Vehicle cabin evaluation method and device, vehicle and storage medium

Abstract

The invention provides a vehicle cabin evaluation method and device, a vehicle and a storage medium, and the method comprises the steps: firstly obtaining a to-be-evaluated decision which is outputted by a vehicle cabin for a preset function in a current scene and is used for adjusting a control parameter of the vehicle cabin; then respectively calculating a first dimension score for evaluating the matching degree between the decision to be evaluated and an expected decision of a user for a preset function, a second dimension score for measuring the incremental learning efficiency of a decision output model in the vehicle cabin, and a third dimension score for evaluating the anti-interference capability of the vehicle cabin; and finally, generating a comprehensive score of the vehicle cabin by integrating the scores of the three dimensions. According to the method, the dynamic adaptability of the vehicle cabin can be evaluated from different dimensions, a basis is provided for research and development personnel to carry out targeted optimization on the vehicle cabin, and the intelligence and the dynamic adaptability of the vehicle cabin can be improved.

Description

Technical Field

[0001] The present application relates to the field of vehicles, and more particularly, to a method, device, vehicle, and storage medium for evaluating a vehicle cabin in the field of vehicles. Background Art

[0002] With the advancement of intelligent vehicles, the smart cockpit has become a key hub for human-vehicle interaction. Building on traditional in-car entertainment and navigation capabilities, the smart cockpit integrates voice interaction, personalized recommendations, driver assistance, and other features to meet users' higher demands for comfort and safety.

[0003] However, the current industry evaluation of smart cockpits still mainly focuses on offline functional accuracy and static scenario testing, and lacks evaluation of the dynamic adaptability of smart cockpits. Summary of the Invention

[0004] This application provides a vehicle cockpit evaluation method, device, vehicle and storage medium. The method can evaluate the dynamic adaptability of the vehicle cockpit from different dimensions, providing a basis for R&D personnel to perform targeted optimization of the vehicle cockpit, and helping to improve the intelligence and dynamic adaptability of the vehicle cockpit.

[0005] In a first aspect, a method for evaluating a vehicle cockpit is provided, the method comprising: obtaining a decision to be evaluated output by the vehicle cockpit for a preset function in a current scenario; wherein the decision to be evaluated is used to adjust control parameters of the preset function; based on the decision to be evaluated, calculating a first dimension score for evaluating the vehicle cockpit, and obtaining a second dimension score and a third dimension score for evaluating the vehicle cockpit; wherein the first dimension score is used to evaluate the degree of match between the decision to be evaluated and the user's expected decision for the preset function; the second dimension score is used to evaluate the incremental learning efficiency of the decision output model in the vehicle cockpit; the third dimension score is used to evaluate the anti-interference ability of the vehicle cockpit; and generating a comprehensive score for the vehicle cockpit based on the first dimension score, the second dimension score and the third dimension score.

[0006] The above technical solution obtains the decision output by the vehicle cockpit for a preset function in the current scenario and, based on the decision, calculates a first-dimension score to evaluate the degree of match between the decision output and the user's desired decision for the preset function. This accurately measures user satisfaction with certain function adjustment decisions in the vehicle cockpit, facilitating subsequent optimization of the vehicle cockpit's decision output capabilities based on the first-dimension score. It also obtains a second-dimension score to evaluate the incremental learning efficiency of the vehicle cockpit's decision output model and a third-dimension score to evaluate the vehicle cockpit's anti-interference ability. Evaluating the incremental learning efficiency of the decision output model helps measure the vehicle cockpit's ability to quickly adapt to different user habits and new scenario changes. Evaluating the vehicle cockpit's anti-interference ability effectively assesses the vehicle cockpit's ability to maintain normal functionality and stable decision output in the face of complex environmental factors or other abnormal situations. A comprehensive score is generated based on the first, second, and third-dimension scores. This comprehensive score fully reflects the dynamic adaptability of the vehicle cockpit and provides a basis for R&D personnel to conduct targeted optimization of the vehicle cockpit, helping to improve the vehicle cockpit's intelligence and dynamic adaptability.

[0007] In combination with the first aspect, in certain implementations of the first aspect, a first dimension score for evaluating a vehicle cabin is calculated based on the decision to be evaluated, including: processing the decision to be evaluated to obtain a decision vector to be evaluated; obtaining the user's expected decision for a preset function, and processing the expected decision to obtain an expected decision vector; obtaining a preset deviation allowed when adjusting the control parameters of the preset function based on the decision to be evaluated; and calculating the first dimension score based on the decision vector to be evaluated, the expected decision vector, and the preset deviation.

[0008] The above technical solution processes the decision to be evaluated to obtain a decision vector for evaluation, and processes the expected decision to obtain an expected decision vector. This quantifies the decision, facilitating subsequent precise assessment of the gap between the decision to be evaluated and the user's expected decision for a preset function. Calculating a first-dimension score based on the decision vector to be evaluated, the expected decision vector, and the preset deviation converts the vehicle cockpit's decision-making performance into a specific numerical value, intuitively reflecting the degree of match between the decision to be evaluated and the user's expected decision for the preset function. Furthermore, the calculated first-dimension score provides the vehicle cockpit with clear feedback tailored to user needs, facilitating subsequent targeted optimization and improvement of the decision output model. This helps optimize and improve the vehicle cockpit's output decisions, thereby increasing the accuracy and effectiveness of the output decisions and, in turn, enhancing the overall performance and user satisfaction of the vehicle cockpit.

[0009] In combination with the first aspect and the above-mentioned implementation methods, in certain implementation methods of the first aspect, obtaining the user's expected decision for a preset function includes: after the vehicle cockpit executes the decision to be evaluated, obtaining the interaction data between the user in the vehicle cockpit and the vehicle cockpit in response to the decision to be evaluated; based on the interaction data, determining the user's expected decision for the preset function.

[0010] This technical solution captures user interaction data with the vehicle cockpit after the vehicle cockpit executes the decision to be evaluated. This interaction data directly reflects the user's actual reaction and perception of the decision. Determining the desired decision based on this interaction data helps provide an accurate basis for subsequent evaluation of the vehicle cockpit's output of the decision to be evaluated, providing data support for vehicle cockpit optimization.

[0011] In combination with the first aspect and the above-mentioned implementation methods, in certain implementation methods of the first aspect, obtaining a second dimension score for evaluating a vehicle cockpit includes: obtaining interactive command data, environmental data, and functional feedback data in the vehicle cockpit; training a decision output model based on the interactive command data, environmental data, and functional feedback data, and storing the trained decision output model in an on-board processing unit in the vehicle cockpit; obtaining the actual learning rate of the decision output model stored in the on-board processing unit in the vehicle cockpit for the user's expected decision on a preset function during the training process, and obtaining the number of training rounds after which the decision output model training is completed; and calculating the second dimension score based on the built-in scoring calculation algorithm of the on-board processing unit and combining the actual learning rate and the number of training rounds.

[0012] The above technical solution, by obtaining the actual learning rate of the decision output model stored in the vehicle's onboard processing unit during training for the user's desired decision for a preset function and the number of training rounds at the completion of training, can intuitively assess the learning progress of the decision output model during training. The actual learning rate reflects how quickly the decision output model absorbs knowledge, while the number of training rounds indicates the depth of the learning process. The combination of the two accurately measures the incremental learning efficiency of the decision output model, providing a powerful basis for evaluating vehicle cockpits based on this dimension. Calculating a second dimension score based on the actual learning rate and the number of training rounds quantifies the incremental learning efficiency of the decision output model. This quantification provides a comparable and analyzable numerical metric for vehicle cockpit evaluation. When comparing the performance of different vehicle cockpits or different configurations of the same cockpit, this second dimension score clearly demonstrates the differences, facilitating targeted analysis of the decision output model's performance and subsequent optimization to improve vehicle cockpit performance.

[0013] In combination with the first aspect and the above-mentioned implementation methods, in certain implementation methods of the first aspect, obtaining a third dimension score for evaluating a vehicle cabin includes: obtaining multiple sets of sample environmental data in the vehicle cabin when no interference conditions are applied, and processing each set of sample environmental data in the multiple sets of sample environmental data separately to obtain multiple sets of sample environmental vectors; obtaining actual environmental data in the vehicle cabin after interference conditions are applied, and processing the actual environmental data to obtain actual environmental vectors; and calculating the third dimension score based on the multiple sets of sample environmental vectors and the actual environmental vectors.

[0014] The above technical solution, by obtaining multiple sets of sample environmental data in the vehicle cockpit when no interference conditions are applied, can fully understand the environmental status of the vehicle cockpit under different times, locations, working conditions, etc., and the multiple sets of sample environmental data can also more realistically reflect the environmental characteristics of the cockpit under normal conditions, avoiding the one-sidedness caused by a single data, and can provide a rich and accurate data basis for the subsequent evaluation of the vehicle cockpit. By obtaining the actual environmental data in the vehicle cockpit after the interference conditions are applied, the actual situation of the vehicle cockpit after the interference is applied can be known. In addition, the multiple sets of sample environmental data and the actual environmental data can be processed separately to obtain multiple sets of sample environmental vectors and actual environmental vectors, and the data can be vectorized to facilitate the subsequent measurement of the gap between the sample environmental data and the actual environmental data. Finally, based on the multiple sets of sample environmental vectors and the actual environmental vectors, the third dimension score is calculated. The anti-interference ability of the vehicle cockpit can be intuitively understood through the third dimension score, so that the vehicle cockpit can be reasonably optimized.

[0015] In combination with the first aspect and the above-mentioned implementation methods, in some implementation methods of the first aspect, a third dimension score is calculated based on multiple groups of sample environment vectors and actual environment vectors, including: calculating a sample mean vector based on multiple groups of sample environment vectors; calculating the difference between the sample mean vector and the actual environment vector to obtain a difference vector; calculating a sample covariance matrix based on the sample mean vector; and calculating a third dimension score based on the difference vector and the sample covariance matrix.

[0016] The above technical solution calculates the sample mean vector to obtain the environmental state representing the normal state of the vehicle cabin. By calculating the difference vector between the sample mean vector and the actual environmental vector, it is possible to clearly determine the deviation of the actual cabin environmental state from the normal state in various dimensions after the interference condition is applied. By calculating the sample covariance matrix, the data distribution characteristics of multiple groups of sample environmental vectors can be measured, thereby more reasonably assessing the degree of deviation of the actual environmental data from the sample environmental distribution. Finally, based on the difference vector and the sample covariance matrix, a third-dimensional score measuring the vehicle cabin's anti-interference ability can be more reasonably calculated. This helps to understand the vehicle cabin's anti-interference ability during testing, provides a data basis for subsequent optimization of the vehicle cabin, and helps improve the overall performance of the vehicle cabin.

[0017] In combination with the first aspect and the above-mentioned implementation methods, in some implementation methods of the first aspect, a comprehensive score for the vehicle cabin is generated based on the first dimension score, the second dimension score and the third dimension score, including: weighted fusion of the first dimension score, the second dimension score and the third dimension score according to preset weights to generate a comprehensive score for the vehicle cabin.

[0018] In this technical solution, the first, second, and third dimension scores evaluate vehicle cabin characteristics from different perspectives. Through weighted fusion, this scattered information is integrated, avoiding the one-sidedness of single-dimensional evaluations and providing a comprehensive picture of the vehicle cabin's overall condition. Furthermore, by integrating these multi-dimensional scores into a single comprehensive score, a unified, quantitative standard is provided for evaluating different vehicle cabins. Whether comparing cabins of different models or evaluating the cabins of the same model with different configurations, the comprehensive score intuitively demonstrates the differences, facilitating horizontal comparisons across models and solutions, and providing a more intuitive basis for evaluating vehicle cabin performance.

[0019] In a second aspect, a vehicle cockpit evaluation device is provided, which includes: an acquisition module for acquiring a decision to be evaluated output by the vehicle cockpit for a preset function in a current scenario; wherein the decision to be evaluated is used to adjust the control parameters of the preset function; a calculation module for calculating a first dimension score for evaluating the vehicle cockpit based on the decision to be evaluated, and acquiring a second dimension score and a third dimension score for evaluating the vehicle cockpit; wherein the first dimension score is used to evaluate the degree of match between the decision to be evaluated and the user's expected decision for the preset function; the second dimension score is used to evaluate the incremental learning efficiency of the decision output model in the vehicle cockpit; the third dimension score is used to evaluate the anti-interference ability of the vehicle cockpit; and a generation module for generating a comprehensive score for the vehicle cockpit based on the first dimension score, the second dimension score and the third dimension score.

[0020] In combination with the second aspect, in certain implementations of the second aspect, the calculation module includes a first calculation unit, which is specifically used to: process the decision to be evaluated to obtain a decision vector to be evaluated; obtain the user's expected decision for the preset function, and process the expected decision to obtain an expected decision vector; obtain the preset deviation allowed when adjusting the control parameters of the preset function based on the decision to be evaluated; and calculate the first dimension score based on the decision vector to be evaluated, the expected decision vector and the preset deviation.

[0021] In combination with the second aspect and the above-mentioned implementation methods, in some implementation methods of the second aspect, the first computing unit includes an acquisition subunit, which is specifically used to: after the vehicle cockpit executes the decision to be evaluated, obtain the interaction data between the user in the vehicle cockpit and the vehicle cockpit in response to the decision to be evaluated; based on the interaction data, determine the user's expected decision for the preset function.

[0022] In combination with the second aspect and the above-mentioned implementation methods, in some implementation methods of the second aspect, the calculation module includes a second calculation unit, which is specifically used to: obtain interaction command data, environmental data and function feedback data in the vehicle cabin; train the decision output model based on the interaction command data, environmental data and function feedback data, and store the trained decision output model in the on-board processing unit in the vehicle cabin; obtain the actual learning rate of the decision output model stored in the on-board processing unit in the vehicle cabin for the user's expected decision on the preset function during the training process, and obtain the number of training rounds when the decision output model training is completed; based on the scoring calculation algorithm built into the on-board processing unit, combined with the actual learning rate and the number of training rounds, calculate the second dimension score.

[0023] In combination with the second aspect and the above-mentioned implementation methods, in some implementation methods of the second aspect, the calculation module includes a third calculation unit, which is specifically used to: obtain multiple sets of sample environmental data in the vehicle cabin when no interference conditions are applied, and process each set of sample environmental data in the multiple sets of sample environmental data to obtain multiple sets of sample environmental vectors; obtain actual environmental data in the vehicle cabin after the interference conditions are applied, and process the actual environmental data to obtain actual environmental vectors; based on the multiple sets of sample environmental vectors and the actual environmental vectors, calculate the third dimension score.

[0024] In combination with the second aspect and the above-mentioned implementation methods, in some implementation methods of the second aspect, the third calculation unit includes a calculation subunit, which is specifically used to: calculate the sample mean vector based on multiple groups of sample environment vectors; calculate the difference between the sample mean vector and the actual environment vector to obtain a difference vector; calculate the sample covariance matrix based on the sample mean vector; and calculate the third dimension score based on the difference vector and the sample covariance matrix.

[0025] In combination with the second aspect and the above-mentioned implementation methods, in some implementation methods of the second aspect, the generation module is further specifically used to: perform weighted fusion of the first dimension score, the second dimension score and the third dimension score according to preset weights to generate a comprehensive score for the vehicle cabin.

[0026] In a third aspect, a vehicle is provided, comprising a memory and a processor. The memory is configured to store executable program code, and the processor is configured to retrieve and execute the executable program code from the memory, so that the vehicle executes the vehicle cabin evaluation method of the first aspect and any possible implementation of the first aspect.

[0027] In a fourth aspect, a computer program product is provided, comprising: a computer program code, which, when executed on a computer, enables the computer to execute the vehicle cabin evaluation method in the above-mentioned first aspect and any possible implementation of the first aspect.

[0028] In a fifth aspect, a computer-readable storage medium is provided, which stores a computer program code. When the computer program code runs on a computer, the computer executes the vehicle cabin evaluation method in the above-mentioned first aspect and any possible implementation of the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a schematic diagram of an evaluation process of a vehicle cockpit evaluation system in the prior art;

[0030] Figure 2 is a schematic flow chart of a vehicle cabin evaluation method provided in an embodiment of the present application;

[0031] Figure 3 This is a schematic diagram of an evaluation process of a vehicle cabin evaluation system provided by an embodiment of the present application;

[0032] Figure 4 1 is a schematic structural diagram of a vehicle cabin evaluation device provided in an embodiment of the present application;

[0033] Figure 5 It is a structural schematic diagram of a vehicle provided in an embodiment of the present application. DETAILED DESCRIPTION

[0034] The following will clearly and thoroughly describe the technical solutions in this application in conjunction with the accompanying drawings. In the description of the embodiments of this application, unless otherwise specified, " / " means or, for example, A / B can mean A or B: "and / or" in the text is only a description of the association relationship of associated objects, indicating that there can be three relationships, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more than two.

[0035] In the following, the terms "first" and "second" are used for descriptive purposes only and should not be understood to imply or suggest relative importance or implicitly indicate the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features.

[0036] With the advancement of intelligent vehicles, the smart cockpit has become a key hub for human-vehicle interaction. Building on traditional in-car entertainment and navigation capabilities, the smart cockpit integrates voice interaction, personalized recommendations, driver assistance, and other features to meet users' higher demands for comfort and safety.

[0037] However, the current industry evaluation of smart cockpits still mainly focuses on offline functional accuracy and static scenario testing, and lacks a systematic quantitative assessment of the dynamic adaptability of smart cockpits.

[0038] Among them, the above-mentioned dynamic adaptability is used to measure the ability of the smart cockpit system to quickly learn and maintain robust output when user behavior habits, environmental conditions or sensor status change.

[0039] In the process of evaluating smart cockpits, if there is a lack of effective assessment of the dynamic adaptability of smart cockpits, the smart cockpits may fail to meet user needs during actual use, or make wrong decisions under noise interference, resulting in poor user experience.

[0040] For example, Figure 1 As shown, in the prior art, an offline evaluation system usually receives one-time test data provided by the user, and performs offline training based on the one-time test data provided by the user. After the training is completed, the model is solidified to form a fixed model, and the fixed model will not be adjusted and updated according to new real-time data (i.e., there is no online update). After the fixed model is formed, the user will use the smart cockpit system containing the fixed model for a long time. However, due to the lack of real-time adaptive updates of the fixed model, in the subsequent long-term use process, even in the face of new scenarios, changes in user behavior, etc., the model cannot self-optimize and adjust based on the new data, that is, the dynamic adaptability of the smart cockpit system containing the fixed model is poor.

[0041] It can be understood that the above-mentioned offline evaluation refers to the evaluation of the effectiveness of models, systems or algorithms in an experimental environment or offline environment using existing labeled data (such as training sets and test sets) without the real-time participation of real users and without relying on online real-time data and environment.

[0042] It can be seen that the above Figure 1 When evaluating the smart cockpit system according to the evaluation process in the previous section, since offline evaluation systems are usually used for evaluation, it may be difficult to measure the adaptive ability of the smart cockpit system in the face of evolving user behavior and environmental changes during long-term use. The above evaluation process also lacks a quantitative description of the continuous learning efficiency of the models contained in the smart cockpit system, which may result in the smart cockpit system being unable to quickly adapt to new needs after going online. The above evaluation process also lacks an online feedback mechanism from real-time user data collection to model updates, and cannot guarantee the continuous optimization of the smart cockpit system in a dynamic environment. In addition, the existing industry specifications are relatively general in their requirements for the performance of smart cockpit systems under noise and abnormal data, making it difficult to conduct a horizontal comparison of the performance of smart cockpit systems.

[0043] In summary, the above Figure 1 The evaluation process in the evaluation of the smart cockpit system lacks an assessment of the dynamic adaptability of the smart cockpit, which may cause the smart cockpit to fail to meet user needs in actual use, or make wrong decisions under noise interference, resulting in poor user experience.

[0044] Based on this, an embodiment of the present application provides a vehicle cockpit evaluation method, executed by the vehicle, specifically its controller. This method evaluates the vehicle cockpit based on three core metrics: contextual relevance, incremental learning efficiency, and anti-interference capability. Using a weighted fusion approach, these three metrics are unified into a single adaptability score, helping the industry accurately measure and continuously optimize the adaptability of vehicle cockpits.

[0045] Figure 2 This is a schematic flowchart of a vehicle cabin evaluation method provided in an embodiment of the present application.

[0046] For example, Figure 2 As shown, the method 200 includes:

[0047] S201, obtaining a decision to be evaluated output by the vehicle cockpit for a preset function in the current scenario.

[0048] The above decision to be evaluated is used to adjust the control parameters of the preset function;

[0049] Regarding the above S201, it can be understood that the above vehicle cockpit is the smart cockpit system in the vehicle. The smart cockpit system is a user-centric vehicle system that provides intelligent and personalized services to drivers and passengers by integrating multiple advanced technologies.

[0050] The functions of the intelligent cockpit system generally include: infotainment function, human-computer interaction function, driver monitoring function, environmental perception and adjustment function, intelligent seat and space management function, advanced driving assistance function and personalized services.

[0051] The above-mentioned current scene can generally be constructed based on at least one of the external environment in which the vehicle is currently located, the current driving state of the vehicle, and the operating behavior of the user in the current vehicle.

[0052] For example, in the hot summer, the temperature inside the car is high, and the vehicle cabin will make temperature adjustment decisions based on the temperature inside the car in this scenario; when the vehicle is at high speed, the vehicle cabin's adjustment decisions for the air-conditioning function may consider factors such as wind resistance; if the user requests to adjust the air-conditioning temperature through voice commands, it will also affect the vehicle cabin's decision on the output of the air-conditioning function.

[0053] The above-mentioned preset functions refer to any functions of the smart cockpit system, such as temperature adjustment function, seat adjustment function, etc.

[0054] Typically, a vehicle's smart cockpit system is equipped with a machine learning model for outputting decisions (i.e., a decision output model). This machine learning model can output decisions based on the current cockpit scenario (including ambient temperature, driver and passenger status, vehicle driving conditions, etc.) to adjust the control parameters of certain functions in the vehicle cockpit to provide a comfortable environment for the driver and passengers. The decision to be evaluated is the control parameter output by the smart cockpit system for the preset function in the current scenario.

[0055] For example, with user authorization, the vehicle's intelligent cockpit system can detect the ambient temperature in the cabin in real time through a temperature sensor. If it detects that the current ambient temperature in the cabin is high, the machine learning model can output a decision based on the user's historical preference data and the current ambient temperature: "Turn on air conditioning cooling and adjust the air conditioning temperature to 20°C." The vehicle's controller can then turn on the air conditioning cooling function and adjust the air conditioning temperature to 20°C.

[0056] S202 : Based on the decision to be evaluated, a first dimension score for evaluating the vehicle cockpit is calculated, and a second dimension score and a third dimension score for evaluating the vehicle cockpit are obtained.

[0057] Among them, the above-mentioned first dimension scoring is used to evaluate the degree of match between the decision to be evaluated and the user's expected decision for the preset function; the above-mentioned second dimension scoring is used to evaluate the incremental learning efficiency of the decision output model in the vehicle cockpit; the above-mentioned third dimension scoring is used to evaluate the anti-interference ability of the vehicle cockpit.

[0058] It is understood that during the evaluation and testing of the intelligent cockpit system, the decision-to-be-evaluated output of the intelligent cockpit system for the preset functions in the current scenario can be obtained and evaluated. Furthermore, the incremental learning efficiency of the decision-making output model in the intelligent cockpit system and the anti-interference capability of the vehicle cockpit can also be evaluated.

[0059] The above-mentioned decision to be evaluated refers to the decision output by the decision output model in the vehicle cockpit for a certain function in the cockpit based on the current environment and user needs.

[0060] The user's desired decision for a preset function can be determined based on the interactive behavior data generated after the user outputs a decision to be evaluated for the vehicle cockpit. For example, if the decision to be evaluated output by the vehicle cockpit is "adjust the air conditioning temperature to 20°C," and after the vehicle controller adjusts the air conditioning temperature to 20°C, if the user is detected to adjust the air conditioning temperature to 18°C ​​using the air conditioning temperature control switch, then the user's desired decision for the air conditioning temperature adjustment function in the current scenario is "adjust the air conditioning temperature to 18°C."

[0061] The degree of match between the decision to be evaluated and the user's desired decision for the preset function can be understood as the similarity between the decision to be evaluated and the user's desired decision for the preset function. The higher the similarity, the higher the degree of match between the decision to be evaluated and the user's desired decision for the preset function. The degree of match between the decision to be evaluated and the user's desired decision for the preset function can generally be measured by calculating the first dimension score of the decision to be evaluated.

[0062] Furthermore, the above-mentioned first dimension score can also be used to measure the degree of matching between the decision to be evaluated and the context when given a context, that is, the first dimension score can also be called the context relevance of the decision to be evaluated.

[0063] Among them, the context can include "above" and "below". The "above" usually refers to the environmental data of the vehicle cabin and the user's historical behavior data; the "below" usually refers to the "user's expected decision for the preset function" mentioned above.

[0064] Exemplarily, the above-mentioned environmental data may include external environmental data and internal environmental data. External environmental data may include but is not limited to weather (such as rain, snow, fog, haze, and high temperature), road conditions (such as congestion, construction, speed limits), and lighting conditions (such as day or night). Internal environmental data may include but is not limited to internal vehicle temperature, humidity, air quality, and seat position.

[0065] The above-mentioned user historical behavior data refers to the behavior data generated by the user in the past when interacting with the vehicle cabin in different scenarios. For example, in a high temperature environment with the same ambient temperature, the user usually turned on the air conditioning cooling function and set the air conditioning temperature to 20°C.

[0066] Furthermore, the decision output model in the vehicle cockpit typically outputs a decision to be evaluated based on the "previous context." The vehicle controller then obtains the user's interactive behavior data generated after the vehicle cockpit outputs the decision to be evaluated and determines the similarity between the decision to be evaluated and the decision corresponding to the interactive behavior data, namely, the user's desired decision as described above. In other words, the first dimension score measures the degree of match between the output decision to be evaluated and the "next context" given the "previous context."

[0067] For example, after the vehicle cockpit executes the decision to be evaluated, user interaction data can be continuously obtained to obtain the user's expected decision for the preset function based on the user interaction data, and then based on the user's expected decision for the preset function, calculate the first dimension score for evaluating the vehicle cockpit.

[0068] In one possible implementation, based on the decision to be evaluated, a first dimension score for evaluating the vehicle cabin is calculated, including: processing the decision to be evaluated to obtain a decision vector to be evaluated; obtaining the user's expected decision for a preset function, and processing the expected decision to obtain an expected decision vector; obtaining a preset deviation allowed when adjusting the control parameters of the preset function based on the decision to be evaluated; and calculating the first dimension score based on the decision vector to be evaluated, the expected decision vector, and the preset deviation.

[0069] It is understood that the aforementioned processing of the decision to be evaluated specifically refers to vectorizing the decision to be evaluated to obtain a corresponding decision vector to be evaluated. The decision vector to be evaluated may also be referred to as a system decision vector.

[0070] Specifically, the vectorization processing of the decision to be evaluated can first determine the relevant features of the decision to be evaluated, then quantify each feature, and finally construct the decision vector to be evaluated.

[0071] For example, if the decision output model in the vehicle cabin outputs the decision of "turn on the air conditioning cooling and adjust the air conditioning temperature to 20°C", the vehicle controller can first determine the relevant features in the decision, including the air conditioning switch status (used to reflect whether the air conditioning is turned on), the air conditioning operating mode (used to clarify whether the air conditioning is in cooling, heating or other modes) and the set temperature (used to indicate the specific temperature value set by the air conditioning).

[0072] Each feature can then be quantified. For the air conditioner on / off state, the corresponding feature can be represented by binary code: the on state corresponds to a code of 1, and the off state corresponds to a code of 0. In the above output decision, if the air conditioner is on, then the feature value corresponding to the air conditioner on / off state is 1. For the air conditioner operating mode, the corresponding feature can also be represented by coding: the cooling mode corresponds to a code of 1, the heating mode corresponds to a code of 2, the dehumidification mode corresponds to a code of 3, and so on. In the above output decision, if the air conditioner is in cooling mode, then the feature value corresponding to the air conditioner operating mode is 1. For the set temperature, the set temperature value can be directly used to represent the corresponding feature. In the above output decision, if the air conditioner temperature is 20 degrees Celsius, then the feature value corresponding to the set temperature is 20.

[0073] Finally, by combining the eigenvalues ​​corresponding to the above features, we can obtain the decision vector to be evaluated as [1,1,20].

[0074] Similarly, for other decisions to be evaluated output by the vehicle cockpit, the above-mentioned vectorization processing method can also be referred to to obtain the decision vector to be evaluated.

[0075] Furthermore, as mentioned above, since the first dimension score is used to evaluate the degree of match between the decision to be evaluated and the user's expected decision for the preset function, in addition to the above-mentioned decision to be evaluated, the user's expected decision for the preset function can also be obtained.

[0076] In one possible implementation, obtaining the user's expected decision regarding a preset function includes: after the vehicle cockpit executes the decision to be evaluated, obtaining interaction data between the user in the vehicle cockpit and the vehicle cockpit in response to the decision to be evaluated; and determining the user's expected decision regarding the preset function based on the interaction data.

[0077] It's understandable that after the vehicle cabin executes the decision to be evaluated, the user might interact with the vehicle cabin in response to that decision. For example, if the vehicle cabin outputs the decision to be evaluated, "Turn on the air conditioning cooling function and adjust the temperature to 20°C," the vehicle controller then executes the decision to be evaluated: turning on the air conditioning cooling function and adjusting the temperature to 20°C. Later, the user might manually adjust the temperature to 18°C.

[0078] The user adjusting the air-conditioning temperature to 18° C. is the interaction data between the user and the vehicle cabin generated for the decision to be evaluated.

[0079] The interaction data related to the user's interactive behavior with the vehicle cabin in response to the decision to be evaluated generally reflects the user's desired decision. Based on this, the vehicle controller can determine the user's desired decision for the preset function based on the interaction data.

[0080] For example, if after the vehicle cabin executes the decision to be evaluated "Turn on the air conditioning and adjust the air conditioning temperature to 20°C", the user initiates a voice command "Adjust the air conditioning temperature to 18°C", then the voice command is the interaction data generated by the user with the vehicle cabin in response to the decision to be evaluated, and "Adjust the air conditioning temperature to 18°C" is the user's expected decision for the preset function.

[0081] After the vehicle cabin executes the decision to be evaluated "turn on the air conditioning and adjust the air conditioning temperature to 20℃", if there is no interaction data between the user and the vehicle cabin in response to the decision to be evaluated, it can be determined that "turn on the air conditioning and adjust the air conditioning temperature to 20℃" is the user's expected decision for the preset function.

[0082] Furthermore, the vectorization process described above can be referred to to perform vectorization processing on the user's expected decision for the preset function to obtain an expected decision vector, which can also be called a context feature vector.

[0083] For example, if the user's expected decision for the preset function is "adjust the air conditioning temperature to 18°C", it means that compared with the decision to be evaluated "turn on the air conditioning cooling and adjust the air conditioning temperature to 20°C", the eigenvalues ​​corresponding to the air conditioning switch state and the eigenvalues ​​corresponding to the air conditioning operation mode in the expected decision have not changed, but the eigenvalues ​​corresponding to the set temperature have changed. In this case, only the eigenvalues ​​corresponding to the set temperature in the decision vector to be evaluated need to be modified, and the expected decision vector can be obtained as [1,1,18].

[0084] Typically, a vehicle cockpit sets a maximum tolerance for adjusting the control parameters of a preset function based on the decision to be evaluated. This is known as the preset deviation. This maximum deviation between the decision to be evaluated and the expected decision is allowed.

[0085] For example, for the air conditioning temperature adjustment function, the maximum temperature deviation allowed between the decision to be evaluated and the expected decision is 10°C, that is, for the air conditioning temperature adjustment function, the allowed preset deviation is set to 10°C.

[0086] Furthermore, after obtaining the decision vector to be evaluated, the expected decision vector and the allowed preset deviation, a first dimension score for evaluating the degree of match between the decision to be evaluated and the user's expected decision for the preset function can be calculated based on the decision vector to be evaluated, the expected decision vector and the preset deviation.

[0087] It is understandable that the first dimension score can also be called the context relevance of the decision to be evaluated (C context ), the first dimension score can reflect the matching degree between the decision to be evaluated output by the vehicle cockpit and the expected decision. For example, if the first dimension score is 0.6, it means that the matching degree between the currently output decision to be evaluated and the expected decision is 60%.

[0088] In some embodiments, the context relevance of the decision to be evaluated may be calculated based on the similarity between the decision to be evaluated and the expected decision, that is, a first dimension score for evaluating the vehicle cabin may be obtained.

[0089] It can be understood that since the calculation of similarity is intuitive and efficient, by calculating the similarity between the decision to be evaluated and the expected decision, the numerical difference between the decision to be evaluated and the expected decision can be quickly quantified.

[0090] Furthermore, the first dimension score is calculated based on the similarity between the decision to be evaluated and the expected decision, which can be referred to the following formula (1):

[0091]

[0092] Where D represents the decision vector to be evaluated; X represents the expected decision vector; D max represents the allowed preset deviation; |DX| represents the modulus of the vector difference between the decision vector to be evaluated and the expected decision vector; |DX|‖ represents the absolute value of the modulus; SimScore(D,X) represents the first dimension score.

[0093] For example, for the air conditioning temperature adjustment function, assuming that the obtained decision vector to be evaluated D is [1,1,20], the obtained expected decision vector X is [1,1,18], and the obtained allowable preset deviation D is max If it is 10℃, then the modulus of the vector difference between the decision vector to be evaluated and the expected decision vector can be calculated: Then the calculated modulus and the obtained preset deviation D max Substituting into the above formula (1), we can get the first dimension score: SimScore(D,X)=0.8.

[0094] The above method processes the decision to be evaluated to obtain a decision vector, and processes the expected decision to obtain an expected decision vector. This quantifies the decision, facilitating subsequent precise assessment of the gap between the decision to be evaluated and user expectations. Calculating a first-dimension score based on the decision vector, the expected decision vector, and a preset deviation converts the vehicle cockpit's decision-making performance into a specific numerical value, intuitively reflecting the degree of match between the decision to be evaluated and the user's expected decision for the preset function. Furthermore, the calculated first-dimension score provides the vehicle cockpit with clear feedback on user needs, facilitating subsequent targeted optimization and improvement of the decision output model. This helps optimize and refine the vehicle cockpit's output decisions, improving their accuracy and effectiveness, and ultimately enhancing the overall performance and user satisfaction of the vehicle cockpit.

[0095] Furthermore, to ensure that a vehicle's intelligent cockpit system can quickly adapt to new requirements after launch, the decision output model within the intelligent cockpit system is typically required to have a high incremental learning efficiency. Based on this, the incremental learning efficiency of the decision output model can be evaluated to obtain a second-dimensional score. Based on this second-dimensional score, targeted optimization measures can be implemented for the decision output model to better adapt the intelligent cockpit system to new requirements.

[0096] Incremental learning efficiency is a key metric for measuring a model's ability to quickly and effectively learn new knowledge and improve its performance to adapt to changes as it continuously receives new data. A higher incremental learning efficiency means the model learns faster and achieves better results when processing new data.

[0097] When stored in the vehicle's in-cabin processing unit, the decision output model is typically a general-purpose model. This general-purpose model is trained on a wide range of data and excels at general tasks, but lacks prior knowledge of the in-cabin scenario. Therefore, after the decision output model is stored in the in-cabin processing unit, it can be trained using in-cabin data and constraints related to the in-cabin scenario, strengthening its decision-making capabilities related to in-cabin functions.

[0098] In one possible implementation, obtaining a second dimension score for evaluating the vehicle cockpit includes: obtaining interactive command data, environmental data, and functional feedback data in the vehicle cockpit; training the decision output model based on the interactive command data, the environmental data, and the functional feedback data, and storing the trained decision output model in an on-board processing unit in the vehicle cockpit; obtaining an actual learning rate of the decision output model stored in the on-board processing unit in the vehicle cockpit regarding the user's expected decision for the preset function during training, and obtaining the number of training rounds after which the decision output model training is completed; and calculating the second dimension score based on a scoring calculation algorithm built into the on-board processing unit, in combination with the actual learning rate and the number of training rounds.

[0099] It is understandable that the above-mentioned interaction command data, environmental data and function feedback data in the vehicle cabin can usually be collected by sensors in the vehicle cabin.

[0100] The above-mentioned interactive command data may be an interactive command initiated by a user when he or she desires to activate a preset function in the vehicle cabin, including but not limited to voice commands, touch operations, and gesture controls, such as a voice command related to "It's too hot, please turn on the air conditioner";

[0101] The above-mentioned environmental data can be scene information inside the vehicle cabin and environmental information outside the vehicle, including but not limited to temperature, humidity, light intensity, seat position, number of passengers, vehicle location, and road conditions.

[0102] The aforementioned function feedback data may include the actual execution results and user feedback after executing certain preset functions. For example, the actual execution result may be adjusting the air conditioning temperature in the vehicle cabin to 20°C. The user feedback information may include adjustments to the preset function initiated by the user after the execution result is output, such as manually adjusting the air conditioning temperature to 20°C after executing the air conditioning adjustment operation. The aforementioned user feedback information can also be understood as the user's expected decision regarding a preset function as described above.

[0103] Furthermore, after obtaining the interactive command data, environmental data and functional feedback data in the vehicle cabin, the obtained data can be processed to form multiple groups of data combinations of the "scene (environmental data) - action (interactive command data) - result (functional feedback data)" type, and a cabin-specific data set can be constructed.

[0104] Constraints can also be formed based on hardware capability limitations in the cabin (such as maximum screen brightness and maximum air conditioning adjustment range), safety rules (such as prohibiting high-risk operations while driving), multi-tasking priorities (such as pausing music when a call comes in), and user permissions (such as disabling certain functions when the child lock is on).

[0105] In the process of training the decision output model based on the constructed cockpit-specific data set, constraints are added and the decision output model is iteratively trained until the accuracy of the decision output by the decision output model reaches a preset accuracy threshold (for example, 80%), and the decision output model training is determined to be completed.

[0106] The accuracy of the decision output by the decision output model may be the similarity between the decision output by the decision output model and the user's expected decision for the preset function. The specific method for calculating the similarity between the decision output by the decision output model and the user's expected decision for the preset function is described above and is not further described here.

[0107] Furthermore, after determining that the decision output model training is completed, the decision output model can be stored in an onboard processing unit in the vehicle cabin.

[0108] It can be understood that the above-mentioned on-board processing unit is a processing unit in the cockpit controller, which is responsible for running the decision output model and processing data.

[0109] The actual learning rate is used to measure the speed of performance improvement brought about by each iteration of the decision output model, or to measure the speed of parameter adjustment during each iteration update of the decision output model.

[0110] As mentioned above, since the above-mentioned decision output model will take into account user feedback information, that is, the user's expected decision for a preset function during the training process, the above-mentioned actual learning rate can also be understood as the learning rate of the decision output model for the user's expected decision for the preset function during the training process.

[0111] The actual learning rate mentioned above can usually be a constant set when building the model.

[0112] The number of training rounds after which the decision output model is trained can also be understood as the number of iterations of the decision output model. The decision output model typically completes its training when the accuracy of its output decision exceeds a preset accuracy threshold (e.g., 80%).

[0113] The above-mentioned scoring calculation algorithm is a mathematical model built into the on-board processing unit, which is used to quantify the incremental learning efficiency of the decision output model.

[0114] For example, after the decision output model is stored in the onboard processing unit in the vehicle cabin, the number of training rounds can be obtained from the training log of the decision output model. Typically, during the training of the decision output model, relevant information during the training process is recorded in the log. The training log will record the relevant indicators of each training round, such as loss value and accuracy, in detail. By viewing the log file, the number of training rounds after which the model training is completed can be found.

[0115] In addition, when the decision output model is saved, relevant information such as the number of training rounds may be directly saved in the metadata of the model file. When the decision output model is loaded, the number of training rounds of the model can be read at the same time.

[0116] Furthermore, after obtaining the actual learning rate of the decision output model and the number of training rounds after which the decision output model is trained, the onboard processing unit can calculate a second dimension score for evaluating the vehicle cabin based on a built-in scoring calculation algorithm and in combination with the actual learning rate and the number of training rounds. This second dimension score can also be called the incremental learning efficiency (E) of the decision output model. learn ).

[0117] In some embodiments, an exponential decay model can be used to describe the incremental learning efficiency of the decision output model. The exponential decay model is a mathematical model used to describe the phenomenon of exponential decay over time or other variables. The general form of the exponential decay model is: y = A·e -kt , where y is the variable value that changes with time t, A is the initial value, k is the decay constant, e is the natural constant, and t is the time variable.

[0118] Furthermore, based on the above exponential decay model, a formula for calculating the second dimension score can be obtained. The calculation of the second dimension score can refer to the following formula (2):

[0119] E learn =1-e -λ×n Formula (2)

[0120] Where λ represents the actual learning rate of the decision output model; n represents the number of training rounds after the decision output model training is completed; E learn It represents the incremental learning efficiency of the decision output model, which is the second dimension score used to evaluate the vehicle cabin.

[0121] It can be understood that based on the definition of the exponential decay model, e -λ×n It can be shown that as the number of training rounds n of the decision output model increases, the proportion of knowledge that the decision output model has not learned decreases. -λ×n , this attenuation can be converted into an increase in learning efficiency. -λ×nAs n increases, it decreases and approaches 0, then 1-e -λ×n It will increase as n increases, approaching 1 from 0, which is consistent with the process of continuous improvement of model learning efficiency as the number of training rounds increases.

[0122] In the above formula (2), λ represents the actual learning rate of the decision output model. When the number of training rounds is the same, the larger λ is, the faster e -λ×n The faster the decay, the learn The closer it is to 1, the greater the incremental learning efficiency of the decision output model.

[0123] In the above formula (2), n represents the number of training rounds. Under the condition of the same actual learning rate, as the number of training rounds n increases, e -λ×n Getting smaller and smaller, E learn The closer it is to 1, the greater the incremental learning efficiency of the decision output model.

[0124] In the above formula (2), when new data is introduced to train the decision output model, as the number of training rounds n increases, e -λ×n will decrease rapidly, E learn It will increase rapidly, which is equivalent to introducing new data into the decision output model. The new data brings new information. The decision output model can quickly absorb this information in the initial training stage, which makes the performance of the decision output model (based on incremental learning efficiency) improve rapidly. As the number of training rounds n increases, e -λ×n will approach 0, E learn It will approach 1, which means that the room for improving the incremental learning efficiency of the decision output model is getting smaller and smaller, and the speed of performance improvement of the decision output model is gradually slowing down and eventually tends to be stable.

[0125] In summary, based on the formula (2) obtained from the above exponential decay model, using the two variables of the actual learning rate λ and the number of training rounds n of the decision output model, we can accurately characterize the process of the incremental learning efficiency of the decision output model from rapid growth to gradual stabilization, that is, it can accurately reflect the process of rapid improvement and gradual stabilization of the performance of the decision output model after the introduction of new data.

[0126] For example, assuming that the decision model in the vehicle cabin is a cabin temperature adjustment decision model for adjusting the cabin temperature, the vehicle cabin can use the cabin temperature adjustment decision model to intelligently adjust the temperature in the vehicle cabin based on interaction command data (such as the voice command "it's too hot"), environmental data (such as the temperature inside and outside the vehicle, sunlight intensity), and the user's past temperature preferences.

[0127] The cabin temperature adjustment decision model may be a neural network model. Initially, the actual learning rate λ of the cabin temperature adjustment decision model for the user's desired decision regarding a preset function may be set to 0.05 before training begins. Before training begins, the number of training rounds n for the model is 0.

[0128] During daily vehicle driving, interaction command data, environmental data, and functional feedback data (including interaction data from users manually adjusting the air conditioning temperature) are continuously collected. After each driving trip, this data is input into the cabin temperature adjustment decision model as training samples to iteratively train the cabin temperature adjustment decision model. The number of training rounds n increases by 1 with each training round.

[0129] After continuous iterative training, the number of training rounds n increases to 10. Substituting n = 10 and λ = 0.05 into the above formula (2), we can calculate E learn =1-e -0.05×10 =1-e -0.5 ≈0.393. It can be seen that after 10 rounds of training, the incremental learning efficiency of the cabin temperature adjustment decision model is still relatively low, and the cabin temperature adjustment decision model cannot accurately adjust the cabin temperature.

[0130] Subsequently, data can be collected and iterative training can be continued. When the number of training rounds n increases to 50, substituting n = 50 and λ = 0.05 into the above formula (2), E can be calculated. learn =1-e -0.05×50 =1-e -2.5 ≈0.918. It can be seen that after 50 rounds of training, the incremental learning efficiency of the cabin temperature adjustment decision model has been greatly improved. The cabin temperature adjustment decision model can effectively output temperature adjustment decisions that are close to user needs based on the input data.

[0131] Furthermore, in the above process, when it is determined that the accuracy of the decision output by the cabin temperature adjustment decision model reaches a preset accuracy threshold (for example, 80%), it is determined that the training of the cabin temperature adjustment decision model is completed, and the cabin temperature adjustment decision model can be stored in the on-board processing unit in the vehicle cabin.

[0132] For example, if the actual learning rate λ of the decision output model stored in the vehicle processing unit in the vehicle cockpit for the user's expected decision for the preset function during the training process is obtained to be 0.2, and the number of training rounds n after the decision output model training is completed is obtained to be 10, then the learning rate constant λ and the number of training rounds n can be substituted into the above formula (2) to calculate the second dimension score E learn =1-e -0.2×10 =1-e -2 ≈0.865.

[0133] In some embodiments, when the vehicle is started, the on-board computing unit in the vehicle cockpit can obtain the actual learning rate of the decision output model for the user's expected decision for the preset function during the last training process, and obtain the number of training rounds after the decision output model training is completed. Then, based on the scoring calculation algorithm built into the on-board processing unit, combined with the actual learning rate and number of training rounds, the second dimension score of the most recent model training is automatically calculated. If the second dimension score is lower than the preset scoring threshold, it means that the current incremental learning efficiency of the model is too low, and the driver can be prompted to upgrade the decision output model through over-the-air technology (OTA) to improve the incremental learning efficiency of the decision output model.

[0134] For example, the preset scoring threshold can be set according to actual needs, for example, it can be set to 0.5.

[0135] It is understood that the OTA upgrade can include sending an optimized learning rate constant (e.g., an increased learning rate constant) or adjusting the upper limit of the number of training rounds. By increasing the learning rate constant and raising the upper limit of the number of training rounds, the incremental learning efficiency of the decision output model is increased.

[0136] Furthermore, through the above-mentioned OTA upgrade, the incremental learning efficiency of the decision output model in the vehicle cockpit can be improved, thereby improving the second dimension score of the vehicle cockpit.

[0137] The above method, by obtaining the actual learning rate during the decision output model training process and the number of training rounds at the completion of training, can intuitively understand the learning progress of the decision output model during training. The actual learning rate reflects the speed at which the decision output model absorbs knowledge, while the number of training rounds indicates the depth of the learning process. The combination of the two accurately measures the incremental learning efficiency of the decision output model, thus providing a strong basis for evaluating vehicle cockpits based on this dimension of incremental learning efficiency. Calculating a second dimension score based on the actual learning rate and the number of training rounds quantifies the incremental learning efficiency of the decision output model. This quantification method provides a specific, comparable and analyzable numerical metric for vehicle cockpit evaluation. When comparing the performance of different vehicle cockpits or different configurations of the same cockpit, the second dimension score can clearly demonstrate the differences, facilitating researchers' targeted analysis of the decision output model's performance and subsequent targeted optimization to improve vehicle cockpit performance.

[0138] Furthermore, to ensure that a vehicle's intelligent cockpit system can maintain stable and highly accurate decision-making despite noise, sensor failures, or extreme environmental conditions, it is typically required to have strong anti-interference capabilities. Based on this, the vehicle cockpit's anti-interference capabilities can be evaluated to obtain a third-dimensional score. Based on this third-dimensional score, targeted optimization measures can be implemented to ensure that the vehicle cockpit has stronger anti-interference capabilities.

[0139] For example, during the test and evaluation of the vehicle cockpit, some interference conditions can be imposed on the vehicle cockpit, and then data from the vehicle cockpit system can be collected. The data collected during the interference period can be compared with the data collected during the normal state (i.e., without interference), so as to know the strength of the vehicle cockpit's anti-interference ability.

[0140] The above-mentioned evaluation of the anti-interference capability of the vehicle cockpit can specifically be an evaluation of the anti-interference capability of the actuators and interactive hardware in the vehicle cockpit, such as the display screen, speakers, air-conditioning system and other sensors in the vehicle cockpit. It can also be used to evaluate the anti-interference capability of the software algorithms in the vehicle cockpit, such as the environmental adjustment function, voice recognition function, scoring calculation method, gesture control function, image display function and warning prompt function in the vehicle cockpit.

[0141] In one possible implementation, obtaining a third dimension score for evaluating the vehicle cabin includes: obtaining multiple sets of sample environmental data in the vehicle cabin when no interference conditions are applied, and processing each set of sample environmental data in the multiple sets of sample environmental data separately to obtain multiple sets of sample environmental vectors; obtaining actual environmental data in the vehicle cabin after the interference conditions are applied, and processing the actual environmental data to obtain actual environmental vectors; and calculating the third dimension score based on the multiple sets of sample environmental vectors and the actual environmental vector.

[0142] It is understandable that the above interference conditions may include noise interference, sensor failure interference, extreme environment interference and other conditions.

[0143] For example, noise interference can be generated by simulating noise sources in different scenarios, and the vehicle cabin's ability to resist noise signals can be tested. For example, the accuracy of the vehicle cabin's recognition of voice commands before and after the noise interference can be compared to determine whether the vehicle cabin's misrecognition rate of user-initiated voice commands under noise interference is less than 5%.

[0144] By simulating sensor anomalies to generate sensor failure interference, the vehicle cabin's anti-interference capability in the event of a sensor failure can be tested. For example, the temperature sensor circuit in the vehicle cabin can be disconnected to simulate a temperature sensor failure. The temperature fluctuations in the vehicle cabin before and after the failure can be compared to determine whether the cabin temperature fluctuation remains within ±2°C when the sensor fails.

[0145] The environmental chamber can be used to simulate extreme conditions such as high temperature, low temperature, high humidity, and strong light to create extreme environmental interference, and test the vehicle cabin's anti-interference ability in extreme environments. For example, the response status of the vehicle's display screen before and after the application of extreme environmental interference can be compared to determine whether the display screen's response delay at low temperatures is less than or equal to 100ms.

[0146] The aforementioned sample environmental data refers to a certain number of sets of environmental data continuously collected by sensors in the vehicle cabin while the vehicle cabin is operating normally (i.e., without interference conditions), including but not limited to temperature, humidity, lighting, and other environmental data. The aforementioned sample environmental data may be environmental data collected within the vehicle cabin after a decision regarding the vehicle cabin output is executed, or it may be environmental data collected within the vehicle cabin before a decision regarding the vehicle cabin output is executed.

[0147] Furthermore, by performing the aforementioned vectorization processing on each of the multiple sets of sample environment data obtained, multiple sets of sample environment vectors can be obtained.

[0148] The aforementioned actual environmental data refers to various environmental data collected by sensors within the vehicle cabin after interference conditions are applied to the vehicle cabin, including but not limited to temperature, humidity, and lighting. This actual environmental data is data describing the same characteristics as the sample environmental data, along the same dimensions. Similarly, this actual environmental data can be collected after a decision regarding the vehicle cabin output is executed, or before a decision regarding the vehicle cabin output is executed.

[0149] Similarly, the actual environment vector can be obtained by performing the vectorization processing described above on the acquired actual environment data.

[0150] Furthermore, after obtaining the sample environment vector and the actual environment vector, a third dimension score for evaluating the vehicle cabin may be calculated based on the sample environment vector and the actual environment vector.

[0151] In one possible implementation, the third dimension score is calculated based on the multiple groups of sample environment vectors and the actual environment vector, including: calculating the sample mean vector based on the multiple groups of sample environment vectors; calculating the difference between the sample mean vector and the actual environment vector to obtain a difference vector; calculating the sample covariance matrix based on the sample mean vector; and calculating the third dimension score based on the difference vector and the sample covariance matrix.

[0152] It will be appreciated that, as previously described, the aforementioned multiple sets of sample environment vectors record multiple sets of environmental data continuously acquired from the vehicle cabin when no interference conditions are applied. By averaging these multiple sets of sample environment vectors, a sample mean vector can be obtained. Specifically, the average value of the vector elements along each dimension can be calculated, and the sample mean vector can be obtained by combining these average values.

[0153] For example, for each feature (such as ambient temperature, ambient humidity, etc.), the average value of all sample data can be calculated based on the following formula (3):

[0154]

[0155] Wherein, the above N represents the total number of sample data, that is, the number of sample data used to calculate the average value; the above a i Represents the value of the i-th sample data, i is used to represent the index of the sample data, and i can be any number between 1 and N; the above μ represents the average value of all sample data for a certain feature (such as ambient temperature, ambient humidity, etc.).

[0156] Furthermore, after calculating the average values ​​of all sample data for each feature, a sample mean vector can be constructed based on these average values.

[0157] It can be understood that the above sample mean vector can comprehensively reflect the general state of the vehicle cabin environment when no interference conditions are applied, and can be used as a benchmark when evaluating the anti-interference capability of the vehicle cabin.

[0158] Furthermore, the difference between the sample mean vector and the actual environment vector obtained after the interference condition is applied can be calculated to obtain a difference vector. This difference vector reflects the difference in various dimensions between the actual cabin environment state after the interference condition is applied and the typical environment state represented by the sample. This difference vector can be used to measure the degree of deviation between the actual environment vector and the sample mean vector.

[0159] In addition, the sample covariance matrix can be calculated based on the sample mean vector, which can reflect the linear correlation between various environmental features.

[0160] For example, the formula for calculating the sample covariance matrix can refer to the following formula (4):

[0161]

[0162] Wherein, the above N represents the total number of sample data, that is, the number of sample data used to calculate the average value; the above a i Represents the value of the i-th sample data, i is used to represent the index of the sample data, i can be any number between 1 and N; the above μ represents the average value of all sample data for a certain feature (such as ambient temperature, ambient humidity, etc.); the above (a i -μ) represents the difference vector between the i-th sample data and the average value of the sample data; the above (a i -μ) T Represents the inverse of the difference vector between the i-th sample data and the mean of the sample data.

[0163] For example, if N sample data are substituted into the above formula (3), the average temperature under normal conditions without interference conditions is calculated to be 22°C, and the average humidity under normal conditions without interference conditions is calculated to be 45%, then the sample mean vector can be constructed: μ = (22, 45) T

[0164] Furthermore, by substituting N sample data and the constructed sample mean vector into the above formula (4), the covariance matrix can be calculated:

[0165]

[0166] Among them, the main diagonal elements Σ in the above covariance matrix are jj (When k=j) represents the variance of the jth feature itself; the non-main diagonal elements Σ jk (When k≠j) represents the covariance between the j-th feature and the k-th feature.

[0167] In the above covariance matrix Σ 11 =1, indicating that the variance between each temperature data is 1; in the above covariance matrix Σ 22 =4, indicating that the variance between each humidity data is 4. The non-main diagonal elements in the above covariance matrix are 0, indicating that the covariance between the two features of temperature and humidity is 0, that is, temperature and humidity are linearly independent.

[0168] It is understood that, in practical applications, the calculation of the above covariance matrix can generally be completed with the help of data processing software or programming language (such as the NumPy library in Python). The above calculation process is only an example.

[0169] Furthermore, calculating the sample covariance matrix facilitates obtaining important statistical characteristics of the sample data distribution, providing a reference for subsequent quantification of the anti-interference capability of the vehicle cabin.

[0170] Generally speaking, the distribution distance method can measure the statistical distribution difference of the data under normal conditions and interference conditions, thereby quantifying the anti-interference ability of the system. Commonly used distribution distance methods include the Mahalanobis distance calculation method and the KL divergence calculation method. The following uses the Mahalanobis distance calculation method as an example to measure the statistical distribution difference of the data under normal conditions and interference conditions in the vehicle cabin, and quantify the anti-interference ability of the vehicle cabin (R robust ), which is to calculate the third dimension score used to evaluate the vehicle cabin.

[0171] In some embodiments, based on the Mahalanobis distance calculation method, the third dimension score for evaluating the vehicle cabin can be calculated according to the following formula (5):

[0172]

[0173] Wherein, the above x represents the actual environment vector obtained based on the actual environment data after the interference condition is applied to the vehicle cabin; the above μ represents the sample mean vector obtained based on the sample environment data when the interference condition is not applied; the above (x-μ) represents the difference vector between the actual environment vector and the sample mean vector; the above (x-μ) T represents the transposed vector of the difference vector; the above Σ represents the sample covariance matrix corresponding to the sample environment data when no interference conditions are applied; the above Σ -1 Represents the inverse matrix corresponding to the sample covariance matrix; the above D M It indicates the degree of deviation between the actual environmental data and the sample environmental data after the interference condition is applied, D M The larger the value is, the further the actual environmental data under interference conditions deviate from the normal distribution of the sample environmental data, that is, the worse the anti-interference ability of the system is.

[0174] For example, assuming that after the interference condition is applied, the actual environmental data obtained by sampling is: the ambient temperature is 24°C and the ambient humidity is 42%, then the actual environmental vector can be obtained as x = (24, 42) T If the sample mean vector is μ=(22,45) T , we can calculate the difference vector between the actual environment vector and the sample mean vector as x-μ=(2,-3) T By transposing the above difference vector, we can get (x-μ) T =(2,-3).

[0175] If the sample covariance matrix is:

[0176]

[0177] The inverse matrix of the sample covariance matrix is ​​calculated (the inverse of the diagonal matrix is ​​the reciprocal of each diagonal element):

[0178]

[0179] Calculate Σ -1 (x-μ):

[0180]

[0181] Calculate D M :

[0182]

[0183] It is understandable that the above D M =2.5 means that the degree of deviation between the environmental data collected after the interference conditions are applied to the vehicle cabin and the data collected under normal conditions is 2.5.

[0184] In some embodiments, a deviation threshold may be set in advance, and the anti-interference capability of the vehicle cabin system may be measured by comparing the difference between the calculated deviation degree and the deviation threshold.

[0185] For example, the above-mentioned deviation threshold can be set according to actual needs, for example, it can be set to 1.5. If the deviation is greater than the deviation threshold, it means that the anti-interference ability of the vehicle cabin is poor.

[0186] In some embodiments, the calculated deviation values ​​may be normalized to obtain a final third dimension score.

[0187] For example, the Sigmoid function can be used, that is, y = 1 / 1 + e -x , normalize the calculated deviation value. For example, if the calculated deviation value is 2.5, then substituting x = 2.5 into the Sigmoid function yields a normalized value of approximately 0.92. This normalized value 0.92 can be determined as the final third dimension score.

[0188] It can be understood that the above-mentioned method of performing normalization processing using the Sigmoid function is only an example, and the embodiments of the present application do not limit the specific normalization processing method.

[0189] The above method obtains multiple sets of sample environmental data without interference conditions, comprehensively covering the environmental conditions of the vehicle cabin at different times, locations, and operating conditions. This method more realistically reflects the environmental characteristics of the cabin under normal conditions, avoiding the bias caused by a single data point, and provides a rich and accurate data foundation for subsequent vehicle cabin evaluation. By calculating the sample mean vector, a representative representation of the vehicle cabin's environmental state under normal conditions is obtained. By calculating the difference vector between the sample mean vector and the actual environmental vector, the deviation of the actual cabin environmental state from the normal state in various dimensions after the interference conditions are applied is clearly determined. By calculating the sample covariance matrix, the data distribution characteristics of multiple sets of sample environmental vectors are measured, allowing for a more reasonable assessment of the degree of deviation of the actual environmental data from the sample environmental distribution. Finally, based on the difference vector and sample covariance matrix, a third dimension score, measuring the vehicle cabin's anti-interference ability, can be more accurately calculated. This helps understand the vehicle cabin's anti-interference ability during testing, provides a data basis for subsequent optimization of the vehicle cabin, and helps improve the overall performance of the vehicle cabin.

[0190] S203 : Generate a comprehensive score for the vehicle cabin based on the first dimension score, the second dimension score, and the third dimension score.

[0191] It is understood that after obtaining the first, second, and third dimension scores, the three dimension scores can be combined into a unified adaptability index to obtain a comprehensive score for the vehicle cabin. This comprehensive score is used to measure the dynamic adaptability of the vehicle cabin.

[0192] In one possible implementation, a comprehensive score for the vehicle cabin is generated based on the first dimension score, the second dimension score, and the third dimension score, including: weighted fusion of the first dimension score, the second dimension score, and the third dimension score according to preset weights to generate a comprehensive score for the vehicle cabin.

[0193] It is understandable that the above preset weights can be set according to actual needs. For example, the weight corresponding to the first dimension score is α, the weight corresponding to the second dimension score is β, and the weight corresponding to the third dimension score is δ. The above weights satisfy: α + β + δ = 1.

[0194] In some embodiments, the preset weights can be set based on the vehicle's varying cabin capability requirements in different scenarios. Typically, the initial weight corresponding to the first dimension score is set to 0.4, the initial weight corresponding to the second dimension score is set to 0.3, and the initial weight corresponding to the third dimension score is set to 0.3. If the vehicle's scenario changes, the weights corresponding to the scores in these different dimensions can be adjusted accordingly.

[0195] It is understandable that when the scenario in which the vehicle is located changes, the weights corresponding to the scores of different dimensions are usually adjusted based on the initial weights, and the sum of the weights corresponding to the scores of each dimension after adjustment is ensured to be still 1.

[0196] For example, when a vehicle is operating in an extreme environment, it is more likely to be affected by environmental interference. To ensure the normal operation of various functions in the vehicle cabin, the weight corresponding to the third dimension score used to evaluate the vehicle cabin's anti-interference ability can generally be appropriately increased, while the weight corresponding to the first dimension score and the second dimension score can be reduced, prioritizing the stability of vehicle cabin function execution. For example, based on the initial weights, the weight corresponding to the third dimension score can be increased to 0.5, the weight corresponding to the first dimension score can be reduced to 0.3, and the weight corresponding to the second dimension score can be reduced to 0.2.

[0197] When driving on urban roads, drivers typically need to interact frequently with the vehicle cockpit, such as for navigation, music control, and air conditioning. To ensure smooth interaction between the vehicle cockpit and the user, the weight of the first dimension score, which evaluates the degree of match between the decision to be evaluated and the user's desired decision for the preset function, can be appropriately increased, while the weights of the second and third dimension scores can be reduced, prioritizing smooth interaction between the vehicle cockpit and the user. For example, based on the initial weights, the weight of the first dimension score can be increased to 0.5, the weight of the second dimension score can be reduced to 0.25, and the weight of the third dimension score can be reduced to 0.25.

[0198] For example, the comprehensive score of the vehicle cabin can be calculated according to the following formula (6):

[0199] A adapter =α·C context +β·E learn +δ·R robust Formula (6)

[0200] Among them, the above A adapter Represents a unified adaptability index, i.e., a comprehensive score for the vehicle cockpit; the above C context represents the context relevance of the decision to be evaluated, i.e., the first dimension score; the above α represents the weight corresponding to the first dimension score; the above E learn represents the incremental learning efficiency of the decision output model, i.e., the second dimension score; the above β represents the weight corresponding to the second dimension score; the above R robust represents the anti-interference ability of the vehicle cabin, that is, the third dimension score; the above δ represents the weight corresponding to the third dimension score.

[0201] For example, assume that the weight α corresponding to the first dimension score is set to 0.4, the weight β corresponding to the second dimension score is set to 0.3, and the weight δ corresponding to the third dimension score is set to 0.3. If the first dimension score is 0.8, the second dimension score is 0.63, and the third dimension score is 0.97, substituting the above values ​​into formula (6), we can calculate A adapter =0.4×0.8+0.3×0.63+0.3×0.97=0.8, that is, the calculated comprehensive score for the vehicle cabin is 0.8.

[0202] In some embodiments, the maximum value corresponding to the comprehensive score for evaluating the dynamic adaptability of the vehicle cabin can be set in advance, that is, the "full score" corresponding to the comprehensive score can be set in advance, and based on the difference between the "full score" corresponding to the comprehensive score and the calculated comprehensive score, the dynamic adaptability of the vehicle cabin in the current test scenario can be evaluated.

[0203] For example, the maximum value corresponding to the above comprehensive score can be set according to actual needs, for example, it can be set to 1.

[0204] Assuming that the "full score" corresponding to the comprehensive score is 1, if the calculated comprehensive score for evaluating the dynamic adaptability of the vehicle cockpit is 0.8, it can be determined that the dynamic adaptability of the vehicle cockpit is good under the current test scenario, that is, the vehicle cockpit has strong dynamic adaptability.

[0205] In some embodiments, a scoring threshold may be set. After obtaining the comprehensive score for evaluating the dynamic adaptability of the vehicle cabin, the comprehensive score may be compared with the scoring threshold to determine whether to further optimize the vehicle cabin.

[0206] For example, the above-mentioned scoring threshold can be set according to actual needs, for example, the scoring threshold is set to 0.75. If the calculated comprehensive score of the vehicle cabin is less than or equal to 0.75, it is determined that the vehicle cabin needs to be optimized; conversely, if the calculated comprehensive score of the vehicle cabin is greater than 0.75, the vehicle cabin may not be optimized.

[0207] In the above method, the first, second, and third dimension scores evaluate vehicle cabin characteristics from different perspectives. Through weighted fusion, this scattered information is integrated, avoiding the one-sidedness of single-dimensional evaluations and providing a comprehensive picture of the vehicle cabin's overall condition. Furthermore, by integrating these multi-dimensional scores into a single comprehensive score, a unified, quantitative standard is provided for evaluating different vehicle cabins. Whether comparing cabins of different models or evaluating the cabins of the same model with different configurations, the comprehensive score clearly demonstrates the differences, facilitating horizontal comparisons across models and solutions, and providing a more intuitive basis for vehicle cabin evaluation.

[0208] In some embodiments, when it is determined that the vehicle cockpit needs to be optimized based on the comprehensive score of the vehicle cockpit, the vehicle cockpit can be optimized in a targeted manner based on the scores in different test scenarios.

[0209] For example, if the air conditioning temperature adjustment function in the vehicle cabin does not accurately match the user's desired decision for the preset function, the decision output model in the vehicle cabin can be iteratively trained based on more user preference data to optimize the decision output model. At the same time, multiple temperature sensors can be added to different areas of the vehicle cabin to monitor temperature changes in various areas in real time to achieve vehicle cabin optimization.

[0210] In the above embodiment, the vehicle cockpit's decision output for a preset function in the current scenario is obtained. Based on this decision, a first-dimension score is calculated to assess the degree of match between the decision output and the user's desired decision for the preset function. This accurately assesses user satisfaction with certain function adjustment decisions in the vehicle cockpit, facilitating subsequent optimization of the vehicle cockpit's decision output capabilities based on the first-dimension score. Furthermore, a second-dimension score is obtained to assess the incremental learning efficiency of the vehicle cockpit's decision output model, and a third-dimension score is obtained to assess the vehicle cockpit's anti-interference ability. Evaluating the incremental learning efficiency of the decision output model helps measure the vehicle cockpit's ability to quickly adapt to different user habits and new scenario changes. Evaluating the vehicle cockpit's anti-interference ability effectively assesses the vehicle cockpit's ability to maintain normal functionality and stable decision output in the face of complex environmental factors or other abnormal situations. A comprehensive score is generated based on the first, second, and third-dimension scores. This comprehensive score fully reflects the dynamic adaptability of the vehicle cockpit and provides a basis for developers to conduct targeted optimization of the vehicle cockpit, helping to enhance its intelligence and dynamic adaptability.

[0211] Figure 3 This is a schematic diagram of the evaluation process of a vehicle cabin evaluation system provided in an embodiment of the present application.

[0212] For example, Figure 3 As shown, the evaluation system 300 may include a data acquisition module 301 , a feature extraction and model updating module 302 , a dynamic evaluation calculation module 303 and a decision output module 304 .

[0213] The data acquisition module 301 is typically used to collect user interaction data and environmental data within the vehicle cabin, and is also used to transmit the collected data to the feature extraction and model updating module 302. The user interaction data refers to data generated when a user interacts with the vehicle cabin, such as when a user adjusts the air conditioning temperature using the vehicle's air conditioning control buttons; the environmental data refers to environmental data within the vehicle cabin, including ambient temperature, humidity, and other data.

[0214] The feature extraction and model updating module 302 is used to clean and extract features from the collected user interaction data and environmental data, obtaining data in a specific format that can be recognized by the model or vehicle controller, and then passing the processed data to the dynamic evaluation calculation module 303. For example, the processed data can be vector data converted from the user interaction data.

[0215] The feature extraction and model updating module 302 is further configured to optimize and update the decision output model in the vehicle cockpit according to the feedback result of the decision output module 304 .

[0216] The dynamic evaluation calculation module 303 is used to evaluate the vehicle cabin and output a first dimension score, a second dimension score, and a third dimension score for evaluating the vehicle cabin. The dynamic evaluation calculation module may include the vehicle-mounted computing unit mentioned above.

[0217] Among them, the first dimension scoring is specifically to evaluate the contextual relevance of the decision to be evaluated output by the decision output model; the second dimension scoring is specifically to evaluate the incremental learning efficiency of the decision output model; and the third dimension scoring is specifically to evaluate the anti-interference ability of the vehicle cabin.

[0218] The decision output module 304 is used to feed back the evaluation result of the vehicle cockpit to the feature extraction and model updating module 302, so that the feature extraction and model updating module 302 can optimize and update the decision output model in the vehicle cockpit according to the comprehensive scoring result of the vehicle cockpit.

[0219] The decision output module 304 is also used to output the optimal decision to the user. The optimal decision is the decision output by the optimized decision output model. The optimal decision is used to adjust the control parameters of the corresponding function in the vehicle cabin in the current scenario.

[0220] Based on the above Figure 3In the system, after the decision output model of the vehicle cockpit outputs the decision to be evaluated, the system's data acquisition module can collect user interaction data and environmental data. This collected data is then passed to the feature extraction and model update module, which performs data cleaning and feature extraction on the user interaction data and environmental data, respectively. The feature extraction and model update module then passes the processed data, including the context vector, to the dynamic evaluation and calculation module. The dynamic evaluation and calculation module evaluates the vehicle cockpit based on the decision to be evaluated output by the decision output model, certain parameters of the decision output model, and the processed data, and outputs first, second, and third dimension scores for the vehicle cockpit to the decision output module. After receiving the vehicle cockpit evaluation results, the decision output module can feed them back to the feature extraction and model update module, allowing it to update and optimize the decision output model based on the evaluation results. Finally, the decision output module can output the optimal decision to the user based on the updated and optimized decision output model.

[0221] In this system, the data acquisition module collects user interaction and environmental data, covering both human-machine interaction and environmental status during cockpit use. This provides a rich and comprehensive data foundation for subsequent analysis, more accurately reflecting the actual cockpit operation. The feature extraction and model update module cleans and extracts features from the collected data, removing interference factors such as noise and outliers, extracting key features, and improving data quality, making subsequent analysis and evaluation based on this data more accurate and reliable. The dynamic evaluation module uses this multifaceted information to generate scores for the first, second, and third dimensions of the vehicle cockpit, enabling quantitative evaluation from various perspectives. The decision output module feeds the evaluation results back to the feature extraction and model update module to update and optimize the decision output model, forming a closed-loop feedback system. Based on the evaluation results, model deficiencies can be identified, enabling targeted adjustments to model parameters and structural improvements, ensuring the model continuously adapts to new data and actual needs, and improving the accuracy and reliability of decision output. Finally, based on the updated and optimized model, the optimal decision is output, providing users with a more tailored cockpit control strategy, enhancing the intelligence and user-friendly nature of the vehicle cockpit and improving the user experience.

[0222] Figure 4 It is a structural schematic diagram of a vehicle cabin evaluation device provided in an embodiment of the present application.

[0223] For example, Figure 4 As shown, the apparatus 400 includes:

[0224] The acquisition module 401 is used to obtain the decision to be evaluated output by the vehicle cabin for the preset function in the current scenario.

[0225] The decision to be evaluated is used to adjust the control parameters of the preset function.

[0226] The calculation module 402 is configured to calculate a first dimension score for evaluating the vehicle cabin based on the decision to be evaluated, and obtain a second dimension score and a third dimension score for evaluating the vehicle cabin.

[0227] Among them, the first dimension scoring is used to evaluate the degree of match between the decision to be evaluated and the user's expected decision for the preset function; the second dimension scoring is used to evaluate the incremental learning efficiency of the decision output model in the vehicle cockpit; the third dimension scoring is used to evaluate the anti-interference ability of the vehicle cockpit.

[0228] The generating module 403 is configured to generate a comprehensive score for the vehicle cabin based on the first dimension score, the second dimension score, and the third dimension score.

[0229] In one possible implementation, the calculation module includes a first calculation unit, which is specifically used to: process the decision to be evaluated to obtain a decision vector to be evaluated; obtain the user's expected decision for the preset function, and process the expected decision to obtain an expected decision vector; obtain the preset deviation allowed when adjusting the control parameters of the preset function based on the decision to be evaluated; and calculate the first dimension score based on the decision vector to be evaluated, the expected decision vector and the preset deviation.

[0230] In one possible implementation, the first computing unit includes an acquisition subunit, which is specifically used to: after the vehicle cockpit executes the decision to be evaluated, obtain the interaction data between the user in the vehicle cockpit and the vehicle cockpit in response to the decision to be evaluated; based on the interaction data, determine the user's expected decision for the preset function.

[0231] In one possible implementation, the calculation module includes a second calculation unit, which is specifically used to: obtain interactive command data, environmental data and functional feedback data in the vehicle cabin; train the decision output model based on the interactive command data, environmental data and functional feedback data, and store the trained decision output model in the on-board processing unit in the vehicle cabin; obtain the actual learning rate of the decision output model stored in the on-board processing unit in the vehicle cabin for the user's expected decision on the preset function during the training process, and obtain the number of training rounds when the decision output model training is completed; based on the scoring calculation algorithm built into the on-board processing unit, combined with the actual learning rate and the number of training rounds, calculate the second dimension score.

[0232] In one possible implementation, the calculation module includes a third calculation unit, which is specifically used to: obtain multiple sets of sample environmental data in the vehicle cabin when no interference conditions are applied, and process each set of sample environmental data in the multiple sets of sample environmental data to obtain multiple sets of sample environmental vectors; obtain actual environmental data in the vehicle cabin after the interference conditions are applied, and process the actual environmental data to obtain actual environmental vectors; and calculate a third dimension score based on the multiple sets of sample environmental vectors and the actual environmental vectors.

[0233] In one possible implementation, the third calculation unit includes a calculation subunit, which is specifically used to: calculate a sample mean vector based on multiple groups of sample environment vectors; calculate the difference between the sample mean vector and the actual environment vector to obtain a difference vector; calculate a sample covariance matrix based on the sample mean vector; and calculate a third dimension score based on the difference vector and the sample covariance matrix.

[0234] In one possible implementation, the generation module is further specifically used to: perform weighted fusion of the first dimension score, the second dimension score, and the third dimension score according to preset weights to generate a comprehensive score for the vehicle cabin.

[0235] Figure 5 It is a structural schematic diagram of a vehicle provided in an embodiment of the present application.

[0236] For example, Figure 5 As shown, the vehicle 500 includes: a memory 501 and a processor 502, wherein the memory 501 stores an executable program code 5011, and the processor 502 is used to call and execute the executable program code 5011 to perform a vehicle cabin evaluation method.

[0237] In addition, an embodiment of the present application also protects a device, which may include a memory and a processor, wherein the memory stores executable program code, and the processor is used to call and execute the executable program code to perform a vehicle cabin evaluation method provided by an embodiment of the present application.

[0238] In this embodiment, the device can be divided into functional modules based on the above-described method examples. For example, each functional module can be mapped to a specific functional module, or two or more functions can be integrated into a single processing module. The integrated module can be implemented in hardware. It should be noted that the module division in this embodiment is illustrative and represents only a logical functional division. In actual implementation, other division methods may be used.

[0239] In the case of dividing each functional module into corresponding functional modules, the device may further include an acquisition module, a calculation module, a generation module, etc. It should be noted that all relevant contents of each step involved in the above method embodiment can be referred to the functional description of the corresponding functional module and will not be repeated here.

[0240] It should be understood that the device provided in this embodiment is used to execute the above-mentioned vehicle cabin evaluation method, and thus can achieve the same effect as the above-mentioned implementation method.

[0241] In the case of an integrated unit, the device may include a processing module and a storage module. When the device is used in a vehicle, the processing module may be used to control and manage the vehicle's movements. The storage module may be used to support the vehicle's execution of relevant program codes and data.

[0242] The processing module may be a processor or controller that implements or executes the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. The processor may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a digital signal processing system (DSP) and a microprocessor, and the storage module may be a memory.

[0243] In addition, the device provided in the embodiments of the present application can specifically be a chip, component or module, and the chip may include a connected processor and memory; wherein the memory is used to store instructions, and when the processor calls and executes the instructions, the chip can execute a vehicle cabin evaluation method provided in the above embodiment.

[0244] This embodiment also provides a computer-readable storage medium, which stores computer program code. When the computer program code runs on a computer, the computer executes the above-mentioned related method steps to implement a vehicle cabin evaluation method provided by the above embodiment.

[0245] This embodiment further provides a computer program product. When the computer program product is run on a computer, it enables the computer to execute the above-mentioned related steps to implement the vehicle cabin evaluation method provided by the above embodiment.

[0246] Among them, the device, computer-readable storage medium, computer program product or chip provided in this embodiment are all used to execute the corresponding methods provided above. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding methods provided above, and will not be repeated here.

[0247] Through the description of the above implementation methods, technical personnel in the relevant field can understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0248] In the embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of modules or units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0249] The above content is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A vehicle cabin evaluation method, characterized in that: The method comprises: Obtaining a decision to be evaluated output by the vehicle cockpit for a preset function in a current scenario; wherein the decision to be evaluated is used to adjust a control parameter of the preset function; Based on the decision to be evaluated, a first dimension score for evaluating the vehicle cockpit is calculated, and a second dimension score and a third dimension score for evaluating the vehicle cockpit are obtained; wherein the first dimension score is used to evaluate the degree of match between the decision to be evaluated and the user's expected decision for the preset function; the second dimension score is used to evaluate the incremental learning efficiency of the decision output model in the vehicle cockpit; and the third dimension score is used to evaluate the anti-interference ability of the vehicle cockpit; A comprehensive score for the vehicle cabin is generated based on the first dimension score, the second dimension score, and the third dimension score.

2. The method according to claim 1, characterized in that The calculating, based on the decision to be evaluated, a first dimension score for evaluating the vehicle cabin, includes: Processing the decision to be evaluated to obtain a decision vector to be evaluated; Obtaining the user's expected decision for the preset function, and processing the expected decision to obtain an expected decision vector; Obtaining a preset deviation allowed when adjusting a control parameter of the preset function based on the decision to be evaluated; A first dimension score is calculated based on the decision vector to be evaluated, the expected decision vector, and the preset deviation.

3. The method according to claim 2, characterized in that The obtaining of the user's expected decision regarding the preset function includes: After the vehicle cockpit executes the decision to be evaluated, acquiring interaction data between the user in the vehicle cockpit and the vehicle cockpit in response to the decision to be evaluated; Based on the interaction data, an expected decision of the user with respect to the preset function is determined.

4. The method according to claim 1, wherein The obtaining of a second dimension score for evaluating the vehicle cabin includes: Acquiring interactive command data, environmental data, and functional feedback data within the vehicle cabin; Training the decision output model based on the interaction instruction data, the environmental data, and the functional feedback data, and storing the trained decision output model in an onboard processing unit in the vehicle cabin; Obtaining an actual learning rate of the decision output model stored in the vehicle processing unit in the vehicle cabin for the user's expected decision for the preset function during training, and obtaining a number of training rounds after which training of the decision output model is completed; Based on the scoring calculation algorithm built into the onboard processing unit, combined with the actual learning rate and the number of training rounds, the second dimension score is calculated.

5. The method according to claim 1, characterized in that The obtaining of a third dimension score for evaluating the vehicle cabin includes: Acquiring multiple sets of sample environmental data in the vehicle cabin when no interference condition is applied, and processing each set of sample environmental data in the multiple sets of sample environmental data to obtain multiple sets of sample environmental vectors; Acquiring actual environmental data within the vehicle cabin after the interference condition is applied, and processing the actual environmental data to obtain an actual environmental vector; The third dimension score is calculated based on the multiple groups of sample environment vectors and the actual environment vector.

6. The method according to claim 5, characterized in that The calculating the third dimension score based on the multiple groups of sample environment vectors and the actual environment vector includes: Based on the multiple groups of sample environment vectors, a sample mean vector is calculated; Calculating the difference between the sample mean vector and the actual environment vector to obtain a difference vector; Based on the sample mean vector, a sample covariance matrix is ​​calculated; The third dimension score is calculated based on the difference vector and the sample covariance matrix.

7. The method according to claim 1, characterized in that Generating a comprehensive score for the vehicle cabin based on the first dimension score, the second dimension score, and the third dimension score includes: According to preset weights, the first dimension score, the second dimension score, and the third dimension score are weightedly fused to generate a comprehensive score for the vehicle cabin.

8. A vehicle cabin evaluation device, characterized in that: The device comprises: An acquisition module, configured to acquire a decision to be evaluated output by the vehicle cockpit for a preset function in a current scenario; wherein the decision to be evaluated is used to adjust a control parameter of the preset function; a calculation module, configured to calculate, based on the decision to be evaluated, a first dimension score for evaluating the vehicle cockpit, and obtain a second dimension score and a third dimension score for evaluating the vehicle cockpit; wherein the first dimension score is used to evaluate the degree of match between the decision to be evaluated and the user's expected decision for the preset function; the second dimension score is used to evaluate the incremental learning efficiency of the decision output model in the vehicle cockpit; and the third dimension score is used to evaluate the anti-interference capability of the vehicle cockpit; A generating module is used to generate a comprehensive score for the vehicle cabin based on the first dimension score, the second dimension score and the third dimension score.

9. A vehicle, characterized in that: The vehicle comprises: a memory for storing executable program code; A processor is configured to call and run the executable program code from the memory, so that the vehicle executes the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed, the method according to any one of claims 1 to 7 is implemented.

Images