no less than K=0.4 to ensure sufficient support stiffness after lowering the vehicle and avoid the risk of chassis bottoming out due to excessively soft damping combined with a low vehicle height; when the vehicle height returns to normal, the damping coefficient simultaneously returns to K=0.3.

[0063] For the cabin domain controller, the system sends environmental adjustment commands to the in-vehicle infotainment system, air conditioning controller, and window controller, adjusting the environmental parameters within the cabin according to the current control mode. Specifically: In Quiet mode, the system automatically reduces the multimedia volume to a preset value (e.g., 50%). If it detects loud music playing, it automatically switches to soft music or white noise. Simultaneously, it automatically closes the windows (if not already closed), switches the air conditioning to recirculation mode to reduce outside noise entering the cabin, and lowers the fan speed to reduce wind noise. In Smooth mode, it maintains the current volume setting, but if it detects an occupant in a reading posture, it automatically brightens the reading lights to provide a suitable lighting environment. In Standard mode, no additional environmental adjustments are performed, or default settings are restored.

[0064] This embodiment covers three dimensions: trajectory re-optimization in the autonomous driving domain, suspension damping adjustment in the chassis domain, and environmental parameter adjustment in the cabin domain. Control commands in the three domains are generated and executed synchronously around the same control mode label, forming a comprehensive and coordinated adjustment system from vehicle motion control to passenger comfort and cabin environment, enabling the vehicle to provide a highly unified and coordinated comfort experience under different control modes.

[0065] In one embodiment, it also includes: The system continuously monitors the occupant's posture category and dynamically updates the control mode and corresponding cross-domain collaborative control commands in response to changes in the posture category.

[0066] In application, after completing the initial attitude recognition and control mode setting, the system continuously monitors the occupant's attitude in real time at a sampling frequency of no less than 1Hz. When the system detects a change in the occupant's attitude category (e.g., the occupant changes from a reading posture to a resting posture, or from a resting posture to a forward-looking posture), the system immediately triggers the entire process of attitude-control mode mapping and control command generation, redetermines the control mode corresponding to the new attitude, and updates the control commands of each domain controller.

[0067] In application, under any control mode, when the safety layer of the autonomous driving system triggers a safety intervention request (such as AEB automatic emergency braking, FCW forward collision warning, or ESC electronic stability control), the comfort adjustment command is immediately overridden by the safety command, the impact constraint no longer applies, and the system responds to the safety event with maximum braking capacity or optimal obstacle avoidance trajectory. After the safety event ends, the system rereads the current occupant posture and restores the corresponding comfort control mode.

[0068] In application, the system can also continuously monitor the stability indicators of the occupant's posture within a preset evaluation window (e.g., 60 seconds) after the control mode switch is completed. The stability indicators include the frequency of posture category changes and the amplitude of skeletal point jitter. If the occupant's posture remains stable within the evaluation window (i.e., the posture category does not change and the amplitude of skeletal point jitter is below a preset threshold), the current control mode is deemed effective. If the occupant frequently adjusts their sitting posture within a short period after the switch (i.e., the amplitude of skeletal point jitter exceeds the preset threshold), the system will further reduce the upper limit of impact based on the current mode and perform a secondary fine-tuning to more precisely adapt to the occupant's comfort needs.

[0069] This embodiment introduces a continuous monitoring and dynamic update mechanism, enabling the comfort adjustment system to form a complete closed-loop control circuit. The system can track changes in occupant posture in real time and respond quickly, while safety priority constraints ensure that comfort adjustments will not affect driving safety under any circumstances. The comfort effect evaluation and secondary fine-tuning mechanism further enhances the system's adaptability, allowing the adjustment of the control mode not only to be based on coarse-grained mapping of posture categories but also to be fine-grained optimized based on actual occupant comfort feedback.

[0070] In one embodiment, a gradual transition is used during the switching of the control mode, the gradual transition including: The impact limit value is linearly transitioned to the first target value within a first preset transition time, and the suspension damping coefficient is gradually changed to the second target value within a second preset transition time.

[0071] In applications, when the system needs to switch control modes due to changes in occupant posture, abruptly changing each control parameter from its current value to the target value could cause discomfort to the occupants. Therefore, the system employs a gradual transition strategy during control mode switching, ensuring a smooth transition of each control parameter from its current value to the target value.

[0072] Specifically, the gradual transition involves the smooth switching of two main control parameters: the impact limit value undergoes a linear transition within a first preset transition time (e.g., 5 seconds), meaning the impact limit value changes from its current value to the target value at a constant speed during the transition time, causing the constraints of the trajectory planning algorithm to change gradually and avoiding sudden changes in the driving trajectory. The suspension damping coefficient undergoes a gradual adjustment within a second preset transition time (e.g., 3 seconds), meaning the damping coefficient gradually changes to the target value during the transition time, allowing the response characteristics of the suspension system to transition smoothly and avoiding changes in vehicle posture caused by abrupt changes in damping.

[0073] In application, the first preset transition time and the second preset transition time can be set to the same value or different values. Preferably, since the change in the impact amplitude value indirectly affects the vehicle motion through the trajectory planning algorithm, its effect transmission has a certain delay, requiring a longer transition time to ensure smoothness; while the change in the suspension damping coefficient directly affects the suspension system, its effect response is more immediate, and the transition time can be relatively shorter. Therefore, the first preset transition time and the second preset transition time are set to different values.

[0074] This embodiment eliminates the secondary discomfort that may be caused by sudden parameter changes by introducing a gradual transition strategy during control mode switching. The linear transition of the impact limit value ensures the continuous change of trajectory planning constraints, while the gradual adjustment of the suspension damping coefficient ensures a smooth transition of chassis response. The synergistic effect of the two ensures that the control mode switching process maintains a high level of comfort.

[0075] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0076] This application also provides a vehicle, including: Image acquisition devices deployed inside the cockpit are used to collect image data of the occupants in real time; A processor communicatively connected to the image acquisition device is configured to: perform attitude recognition processing on the image data to determine the current attitude category of the occupant, determine the corresponding control mode based on the attitude category, and generate cross-domain collaborative control instructions based on the control mode; At least two of the autonomous driving domain controller, chassis domain controller, and cockpit domain controller, which are respectively communicatively connected to the processor, are used to receive and execute the cross-domain cooperative control instructions.

[0077] In this application, the image acquisition device is a camera system deployed inside the vehicle cabin. Its installation location needs to cover the driver and passenger areas to capture images of occupants in various positions within the cabin. The image acquisition device transmits the acquired image data to the processor in real time via an in-vehicle communication interface (such as Ethernet or LVDS).

[0078] The processor is the core computing unit in the vehicle responsible for attitude recognition and control decision-making. It receives image data from the image acquisition device, runs an attitude recognition algorithm to determine the occupant's attitude category, determines the control mode based on the attitude category, and generates cross-domain cooperative control commands. The processor establishes communication connections with each domain controller through the vehicle communication bus and distributes the generated control commands to each domain controller for execution.

[0079] like Figure 5 As shown, the overall system architecture includes an image acquisition module, a posture recognition module, a strategy decision module, and an instruction execution module connected to the autonomous driving domain interface, chassis domain interface, and cockpit domain interface, respectively. In the diagram, the image acquisition module acquires occupant images and transmits the data down to the posture recognition module. The posture recognition module performs skeletal point extraction and posture classification on the image data and then passes the posture labels to the strategy decision module located in the middle layer of the architecture. The strategy decision module simultaneously receives planning data from the autonomous driving domain controller as auxiliary decision input, determines the control mode according to built-in mapping rules, and generates cross-domain collaborative control instructions. The generated instructions are sent out through the three domain interfaces: the autonomous driving domain interface connects to the trajectory planning algorithm, transmitting impact constraint parameters; the chassis domain interface connects to the suspension controller (CDC / air suspension), transmitting damping adjustment instructions; and the cockpit domain interface connects to the infotainment system, air conditioning controller, and window controller, transmitting environmental adjustment instructions. The autonomous driving domain controller is responsible for vehicle path planning and motion control, receives impact constraint parameters sent by the processor, and re-optimizes the driving trajectory accordingly. The chassis domain controller is responsible for adjusting the parameters of the suspension system, receiving suspension damping adjustment commands from the processor, and controlling the damping coefficient of the suspension system. The cockpit domain controller is responsible for the comprehensive adjustment of the cockpit environment, receiving environmental adjustment commands from the processor, and controlling the operating status of cockpit equipment such as the multimedia system, air conditioning system, and window system. The processor and the collaborative work of at least two of the above domain controllers form a complete control link from perception to decision-making to execution.

[0080] In one embodiment, the image acquisition device includes at least one driver monitoring camera facing the driver's position and at least one passenger monitoring camera facing the passenger's position; The processor integrates a neural network acceleration chip for running the attitude recognition model; The autonomous driving domain controller is used to re-optimize the driving trajectory based on the impact constraint parameters in the cross-domain cooperative control command. The chassis domain controller is used to adjust the damping coefficient of the suspension system according to the cross-domain collaborative control command; The cockpit domain controller is used to adjust the environmental parameters within the cockpit according to the cross-domain collaborative control commands.

[0081] In application, the Driver Monitoring Camera (DMS camera) is mounted above the steering wheel or in the dashboard area, facing the driver, to capture facial and upper body image data of the driver. The Passenger Monitoring Camera (OMS camera) is mounted near the headliner or rearview mirror, facing the passenger area, to capture image data of front and rear passengers. Both the DMS and OMS cameras have infrared illumination capabilities, enabling them to acquire clear occupant images even at night or in low-light conditions inside the vehicle through active infrared illumination, ensuring the all-weather operation of the posture recognition system.

[0082] The integrated Neural Processing Unit (NPU) chip in the processor is a hardware acceleration unit specifically optimized for deep learning inference tasks. Through its parallel computing architecture and dedicated instruction set, the NPU can efficiently execute forward inference operations for deep neural network models such as skeleton point recognition networks and pose classifiers, reducing power consumption while meeting real-time requirements. The pose recognition model is deployed on the NPU, enabling the processor to complete high-frame-rate pose recognition processing without affecting other vehicle control tasks.

[0083] Upon receiving the impact constraint parameters, the autonomous driving domain controller incorporates these parameters into the cost function of its internal trajectory planning algorithm, re-optimizing the driving trajectory within the current planning cycle. The optimized trajectory, while meeting safety constraints and traffic rules, ensures that the rate of change of acceleration does not exceed the impact limit set by the current control mode. The chassis domain controller communicates with the CDC suspension or air suspension system via CAN signals, adjusting the damping coefficients of each suspension damper in real time according to received damping adjustment commands, thereby altering the suspension system's response characteristics to road surface excitations. The cockpit domain controller communicates with the in-vehicle infotainment system, air conditioning controller, and window controllers via Ethernet or LIN bus, coordinating and controlling the operating parameters of each cockpit device according to received environmental adjustment commands.

[0084] The configuration of DMS and OMS cameras enables full coverage acquisition of the driver and passengers, while infrared illumination ensures all-weather operation. The integration of the NPU addresses the real-time operation requirements of deep learning models on the in-vehicle embedded platform. The synergistic cooperation of these hardware components allows the comfort adjustment method of this application to be fully and reliably deployed and operated on a real vehicle platform.

[0085] It should be noted that the information interaction and execution process between the processor, image acquisition device and each domain controller in the above-mentioned vehicle are based on the same concept as the method embodiment of this application. For details on their specific functions and the resulting technical effects, please refer to the method embodiment section, which will not be repeated here.

[0086] It should be noted that the collection, processing, and use of sensitive data such as personal information and biometric information involved in this application include, but are not limited to: collecting information based on the user's explicit consent, processing data using anonymization or de-identification technologies, collecting necessary data in accordance with the "minimum necessary" principle, ensuring data security through measures such as encrypted transmission, secure storage, and access control, and does not involve the illegal sale, disclosure, or misuse of personal information. It fully complies with the legal obligations of personal information processors and does not infringe upon user rights or the public interest.

[0087] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for adjusting the comfort of autonomous driving, characterized in that, include: Real-time acquisition of image data of occupants inside the vehicle cabin; The image data is subjected to posture recognition processing to determine the current posture category of the occupant; Based on the attitude category, determine the control mode corresponding to the attitude category; Based on the control mode, cross-domain collaborative control instructions are generated and distributed to at least two of the autonomous driving domain controller, chassis domain controller, and cockpit domain controller for execution.

2. The autonomous driving comfort adjustment method as described in claim 1, characterized in that, The step of performing pose recognition processing on the image data to determine the current pose category of the occupant includes: The image data is input into a pre-trained posture recognition model for processing to obtain the current posture category of the occupant; The pose recognition model includes a skeleton point recognition network module and a pose classifier module. The skeletal point recognition network module is used to identify the spatial coordinates of key skeletal points of the occupant from the image data. The key skeletal points include at least one of the following: eyes, neck, shoulders, elbows, wrists, and hips. The pose classifier module is used to output the pose category based on time series analysis and the spatial coordinates of the key skeleton points in multiple consecutive frames.

3. The autonomous driving comfort adjustment method as described in claim 2, characterized in that, The attitude categories include at least two of the following: The forward-looking posture is determined by the head facing the direction of vehicle travel, and the relative positional relationship between the key skeletal points of the neck and the key skeletal points of the eyes indicates that the face is facing forward. The reading and writing posture is determined by the following conditions: the head is lowered at an angle exceeding a preset angle threshold, the key bone point of the elbow is bent, and the key bone point of the wrist is located above the key bone point of the hip. The resting posture with eyes closed is determined by the following conditions: the key skeletal points of the eye show that the eyelids are closed, the head is tilted back or to the side, and the duration exceeds a preset duration threshold.

4. The autonomous driving comfort adjustment method as described in claim 3, characterized in that, The control mode includes at least one of standard mode, smooth mode and quiet mode; The step of determining the control mode corresponding to the posture category includes: When the posture category is a forward-looking posture, the control mode is determined to be the standard mode; When the posture category is reading and writing posture, the control mode is determined to be smooth mode; When the posture category is a resting posture with eyes closed, the control mode is determined to be the quiet mode.

5. The autonomous driving comfort adjustment method as described in claim 4, characterized in that, The standard mode, the smooth mode, and the quiet mode each correspond to different upper limit constraints on impact. The impact limit of the smooth mode is lower than that of the standard mode, and the impact limit of the quiet mode is lower than that of the smooth mode. When in the smooth mode or the quiet mode, the weight of the impact constraint term is increased and the weight of the time efficiency term is decreased in the cost function of trajectory planning.

6. The autonomous driving comfort adjustment method as described in claim 1, characterized in that, Before generating cross-domain collaborative control instructions based on the control mode, the method further includes: The system acquires planning data from an autonomous driving system and extracts a sequence of dynamic parameters for a future preset time domain from the planning data; wherein the sequence of dynamic parameters includes at least two of the following: target velocity, target acceleration, and target impact. Simultaneously acquire external environmental information; wherein, the external environmental information includes at least one of the following: road curvature ahead, target vehicle distance, traffic light status, and speed limit information.

7. The autonomous driving comfort adjustment method as described in claim 1, characterized in that, The step of distributing the cross-domain collaborative control command to at least two of the autonomous driving domain controller, chassis domain controller, and cockpit domain controller includes at least two of the following: The impact constraint parameters are sent to the autonomous driving domain controller so that the autonomous driving domain controller can re-optimize the driving trajectory based on the impact constraint parameters; Send a suspension damping adjustment command to the chassis domain controller to adjust the damping coefficient of the suspension system according to the control mode; Send an environmental adjustment command to the cockpit domain controller to adjust at least one environmental parameter in the cockpit; the environmental parameter includes at least one of multimedia volume, music type, window status, air conditioning circulation mode, and fan speed.

8. The autonomous driving comfort adjustment method as described in claim 7, characterized in that, Also includes: The system continuously monitors the occupant's posture category and dynamically updates the control mode and corresponding cross-domain collaborative control commands in response to changes in the posture category.

9. The autonomous driving comfort adjustment method as described in claim 8, characterized in that, A gradual transition is used during the switching of the control mode, and the gradual transition includes: The impact limit value is linearly transitioned to the first target value within a first preset transition time, and the suspension damping coefficient is gradually changed to the second target value within a second preset transition time.

10. A vehicle, characterized in that, include: Image acquisition devices deployed inside the cockpit are used to collect image data of the occupants in real time; A processor communicatively connected to the image acquisition device is configured to: perform attitude recognition processing on the image data to determine the current attitude category of the occupant, determine the corresponding control mode based on the attitude category, and generate cross-domain collaborative control instructions based on the control mode; At least two of the autonomous driving domain controller, chassis domain controller, and cockpit domain controller, which are respectively communicatively connected to the processor, are used to receive and execute the cross-domain cooperative control instructions.

Images