Information processing apparatus, information processing method, and information processing program
The information processing device uses RNN and random forest models to analyze occupant input/output data, addressing the challenge of specifying tire performance factors and timing in sensory evaluations, thereby enhancing tire evaluation clarity.
Patent Information
- Application Number
- JP2023086529
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-25
- Publication Date
- 2024-12-05
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing technologies struggle to specify the basis for sensory evaluations of tire performance, such as steering stability, as they rely on occupant feedback during vehicle operation, making it difficult to identify contributing factors and timing.
An information processing device and method that utilizes a recurrent neural network (RNN) and a random forest model to analyze time-series data from vehicle occupants, dividing vehicle travel into intervals and extracting feature information to derive the relevance of input/output to tire evaluation.
Enables estimation of tire evaluation by identifying key factors and timing contributing to sensory evaluations, providing a clearer basis for tire performance assessment.
Smart Images

Figure 2024169231000001_ABST
Abstract
Description
[Technical field]
[0001] The present disclosure relates to an information processing device, an information processing method, and an information processing program. [Background technology]
[0002] There is known an information processing device that uses time series data from a sensor to calculate the importance of the output of a model that uses the time series data as an input (see, for example, Patent Document 1). In this technology, time series data related to a processing target is acquired, and a plurality of time series data with different time widths obtained by cutting out time series data of a predetermined time width from the acquired time series data are set as groups. Then, for each set group, a score related to the importance of the output value of a model that uses the time series data as an input value is calculated.
[0003] Also, a generation device is known that generates a representative feature vector by selecting a feature amount for each time interval from a sequence of feature vectors for each frame (see, for example, Patent Document 2). This technology includes a means for selecting, for each time interval, feature vectors of multiple frames included in the time interval from the sequence of feature vectors for each frame, and a means for selecting, for each time interval, feature amounts of different dimensions of the feature vector from the feature vectors of different frames in the selected time interval, and generating a representative feature vector that represents the time interval. [Prior art documents] [Patent documents]
[0004] [Patent Document 1] Patent Publication No. 2020-166442 [Patent Document 2] International Publication No. 2010 / 087125 Summary of the Invention [Problem to be solved by the invention]
[0005] Incidentally, the performance of a tire mounted on a vehicle, for example, the steering stability performance, may be subject to a sensory evaluation by a vehicle occupant while the vehicle is traveling. However, although this sensory evaluation is obtained as a result of the vehicle occupant's traveling, it has been difficult to identify the basis for the result of the sensory evaluation, for example, what information contributes at what timing. In addition, with recent technological advances, a model for predicting the result of the sensory evaluation has been developed. It is possible to predict the result of the sensory evaluation using this model. However, even if this model is used, it has been difficult to identify the basis for the predicted result of the sensory evaluation, for example, what information contributes at what timing. Therefore, there is room for improvement in evaluating the performance of a tire mounted on a vehicle.
[0006] The present disclosure aims to provide an information processing device, an information processing method, and an information processing program capable of estimating information related to tire evaluation taking into account input and output to a vehicle occupant that changes as the vehicle travels. [Means for solving the problem]
[0007] A first aspect of the present disclosure is a method for manufacturing a semiconductor device comprising: an acquisition unit that acquires a series of a plurality of pieces of time-series information that change in a time series manner in response to the running of a vehicle on which the tire is mounted and that indicates inputs and outputs to an occupant of the vehicle; an extraction unit that uses a first model in which the plurality of series of time-series information acquired by the acquisition unit is input and first evaluation information indicative of an evaluation of the tire is output, to extract a plurality of pieces of feature information indicative of features for the input / output in each of a plurality of driving scenes obtained by dividing the driving of the vehicle into a plurality of time intervals; a derivation unit that uses a second model in which the plurality of pieces of feature information extracted by the extraction unit are input and second evaluation information indicative of an evaluation of the tire is output, to derive association information indicative of an association between the second evaluation information and the input / output in each of the plurality of traveling scenes; The information processing device is provided with:
[0008] A second aspect is the information processing device of the first aspect, The driving scene is a driving state obtained by extracting at least a part of a driving state of the vehicle, The extraction unit extracts, as the plurality of pieces of feature information, information indicating an internal state of the first model at a plurality of different predetermined timings corresponding to each of the plurality of driving scenes.
[0009] A third aspect is an information processing device according to the first or second aspect, The derivation unit derives, as the relevance information, an importance indicating a degree of relevance of the input / output to the second evaluation information, based on information indicating an internal state of the second model in each of the plurality of driving scenes.
[0010] A fourth aspect is an information processing device according to any one of the first to third aspects, the time-series information includes arm information indicating inputs and outputs related to the arms of the occupant, waist information indicating inputs to the waist of the occupant, and head information indicating inputs to the head of the occupant; The arm information is data indicating at least one of steering angle and steering torque, the waist information is data indicating the position of the center of gravity of the occupant on the seat, and the head information is data indicating head acceleration and angular velocity.
[0011] A fifth aspect is an information processing device according to any one of the first to fourth aspects, the first model is a recurrent neural network model; The second model is a random forest model.
[0012] A sixth aspect is the information processing device of the fifth aspect, the information indicating the internal state of the first model is a neuron value in the recurrent neural network model; The information indicating the internal state of the second model is information indicating the result of conditional branching in the random forest model.
[0013] The seventh aspect is The computer acquiring a series of a plurality of time-series information items which change over time in response to the running of a vehicle on which the tire is mounted and which indicate inputs and outputs to an occupant of the vehicle; extracting a plurality of pieces of feature information indicating features of the input and output in each of a plurality of driving scenes obtained by dividing the driving of the vehicle into a plurality of time intervals using a first model that receives the plurality of pieces of acquired time-series information as an input and outputs first evaluation information indicating an evaluation of the tire; deriving association information indicating an association between the second evaluation information and the input / output in each of the plurality of driving scenes using a second model that receives the extracted plurality of pieces of feature information as an input and outputs second evaluation information indicating an evaluation of the tire; This is an information processing method for processing the above.
[0014] The eighth aspect is To your computer acquiring a series of a plurality of time-series information items which change over time in response to the running of a vehicle on which the tire is mounted and which indicate inputs and outputs to an occupant of the vehicle; extracting a plurality of pieces of feature information indicating features of the input and output in each of a plurality of driving scenes obtained by dividing the driving of the vehicle into a plurality of time intervals using a first model that receives the plurality of pieces of acquired time-series information as an input and outputs first evaluation information indicating an evaluation of the tire; deriving association information indicating an association between the second evaluation information and the input / output in each of the plurality of driving scenes using a second model that receives the extracted plurality of pieces of feature information as an input and outputs second evaluation information indicating an evaluation of the tire; It is an information processing program that processes the following:
[0015] The time-series information may include vehicle information indicating the behavior of the vehicle. Effect of the Invention
[0016] According to the present disclosure, it is possible to estimate information related to evaluation of a tire by taking into account input and output to a vehicle occupant that changes as the vehicle travels. [Brief description of the drawings]
[0017] [Figure 1] 1 is a block diagram showing a schematic configuration of an estimation device according to an embodiment. [Diagram 2] FIG. 2 is a conceptual diagram showing the correlation between an occupant and a vehicle. [Diagram 3] FIG. 2 is a diagram showing an example of the arrangement of measurement sensors. [Figure 4] FIG. 1 is a diagram illustrating an example of vehicle travel. [Diagram 5] FIG. 4 is a diagram showing an example of a change in steering angle when changing lanes; [Figure 6] FIG. 2 is a diagram illustrating a conceptual configuration of an estimation model. [Figure 7] FIG. 2 is a diagram showing a conceptual configuration of a first model. [Figure 8] FIG. 13 is a diagram showing a conceptual configuration of a second model. [Figure 9] FIG. 2 is a diagram illustrating an example of an electrical configuration of the estimation device. [Figure 10] 13 is a flowchart showing an example of the flow of an estimation process. [Figure 11] FIG. 1 is a diagram showing distribution characteristics of measurement data of a tire used in verification. [Figure 12] 11 is a conceptual diagram showing the results of subjective sensory evaluation by passengers. FIG. [Figure 13] 11 is a conceptual diagram showing the correspondence between the result of the sensory evaluation by an occupant and the estimation result of an estimation model. FIG. [Figure 14] 11 is a conceptual diagram showing the relationship between the importance of a feature amount and the amount of input and output to an occupant. FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0018] Hereinafter, embodiments for implementing the technology of the present disclosure will be described in detail with reference to the drawings. In addition, components and processes that have the same action and function are given the same reference numerals throughout the drawings, and duplicated explanations may be omitted as appropriate. In addition, the present disclosure is not limited to the following embodiments, and can be implemented with appropriate modifications within the scope of the purpose of the present disclosure.
[0019] In the present disclosure, a tire is an elastically deformable member attached to a vehicle, which is an example of a moving body, and is a concept including a member that can be attached and detached from the vehicle. The steering stability performance of a tire is an example of tire performance, and is also an example of a characteristic related to the motion performance of a vehicle. Tire performance is also a concept including a characteristic related to the performance of a vehicle to which the tire is attached. In the present disclosure, feature information is a concept indicating a characteristic of an occupant's input / output related to the tire's evaluation regarding the running of the tire. In the present disclosure, related information is a concept indicating a relationship between the tire evaluation and the occupant's input / output. The related information is a concept indicating the importance of the feature information, and information indicating the degree of association between the tire evaluation and the occupant's input / output can be applied as the importance. Time-series information can be applied as information indicating input / output to the occupant, and as an example of information indicating input / output, data indicating a steering angle and a steering torque can be applied as information indicating input / output related to the occupant's arms, data indicating a center of gravity position on the seat can be applied as information indicating an input to the occupant's waist, and head acceleration and angular velocity can be applied as information indicating an input to the occupant's head.
[0020] In addition, in this disclosure, a vehicle is a concept that includes a moving body such as an automobile having tires.
[0021] <Estimation device> 1 is a diagram showing an example of the configuration of an estimation device 1 capable of executing an estimation process for estimating information related to tire evaluation. The estimation device 1 includes an estimation unit 5. The estimation device 1 can be realized by a computer including a CPU as an execution device that executes the process described below.
[0022] Information indicating inputs and outputs to and from a vehicle occupant that changes according to a traveling state 3 indicating the traveling of the vehicle 2 is input as input data to the estimation unit 5. The information indicating the inputs and outputs to and from the occupant includes information indicating the behavior of the occupant due to the movement of multiple parts of the occupant and the exchange of forces. This information indicating the inputs and outputs to and from the occupant is a series of time-series information, and time-series measurement data 4 obtained by measuring the inputs and outputs to and from the occupant (occupant behavior) by a measurement sensor is applied. The measurement data 4 is acquired by measurement sensors 21, 22, 23 (FIG. 3) that function as a detection unit 118 (FIG. 9).
[0023] The estimation unit 5 estimates information related to the evaluation of the tire during vehicle travel by an estimation process described later using the estimation model 51 based on a series of input time-series information (i.e., time-series measurement data), and outputs the information as output data. In this embodiment, a case will be described in which evaluation data as first output data indicating the evaluation of the tire and associated data as second output data relating to the input / output of the occupant and the evaluation of the tire are applied as information related to the evaluation of the tire during vehicle travel. That is, the estimation unit 5 can estimate evaluation data 6 indicating the evaluation of the tire for the travel of the vehicle 2 using the time-series measurement data 4, and can also estimate associated data 7 related to the evaluation of the tire. The estimation model 51 includes a first model 51A and a second model 51B. The estimation model 51 including the first model 51A and the second model 51B will be described later.
[0024] Specifically, the estimation unit 5 is a functional unit that estimates information related to the evaluation of a tire. The estimation unit 5 is connected to a vehicle 2 equipped with a measurement sensor (FIG. 3), and receives time-series measurement data 4 as an input. The time-series measurement data 4 corresponds to measurement values obtained by measuring the input and output of an occupant of the vehicle 2 in a time series. The estimation unit 5 estimates and outputs output data related to the evaluation of a tire using an estimation model 51 based on the time-series measurement data 4 indicating the input and output of the occupant measured by the measurement sensor. The input and output of the occupant of the vehicle 2 changes according to a running state 3 indicating the running of the vehicle 2. The estimation unit 5 outputs, as a result of estimation using the estimation model 51, evaluation data 6 indicating an evaluation of the tire as first output data, and associated data 7 relating to the input and output of the occupant and the evaluation of the tire.
[0025] Here, the correlation between the occupant and the vehicle will be described. When evaluating tire performance that affects steering stability and the like, a sensory evaluation by a vehicle occupant may be performed. The sensory evaluation by the occupant is assumed to be based on (for example, dominant) the amount felt by the occupant (hereinafter, referred to as sensory amount). For this reason, in this embodiment, data related to input and output to the occupant is applied as a series of time-series information (i.e., time-series measurement data 4) input to the estimation unit 5.
[0026] Fig. 2 is a conceptual diagram showing a correlation between an occupant and the vehicle 2. Fig. 2 shows an example of a correlation regarding input and output to the occupant when the occupant steers the vehicle 2.
[0027] The occupant steers the vehicle 2 by operating the steering wheel or the like while assuming a target driving state. The steering is output from the occupant by operating the steering wheel or the like, and is reflected in the steering angle. In the vehicle 2, the behavior of the vehicle 2 is reflected according to the steering angle by the occupant, and in response, a steering torque is generated in the steering wheel or the like, and is input to the occupant. These inputs and outputs to the occupant can be measured as sensory quantities in the arms of the occupant.
[0028] In addition, the vehicle behavior is affected by the steering by the occupant, and is reflected in the actual driving state, affecting the occupant's behavior. In the example shown in the figure, the occupant's head and waist are applied as an example of the parts of the occupant's behavior that are affected. Specifically, the occupant sitting on the seat changes in posture and position depending on the actual driving state, and is input to the occupant. This occupant's posture change and position change can be measured as the occupant's center of gravity position and the change in center of gravity position (or either one) as a sensory amount of the occupant's waist. Note that the occupant's center of gravity position may be measured by the moving speed and moving acceleration of the center of gravity position (or either one).
[0029] In addition, the behavior of the occupant's head changes depending on the actual driving state. For example, a change in the position of the occupant's head occurs and is input to the occupant. In this embodiment, the acceleration and angular velocity of the occupant's head are applied as the behavior of the occupant's head. The acceleration and angular velocity of the occupant's head, which are input to the occupant, can be measured as a sensory amount of the occupant's head. At least one of the acceleration and angular velocity of the occupant's head can be applied as the behavior of the occupant's head. Note that the measurement value obtained by measuring the position and position change (or either one) of the occupant's head may be applied as the behavior of the occupant's head. In addition, the occupant's line of sight and field of view may be included as the behavior of the occupant's head.
[0030] The target driving state assumed by the occupant described above can be considered as a driving condition. In other words, the target driving state can be applied as a driving condition indicating that the steering should be performed assuming the driving state desired by the occupant.
[0031] While the vehicle 2 is traveling, the occupant performs a sensory evaluation based on the input and output to the occupant described above (i.e., the amount of force that corresponds to the measurement data 4 and the amount of movement of the occupant that they feel), and provides the result as a sensory evaluation value.
[0032] In the above, the case where the arm, waist, and head of the occupant are applied as an example of a part that affects the behavior of the occupant, i.e., the input and output to the occupant, has been described, but the present disclosure is not limited thereto. For example, the leg of the occupant may be included, or other parts may be included. In addition, each part may be subdivided, or a combination of a plurality of these parts may be used as the application part.
[0033] Next, the measurement sensor will be described. The measurement sensor is capable of acquiring a series of time-series information (i.e., time-series measurement data 4). The measurement sensors are arranged corresponding to the types of behavior of the occupant, and measure inputs and outputs to the occupant as the behavior of the occupant aboard the vehicle 2. 3 is a diagram showing an example of the arrangement of the measurement sensors. In this embodiment, measurement sensors 21, 22, and 23 that obtain time-series information corresponding to the type of input / output to the occupant, which is the type of behavior of the occupant, are used as an example of the measurement sensors.
[0034] The measurement sensor 21 is a sensor that measures the behavior of the occupant's arm, and the measurement sensor 21 can obtain measurement data that indicates information that indicates input and output related to the occupant's arm. Data that indicates a steering angle and / or steering torque can be applied to the measurement data measured by the measurement sensor 21. For example, the measurement sensor 21 may be a sensor that measures a steering angle and / or steering torque.
[0035] The measurement sensor 22 is a sensor that measures the behavior of the occupant's waist, and the measurement sensor 22 makes it possible to obtain measurement data that indicates information about input related to the occupant's waist. The measurement data measured by the measurement sensor 22 can be data that indicates the load and center of gravity position of the occupant on the seat. For example, the measurement sensor 22 can be a sensor that measures the load and center of gravity position of the occupant on the seat. One example of the sensor is a six-component force gauge, and for example, multiple (for example, four) six-component force gauges can be installed on the seat or the like, and the load and center of gravity position can be obtained from the values obtained.
[0036] The measurement sensor 23 is a sensor that measures the behavior of the occupant's head, and the measurement sensor 23 makes it possible to obtain measurement data that indicates information about input related to the occupant's head. The measurement data measured by the measurement sensor 23 can be data that indicates the acceleration and angular velocity (or either one) of the occupant's head. For example, the measurement sensor 23 can be a sensor that measures the acceleration and angular velocity (or either one) of the occupant's head. As an example of the sensor, a glasses-type device can be used as a motion sensor, and the acceleration and angular velocity of the motion sensor can be measured.
[0037] The above-mentioned measurement sensors 21, 22, 23 may be dedicated measurement sensors such as position detection sensors, motion sensors, and torque sensors, or may measure from images of the occupants. The measurement sensors are not limited to directly measuring the behavior of the occupants, and may obtain data detectable by the vehicle 2, such as steering angle and steering torque, and use the data as measurement data. In other words, the measurement data 4 as time-series information can include vehicle information indicating the behavior of the vehicle 2.
[0038] Next, the traveling state of the vehicle 2 will be described. Meanwhile, the vehicle 2 runs with its behavior changing from moment to moment by controlling the vehicle speed, steering, etc., according to the occupant's instructions and the state of the road, etc. For example, the vehicle 2 runs in various running conditions such as turning right or left and changing lanes. Therefore, the occupant performs a sensory evaluation in response to various running conditions. However, it has been difficult to identify factors that contribute to the result of the sensory evaluation. Therefore, in this embodiment, a plurality of different timings related to the result of the occupant's sensory evaluation (for example, the basis for judgment) in the running state of the vehicle 2 are considered.
[0039] Fig. 4 is a diagram showing a schematic example of a driving (driving state) of the vehicle 2. The example in the figure shows a driving state when changing lanes from one lane (the right lane in the figure) to the other lane (the left lane in the figure) on a two-lane road. Fig. 5 is a diagram showing a change in steering angle when changing lanes shown in Fig. 4.
[0040] Specifically, the occupant performs a lane change by rotating the vehicle's steering wheel (not shown) clockwise and counterclockwise. The lane change can be classified into three driving scenes, for example, an early stage, a middle stage, and a late stage. A driving scene is a driving state extracted from at least a part of the driving of the vehicle 2. The early stage driving scene is a driving state at the stage where the lane change is started. The middle stage driving scene is a driving state at the stage where the vehicle 2 is moved by the lane change. The late stage driving scene is a driving state at the stage where the posture of the vehicle 2 is adjusted in the later stage of the lane change. It is preferable that the early stage driving scene and the middle stage driving scene include an inflection point of a maximum value or a minimum value in the steering angle characteristic.
[0041] In this embodiment, a time section up to a predetermined timing is defined as a section corresponding to a driving scene. Specifically, a time section up to a timing ts is defined as a section Ts corresponding to an early driving scene, a time section up to a timing tm is defined as a section Tm corresponding to a mid-stage driving scene, and a time section up to a timing te is defined as a section Te corresponding to a later driving scene. Hereinafter, the sections Ts, Tm, and Te may be referred to as driving scenes Ts, Tm, and Te. Note that the above-mentioned predetermined timings may be determined from experimental values or statistical values, for example, timings including inflection points of maximum and minimum values in the above-mentioned steering angle characteristics. These timings may be changed as parameters using behavior values of the vehicle 2, such as the vehicle speed, assuming a case where they vary depending on the environment during driving of the vehicle 2, for example, the vehicle speed.
[0042] In this way, the running state of the vehicle 2 is a part of a series of runs in which the vehicle runs, and is a state in which different running scenes continue in a time series. Therefore, as will be described in detail later, by using the measurement data 4 in each running scene, it becomes possible to identify factors that contribute to the results of the sensory evaluation in each running scene.
[0043] The driving scenes are not limited to the above-mentioned three driving scenes of the early, middle, and late stages, and may be, for example, two or more driving scenes. In addition, the multiple driving scenes may be arranged so that adjacent driving scenes partially overlap, or adjacent driving scenes may be separated from each other. In addition, the above-mentioned timings ts, tm, and te may be timings that represent each driving scene, and are not limited to the above-mentioned timings ts, tm, and te. The timings ts, tm, and te are examples of multiple different timings in the technology of the present disclosure.
[0044] <Estimation model> Next, the estimation model 51 will be described. The estimation model 51 is a machine learning model that models an algorithm related to the sensory evaluation of an occupant who boards the vehicle 2. In other words, the estimation model 51 is a machine learning model that estimates output data (evaluation data 6 and related data 7) indicating information related to the evaluation of tires in a vehicle running state, from the behavior (i.e., input / output) of the occupant of the vehicle 2 indicated by the input time-series measurement data 4.
[0045] FIG. 6 is a diagram showing a conceptual configuration of the estimation model 51. As shown in FIG. The estimation model 51 receives time-series measurement data 4 of the occupant's arms, waist, and head, and outputs evaluation data 6 indicating a sensory evaluation value as first output data. The estimation model 51 can also output related data 7 as second output data. The evaluation data 6 is an example of at least one of the first evaluation information and the second evaluation information of the present disclosure, and the related data 7 is an example of the related information of the present disclosure. The time-series measurement data 4 of the arms, waist, and head of the occupant are an example of the arm information, waist information, and head information of the present disclosure.
[0046] The estimation model 51 takes into account time-series measurement data 4. The past history of the time-series measurement data 4 is important, and a recurrent neural network model using a recurrent neural network (RNN) can be applied as a model that enables the use of the past history. In this embodiment, an Echo State Network (ESN), which is an example of the RNN, is used. The ESN is also an example of a neural network called Reservoir Computing.
[0047] In this embodiment, the estimation model 51 includes a first model 51A and a second model 51B. The first model 51A and the second model 51B are linked to form the estimation model 51. An ESN is applied to this first model 51A. Note that, although details will be described later, the first model 51A linked to the second model 51B is a part of the ESN.
[0048] 7 is a diagram showing the conceptual configuration of an ESN applied to the first model 51A. The ESN is expressed as a collection of information on the weights (strengths) of connections between nodes (neurons) that constitute a neural network, and the internal state of the ESN can be obtained as state data indicating the value of each node (neuron). Hereinafter, a node may be referred to as a neuron.
[0049] The ESN applicable to the first model 51A can learn to receive measurement data 4 indicating input and output to the occupant as input, and to output an output value indicating the evaluation of the tire, which is a sensory evaluation value. That is, the ESN applicable to the first model 51A is a model that models the sensory evaluation of the occupant. In this embodiment, three indicators, good (suitable), normal (medium), and poor (unsuitable), are used as examples of output values indicating the evaluation of the tire.
[0050] The ESN can be applied to a model that models the sensory evaluation of the occupant, with the measurement data 4 as input and an output value indicating the evaluation of the tire, which is a sensory evaluation value, as output. The first model 51A uses a part of the ESN, as described later. In this embodiment, three indicators, good (suitable), normal (medium), and poor (unsuitable), are used as an example of the output value indicating the evaluation of the tire.
[0051] Specifically, the ESN applied to the first model 51A is composed of an input layer 510, an output layer 514, and an intermediate layer called a reservoir layer 512. The reservoir layer 512 includes randomly connected neurons, and as is known in reservoir computing, the weight Wres of the reservoir layer 512 and the weight Win of the connection between the input layer 510 and the reservoir layer 512 are not subject to learning and are randomly initialized in advance. This random structure makes it possible to form a state in which past information is propagated when appropriate parameters such as the number of neurons are set, and to represent long-term dependency. As will be described later, in this embodiment, the ESN applied to the first model 51A does not use the output layer 514, but uses the neuron configuration up to the reservoir layer 512 (specifically, only the neuron value indicating the internal state).
[0052] The above ESN (first model 51A) can be expressed by the following equations (1) and (2). x(t) = (1-α) x(t-1) + α f(Win u(t) + Wres x(t-1)) -(1) y(t) = fout(Wout x(t)) - (2)
[0053] Here, u(t) is the measurement data 4 which is the input data at time t, and in this embodiment, each measurement value which is the input, the measurement data 4, is treated as a scalar quantity. y(t) is the output data, and is expressed by the function fout. If the number of neurons in the reservoir layer 512 is N, x(t) is state data indicating the state quantity held by the neurons in the reservoir layer 512 at time t, and is an N-dimensional vector. Win is the weight of the connection between the input layer 510 and the reservoir layer 512, and is an N-dimensional vector. Wres is the weight of the reservoir layer 512, and is an NxN-dimensional matrix indicating the amount of connection between the neurons in the reservoir layer 512. α is called the Leaky Integrator (LI), and controls the ratio when the internal parameters are updated sequentially. In other words, α controls the forgetting speed of the neuron. Therefore, the ESN holds the internal state at each time step, and the internal state evolves over time according to the formula (1) including the above function f.
[0054] In this embodiment, an ESN is constructed for each type of measurement data 4, and a first model 51A is formed.
[0055] The model applied to the above-described first model 51A is not limited to the ESN, but may be another RNN, which is a regression model capable of utilizing past history based on the time-series measurement data 4.
[0056] 6, the estimation model 51 includes a memory 52 that stores state data at any timing for the ESN, which is the first model 51A. That is, the estimation model 51 can extract the internal state of the ESN at any timing and store it in the memory 52. In this embodiment, the internal state of the ESN is extracted at each of the timings ts, tm, and te that define the above-mentioned driving scenes, for example, the early driving scene, the middle driving scene, and the late driving scene of a lane change, and stored in the memory 52. Therefore, the internal state of the ESN is extracted for each type of measurement data 4 and stored in the memory 52.
[0057] The state data indicating the internal state of the ESN at these timings corresponds to the feature quantity indicating the feature of the measurement data 4 (i.e., time-series information) in each driving scene. The feature quantity, which is the state data, corresponds to information that quantifies the relationship between the measurement data 4 (i.e., time-series information) indicating the input and output to the occupant and the evaluation value for the tires, which is a sensory evaluation value. The feature amount is an example of feature information of the present disclosure.
[0058] Incidentally, in estimating an output using an ESN, the output layer 514 and its weights Wout are generally used. However, the first model 51A to which the ESN of this embodiment is applied does not use the output layer 514 and its weights Wout. Specifically, state data extracted at a predetermined timing and stored in the memory 52 is passed as an output to the second model 51B. That is, in this embodiment, the output layer 514 of the ESN and its weights Wout are not used, and the internal state of the neurons up to the reservoir layer 512 is used.
[0059] Next, the second model 51B of the estimation model 51 will be described.
[0060] The second model 51B receives as input the state data of the first model 51A based on the measurement data 4 indicating the input and output to the occupant, and outputs the evaluation data 6 which is the occupant's sensory evaluation value of the tire. The second model 51B is also a model that models the occupant's sensory evaluation. In this embodiment, three indicators, good (suitable), normal (medium), and poor (unsuitable), are used as an example of the evaluation data 6.
[0061] 8 is a diagram showing a conceptual configuration of the second model 51B. In this embodiment, a case will be described in which a random forest model is applied to the first model 51B as an example of a classification model.
[0062] The random forest model is a well-known technology and will not be described in detail here, but it is an ensemble learning algorithm that uses bagging and decision trees as learning devices. In this random forest model, multiple decision trees trained in parallel make predictions for an input, and the results of majority votes or average values are output.
[0063] For example, the second model 51B includes functional units for performing grouping, modeling, provisional estimation, and final estimation.
[0064] In the second model 51B, first, data is randomly sampled from the original data (state data) by bootstrap (a process called sampling with replacement) to generate a predetermined number (here, M) of data groups (grouping). Next, a decision tree 516 is generated for each of the M groups (modeling). The decision tree 516 is a model that selects an output by conditional branching (for example, if~then~). Next, an estimate is made once using the decision tree for each of the M groups (provisional estimation). Then, a majority vote (or average) of the M groups is taken to make a final prediction of the output data (evaluation value) (final prediction).
[0065] In the second model 51B to which this random forest model is applied, for example, each of the values of N pieces of original data (state data) is used as a feature, and some of the feature values are randomly selected and used for splitting the decision tree 516 in modeling. Then, the original data is slightly changed by bagging, and the feature values are slightly changed by randomness, thereby generating a plurality of decision trees 516 with low correlation with each other. In generating this decision tree 516, the split is performed by conditional branching, and the split is repeated using a predetermined index (for example, error or Gini impurity) that evaluates the split point. This index is a value according to the appropriateness of the conditional branching by the split point. Specifically, the smaller the index, the more appropriate the conditional branching. Therefore, the smaller the index, the higher the importance of the feature. Therefore, it is possible to derive the importance of the feature by calculating the degree of reduction in the index at the time of splitting for each feature of the decision tree 516 (for example, the total value). In other words, it is possible to derive it as related information indicating the importance indicating the degree of relevance of the input / output to the occupant to the evaluation data. The importance of a feature is an example of related information in the present disclosure.
[0066] The second model 51B of this embodiment includes a derivation unit 518 that derives the importance of the above-mentioned feature amount. Specifically, the value of the original data (state data) is taken as the feature amount, and the degree of decrease in the index for each feature amount of the decision tree 516 is calculated to derive the importance of the feature amount. The derivation section 518 is an example of the derivation section of the present disclosure.
[0067] In this way, a random forest model is applied to the second model 51B, which receives as input state data extracted from the internal state of the first model 51A (ESN). In the second model 51B, when a series of time-series measurement data 4 has been input to the first model 51A (ESN), a random forest model is constructed which receives as input state data indicating the internal state of the ESN extracted corresponding to each driving scene, and outputs evaluation data 6 indicating a sensory evaluation value. The second model 51B also includes a derivation unit 518 which derives the importance of the above-mentioned feature amount.
[0068] Therefore, for example, by using state data (neuron information) at each timing indicating the early, middle, and late stages of a driving scene with respect to steering during a lane change, it becomes possible to output information indicating which data at which timing is important to the results of the sensory evaluation.
[0069] Note that a deep forest model may be applied to the first model 51B instead of the random forest model described above.
[0070] As described above, in the first model 51A of the estimation model 51, a model is constructed in which time-series measurement data 4 that changes according to the driving state of the vehicle 2 is input, and state data indicating the internal state value of the first model 51A (ESN) at a predetermined timing corresponding to the driving scene is output.
[0071] In addition, in the second model 51B of the estimation model 51, a model is constructed in which state data indicating the internal state value of the first model 51A (ESN) is input and output data is output. The state data, which is the input, indicates the internal state value of the first model 51A (ESN) at a predetermined timing corresponding to the driving scene that progresses according to the time-series measurement data 4 that changes according to the driving state of the vehicle 2. In addition, the output data of the second model 51B includes evaluation data 6 and related data 7. The evaluation data 6 is data corresponding to the result of the sensory evaluation of the occupant (i.e., the sensory evaluation value). The related data 7 is data corresponding to the driving scene (e.g., timing) that contributes to the result of the sensory evaluation of the occupant (e.g., the judgment basis) and the feature amount (i.e., the importance of the feature amount) that is considered important in that driving scene. In other words, the related data 7 corresponds to information that is related to the result of the sensory evaluation of the occupant and indicates the input / output to the occupant that is considered important for each driving scene.
[0072] As described above, the estimation model 51, i.e., the first model 51A and the second model 51B, are stored in a memory (not shown) of the estimation device 1. In addition, by using the estimation model 51, the estimation device 1 can estimate a result corresponding to the occupant's sensory evaluation regarding the traveling of the vehicle 2, and can also estimate information indicating input / output to the occupant that is related to the result of the occupant's sensory evaluation and is considered important for each traveling scene.
[0073] <Configuration of the estimation device> Next, an example of a specific configuration of the above-mentioned estimation device 1 will be further described. Fig. 9 is a diagram showing an example of the electrical configuration of the estimation device 1. The estimation device 1 shown in Fig. 9 is configured to include a computer as an execution device that executes processes to realize the various functions described above. The estimation device 1 described above can be realized by causing the computer to execute a program representing each of the functions described above.
[0074] The computer functioning as the estimation device 1 includes a computer main body 100. The computer main body 100 includes a CPU 102, a RAM 104 such as a volatile memory, a ROM 106, an auxiliary storage device 108 such as a hard disk drive (HDD), and an input / output interface (I / O) 110. The CPU 102, the RAM 104, the ROM 106, the auxiliary storage device 108, and the input / output I / O 110 are connected via a bus 112 so as to be able to exchange data and commands with each other. In addition, the input / output I / O 110 is connected to a communication unit 114 for communicating with an external device, an operation display unit 116 such as a display and a keyboard, and a detection unit 118 including the above-mentioned measurement sensor. The detection unit 118 has a function of acquiring measurement data 4 from a passenger aboard the vehicle 2 via a wireless connection or a wired connection. The measurement data 4 may be acquired via the communication unit 114.
[0075] The auxiliary storage device 108 stores an estimation program 108P for causing the computer main body 100 to function as an example of an information processing device of the present disclosure. The CPU 102 reads out the estimation program 108P from the auxiliary storage device 108, loads it into the RAM 104, and executes processing. As a result, the computer main body 100 that has executed the estimation program 108P operates as the estimation device 1. The estimation program 108P is an example of an information processing program of the present disclosure.
[0076] The estimation model 51 including the first model 51A and the second model 51B, and data 108D including various data are stored in the auxiliary storage device 108. The estimation program 108P may be provided by a recording medium such as a CD-ROM.
[0077] <Estimation process> Next, the estimation process in the estimation device 1 implemented by a computer will be further described. Fig. 10 is a diagram showing an example of the flow of an estimation process by the estimation program 108P executed by the computer main body 100. The estimation process shown in Fig. 10 is executed by the CPU 102 when the computer main body 100 is powered on. The CPU 102 reads out the estimation program 108P from the auxiliary storage device 108, loads it into the RAM 104, and executes the process. The estimation process shown in FIG. 10 includes a process corresponding to a processing flow capable of implementing the information processing method of the present disclosure.
[0078] First, the CPU 102 reads out the estimation model 51 from the estimation model 108M in the auxiliary storage device 108, and loads it in the RAM 104 to acquire the estimation model 51 (step S200). Specifically, by loading a neural network model (see FIG. 6) as the estimation model 51 in the RAM 104, the estimation model 51 including the first model 51A and the second model 51B is constructed.
[0079] Next, the CPU 102 acquires the measurement data 4, which is time-series information indicating input and output to the occupant, in a time series manner via the detection unit 118 (step S202), and sequentially inputs the acquired time-series measurement data 4 to the first model 51A of the estimation model 51 (step S204). Thus, the time-series measurement data 4 is applied to the ESN, which is the first model 51A, and the neuron value is updated. The above-described measurement sensors 21, 22, and 23 and the detection unit 118 are an example of an acquisition unit in the technology of the present disclosure. The process of step S202 is an example of a process by the acquisition unit of the present disclosure.
[0080] Next, the CPU 102 stores state data indicating the internal state of the first model 51A at a predetermined timing (step S206). The predetermined timing is a predetermined timing corresponding to the driving scene as described above. That is, the measurement data 4 is a physical quantity that changes in time series as the vehicle 2 drives. In addition, the driving of the vehicle can be classified into a plurality of driving scenes. For example, the driving of the vehicle 2 due to the lane change described above can be classified into an early driving scene, a middle driving scene, and a late driving scene. At a timing determined in correspondence with these driving scenes, the first model 51A, that is, state data indicating the internal state which is the neuron value of the ESN, is extracted. In addition, the state data is extracted for each type of the measurement data 4. The process of step S206 is an example of a process by an extraction unit of the present disclosure.
[0081] Next, the CPU 102 inputs state data indicating the internal state of the first model 51A stored in the memory to the second model 51B (step S208). Specifically, when the measurement data 4, which is a series of time-series information on the running of the vehicle 2, i.e., from the initial running scene to the later running scene, is input to the first model 51A (ESN), the state data is input to the second model 51B, i.e., the random forest model. This state data is the internal state of the first model 51A (ESN) for each type of the measurement data 4, and is data extracted at a timing corresponding to each running scene.
[0082] When this series of time-series measurement data 4 has been input into the first model 51A (ESN), a second model 51B (random forest model) is constructed that takes state data indicating the internal state of the first model 51A as input and outputs evaluation data 6 indicating the sensory evaluation value.
[0083] Next, the CPU 102 estimates the evaluation data 6 (step S210). Specifically, the CPU 102 obtains the output of the constructed second model 51B (random forest model) to obtain the evaluation data 6 of the estimation result.
[0084] Next, CPU 102 estimates related data 7 (step S212). Specifically, derivation unit 518 calculates the index for each feature amount of decision tree 516 of second model 51B as described above, thereby deriving the importance of the feature amount. For example, in second model 51B, the degree of reduction in the index for the feature amount is calculated, that is, the degree of reduction in the index for evaluating the conditional branch of decision tree 516 is calculated (for example, the sum).
[0085] The derived related data 7 is the importance of the feature, and corresponds to the importance of the type of input / output to the occupant that contributes to the evaluation data 6 indicating the sensory evaluation value. Therefore, in each driving scene included in the traveling of the vehicle, the higher the importance of the feature corresponding to the type of input / output to the occupant, the higher the degree of contribution to the evaluation data 6. Therefore, it is possible to identify a feature whose importance in a driving scene is the maximum value or exceeds a predetermined threshold, i.e., the type of input / output to the occupant, as information that contributes to the evaluation data 6 in the corresponding driving scene. The process of step S210 and the process of step S212 are examples of processes performed by the derivation unit of the present disclosure.
[0086] Next, CPU 102 outputs the output data (evaluation data 6 and related data 7) of the estimation results obtained in steps S210 and S212 (step S214), and ends this processing routine. The output data may be displayed on operation display unit 116, or may be output to an external device via communication unit 114.
[0087] In this way, the estimation device 1 can estimate information related to the result of the occupant's sensory evaluation (for example, the basis for judgment) regarding vehicle driving. Specifically, the estimation device 1 uses the internal state (neuron information) of the model at the timing indicating the driving scene regarding vehicle driving such as lane changes, making it possible to output information indicating which information (type of input / output to the occupant) is important at what timing for the result of the sensory evaluation.
[0088] Next, the results of verifying the evaluation data 6 and the related data 7 using the estimation device 1 according to the present embodiment will be described.
[0089] FIG. 11 is a diagram showing distribution characteristics of measurement data 4 of a tire used in the verification. The tires used in the verification are three different types of tires. The distribution characteristics of the measurement data 4 show the change characteristics of the steering angle among the results of measuring the input / output data to the occupant by performing 50 lane change driving states as the vehicle driving. The distribution characteristics of the measurement data 4 show the characteristics of the first tire A with a solid line, the characteristics of the second tire B with a dotted line, and the characteristics of the third tire C with a dashed line according to the type of tire. Also, FIG. 11 shows the timing ts in the early driving scene Ts, the timing tm in the middle driving scene Tm, and the timing te in the later driving scene Te. Also, since the measurement data of each tire has a variation, the characteristics of the average value are shown. The variation in the characteristics of the first tire A is abbreviated by up and down arrows. As shown in FIG. 11, by referring to the average value of the measurement data 4, it can be understood that the characteristics are roughly distributed by tire type.
[0090] Fig. 12 is a conceptual diagram showing the results of subjective sensory evaluation (sensory evaluation value) by occupants for three types of tires. For the three types of tires, a sensory evaluation value D was used that allowed discrimination between three levels of evaluation: good (suitable), normal (medium), and poor (unsuitable). The subjective evaluation results of the occupants who performed the evaluation showed that the sensory evaluation value D was roughly the same for each tire in each driving scenario, but the sensory evaluation value differed depending on the type of tire.
[0091] Fig. 13 is a conceptual diagram showing the correspondence between the result of the sensory evaluation by the occupant and the estimation result using the estimation model 51. As shown in Fig. 13, it can be understood that the result of the sensory evaluation by the occupant and the estimation result using the estimation model 51 are roughly in correspondence with each other.
[0092] Fig. 14 is a conceptual diagram showing the relationship between the importance of feature values for each driving scene and the input / output to the occupant. As shown in Fig. 14, it was estimated that the importance of steering angle is large in early driving scenes, the importance of steering angle and steering torque is large in mid-stage driving scenes, and the importance of head lateral acceleration is large in late driving scenes. The above-mentioned verification results were found to be natural when compared with previously acquired information such as comments that describe the feelings of the occupants, and it was confirmed that the estimation results are valid.
[0093] Furthermore, the technical scope of the present disclosure is not limited to the scope described in the above embodiment. Various modifications or improvements can be made to the above embodiment without departing from the gist of the present disclosure, and the modifications or improvements are also included in the technical scope of the present disclosure.
[0094] In addition, in the above embodiment, the estimation process is described as being realized by a software configuration using a flowchart, but this is not limited to this, and for example, each process may be realized by a hardware configuration.
[0095] Furthermore, a part of the estimation device, for example a neural network such as an estimation model, may be configured as a hardware circuit.
[0096] Although the technology of the present disclosure has been described above using the embodiments, the technical scope of the present disclosure is not limited to the scope described in the above embodiments. Various modifications or improvements can be made to the above embodiments without departing from the gist of the disclosure, and the forms with such modifications or improvements are also included in the technical scope of the disclosure.
[0097] In the above embodiment, the process is performed by executing the program stored in the auxiliary storage device, but at least a part of the process of the program may be realized by hardware. The process flow of the program described in the above embodiment is also an example, and unnecessary steps may be deleted, new steps may be added, or the process order may be changed without departing from the spirit of the invention.
[0098] Furthermore, in order to cause a computer to execute the processing in the above-described embodiment, a program describing the above-described processing in code that can be processed by a computer may be stored on a storage medium such as an optical disk and distributed.
[0099] In the above-described embodiment, a CPU has been used as an example of a general-purpose processor. However, in the above-described embodiment, the term "processor" refers to a processor in a broad sense, and includes general-purpose processors (e.g., CPU: Central Processing Unit, etc.) and dedicated processors (e.g., GPU: Graphics Processing Unit, ASIC: Application Specific Integrated Circuit, FPGA: Field Programmable Gate Array, programmable logic device, etc.).
[0100] Furthermore, the operations of the processors in the above-described embodiments may not only be performed by a single processor, but may also be performed by multiple processors working together, or may be performed by multiple processors located at physically separate locations working together.
[0101] All publications, patent applications, and standards mentioned in this specification are herein incorporated by reference to the same extent as if each individual publication, patent application, or standard was specifically and individually indicated to be incorporated by reference.
[0102] The SDGs have been proposed to realize a sustainable society. One embodiment of the present invention is thought to be a technology that can contribute to "No.9" and other goals. [Explanation of symbols]
[0103] 1 Estimation device 2 Vehicles 3. Driving condition 4. Measurement data 5 Estimation part 6. Evaluation Data 7 Related Data 21, 22, 23 Measurement sensor 51 Estimation model 51A 1st model 51B 2nd model 52 Memory 100 Computer main unit 108 Auxiliary storage 108P Estimation Program 114 Communications Department 116 Operation display section 118 Detection unit 510 Input Layer 512 Reservoir Layer 516 Decision Tree 518 Derivation part
Claims
1. an acquisition unit that acquires a series of a plurality of pieces of time-series information that change in a time series manner in response to the running of a vehicle on which the tire is mounted and that indicates inputs and outputs to an occupant of the vehicle; an extraction unit that extracts a plurality of pieces of feature information indicating features for the input and output in each of a plurality of driving scenes obtained by dividing a driving of the vehicle into a plurality of time intervals, using a first model that receives the plurality of pieces of time-series information acquired by the acquisition unit as an input and outputs first evaluation information indicating an evaluation of the tire; a derivation unit that uses a second model in which the plurality of pieces of feature information extracted by the extraction unit are input and second evaluation information indicative of an evaluation of the tire is output, to derive association information indicative of an association between the second evaluation information and the input / output in each of the plurality of traveling scenes; An information processing device comprising:
2. The extraction unit extracts, as the plurality of pieces of feature information, information indicating an internal state of the first model at a plurality of predetermined different timings corresponding to each of the plurality of driving scenes. The information processing device according to claim 1 .
3. The derivation unit derives, as the relevance information, an importance indicating a degree of relevance of the input / output with respect to the second evaluation information, based on information indicating an internal state of the second model in each of the plurality of traveling scenes. The information processing device according to claim 1 .
4. the time-series information includes arm information indicating inputs and outputs related to the arms of the occupant, waist information indicating inputs to the waist of the occupant, and head information indicating inputs to the head of the occupant; The arm information is data indicating at least one of a steering angle and a steering torque, the waist information is data indicating a center of gravity position of an occupant on a seat, and the head information is data indicating a head acceleration and an angular velocity. The information processing device according to claim 1 .
5. the first model is a recurrent neural network model; The second model is a random forest model. The information processing device according to claim 1 .
6. the information indicating the internal state of the first model is a neuron value in the recurrent neural network model; The information indicating the internal state of the second model is information indicating a result of a conditional branch in the random forest model. The information processing device according to claim 5 .
7. The computer acquiring a series of a plurality of time-series information items which change over time in response to the running of a vehicle on which the tire is mounted and which indicate inputs and outputs to an occupant of the vehicle; extracting a plurality of pieces of feature information indicating features of the input and output in each of a plurality of driving scenes obtained by dividing the driving of the vehicle into a plurality of time intervals using a first model that receives the plurality of pieces of acquired time-series information as an input and outputs first evaluation information indicating an evaluation of the tire; deriving association information indicating an association between the second evaluation information and the input / output in each of the plurality of driving scenes using a second model that receives the extracted plurality of pieces of feature information as an input and outputs second evaluation information indicating an evaluation of the tire; An information processing method for processing the above.
8. To your computer acquiring a series of a plurality of time-series information items which change over time in response to the running of a vehicle on which the tire is mounted and which indicate inputs and outputs to an occupant of the vehicle; extracting a plurality of pieces of feature information indicating features of the input and output in each of a plurality of driving scenes obtained by dividing the driving of the vehicle into a plurality of time intervals using a first model that receives the plurality of pieces of acquired time-series information as an input and outputs first evaluation information indicating an evaluation of the tire; deriving association information indicating an association between the second evaluation information and the input / output in each of the plurality of driving scenes using a second model that receives the extracted plurality of pieces of feature information as an input and outputs second evaluation information indicating an evaluation of the tire; An information processing program that processes information.
Citation Information
Patent Citations
Information processing apparatus, calculation method, and calculation program
JP2020166442A
Time segment representative feature vector generation device
WO2010087125A1
