Vehicle control device and vehicle control method

Abstract

This technology provides a way to reduce the number of omissions in limiting control values ​​when executing in-vehicle functions. [Solution] The vehicle control device 10 includes a model holding unit 36 ​​that holds a learning model which has learned control values ​​for controlling a predetermined in-vehicle function; an acquisition unit that acquires vehicle state information and environmental information around the vehicle; a derivation unit 26 that derives a first control value for executing an in-vehicle function based on the vehicle state information and environmental information; a threshold output unit 28 that inputs vehicle state information and environmental information to the learning model and outputs a threshold; a limiting unit 30 that derives a second control value which is limited by the threshold for the first control value; and a control unit 32 that executes an in-vehicle function based on the second control value.

Description

Technical Field

[0001] The present invention relates to a technique for using a learning model to execute a predetermined in-vehicle function.

Background Art

[0002] Patent Document 1 discloses that a vehicle accident prediction system acquires a learning dataset including feature group data including a first feature amount representing an attribute of a driver of a vehicle, a second feature amount representing a state of the vehicle, and a third feature amount obtained by combining a plurality of second feature amounts, and accident data related to accidents of the vehicle, generates a learned model for predicting vehicle accidents from the feature group data by learning using the acquired plurality of learning datasets, inputs the feature group data to be predicted, and predicts vehicle accidents from the input feature group data using the generated learned model.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] When a vehicle control device derives a control value for executing a predetermined in-vehicle function, there is a risk of deriving an inappropriate control value. Therefore, there is a method of restricting the control value by a preset threshold, but in a complicated driving situation, there is a risk of leakage in the restriction only by the preset threshold.

[0005] An object of the present invention is to provide a technique for reducing leakage in restriction when restricting a control value for executing an in-vehicle function.

Means for Solving the Problems

[0006] To solve the above problems, a vehicle control device according to one aspect of the present invention includes: a model holding unit that holds a learning model that has learned control values ​​for controlling a predetermined in-vehicle function; an acquisition unit that acquires vehicle state information and environmental information around the vehicle; a derivation unit that derives a first control value for executing an in-vehicle function based on the vehicle state information and environmental information; a threshold output unit that inputs vehicle state information and environmental information to the learning model and outputs a threshold; a limiting unit that derives a second control value obtained by limiting the first control value by the threshold; and a control unit that executes an in-vehicle function based on the second control value. The learning model may be learned using a deep learning method.

[0007] Another aspect of the present invention is a vehicle control method. This method is a vehicle control method in which each step is performed by a computer, and includes the steps of: holding a learning model which has learned control values ​​for controlling a predetermined in-vehicle function; acquiring vehicle state information and environmental information around the vehicle; deriving a first control value for executing an in-vehicle function based on the vehicle state information and environmental information; inputting the vehicle state information and environmental information into the learning model and outputting a threshold; deriving a second control value which is the first control value limited by the threshold; and executing an in-vehicle function based on the second control value. [Effects of the Invention]

[0008] According to the present invention, a technique can be provided to reduce the likelihood of limiting control values ​​when limiting control values ​​for executing in-vehicle functions. [Brief explanation of the drawing]

[0009] [Figure 1] This diagram shows the functional configuration of the vehicle control system in the embodiment. [Figure 2] This is a diagram to explain the learning model. [Figure 3] This diagram illustrates the third input data output by the threshold output unit. [Figure 4] This is a flowchart of the vehicle control method performed by the vehicle control system of the embodiment. [Modes for carrying out the invention]

[0010] Figure 1 shows the functional configuration of the vehicle control system 1 of the embodiment. The vehicle control system 1 is capable of performing autonomous driving control, such as follow control which follows a preceding vehicle and cruise control which drives along a driving lane at a predetermined speed. The vehicle control system 1 comprises a vehicle control device 10, an object detection sensor 12, a vehicle state detection sensor 14, and a driving device 16.

[0011] The object detection sensor 12 includes an in-vehicle camera, millimeter-wave radar, optical radar, and sound wave sensor, and detects objects located around the vehicle. The object detection sensor 12 may transmit information indicating the positional relationship between the object and the vehicle as object-related information to the vehicle control device 10, or it may transmit simple sensor values ​​as object-related information to the vehicle control device 10.

[0012] The vehicle state detection sensor 14 includes an operation detection unit that detects driver operations and a driving state detection sensor that detects the vehicle's driving state. The operation detection unit includes an accelerator operation detection sensor, a brake operation detection sensor, a steering angle detection sensor, etc. The driving state detection sensor includes a vehicle speed sensor, a steering angle sensor, an acceleration sensor, a yaw rate sensor, a brake pressure sensor, etc., and transmits the results of detecting the vehicle's driving state to the vehicle control device 10.

[0013] The running gear 16 includes a driving means that applies driving force to the wheels and rotates the wheels to move the vehicle forward, a steering means that steers the wheels, and a braking means that applies braking force to the wheels. The driving means may be an engine, a motor, or a combination thereof. The running gear 16 may be driven by the driver's operation or by automatic driving control.

[0014] The vehicle control device 10 includes a communication unit 20, an environmental information acquisition unit 22, a vehicle status information acquisition unit 24, an output unit 26, a threshold output unit 28, a limiting unit 30, a control unit 32, a learning unit 34, a model holding unit 36, and a threshold holding unit 38.

[0015] The vehicle control device 10 performs predetermined on-board functions. These predetermined on-board functions may be functions that perform driver assistance control such as automatic driving control, automatic braking function, and anti-lock braking system. In driver assistance control, the vehicle control device 10 controls the running gear 16 according to the target vehicle speed, target steering angle, and target deceleration calculated by the vehicle control device 10.

[0016] The communication unit 20 acquires the detection results of the object detection sensor 12 and the vehicle state detection sensor 14, respectively. The environmental information acquisition unit 22 acquires environmental information around the vehicle based on the detection results of the object detection sensor 12, and in particular acquires information about objects located around the vehicle. The environmental information acquisition unit 22 may identify objects from the captured image using a neural network method, such as a deep learning method. For example, the environmental information acquisition unit 22 acquires environmental information including the type of object, the position of the object, and the relative speed between the object and the vehicle.

[0017] The environmental information acquisition unit 22 acquires the following environmental information: road type such as expressways and general roads; road shape such as straights, curves, branches, merges and intersections; traffic signs; lane type; pedestrian crossings; presence or absence of lane restrictions; number of surrounding vehicles; type of surrounding vehicles; relative distance to surrounding vehicles; lane of surrounding vehicles; relative speed to surrounding vehicles; direction of travel of surrounding vehicles; lights on of surrounding vehicles; presence or absence of pedestrians and cyclists; relative distance to pedestrians and cyclists; and presence or absence of sidewalks.

[0018] The vehicle status information acquisition unit 24 acquires vehicle status information that includes operation information indicating the amount of driver operation, control information added by the vehicle control device 10, and driving status information indicating the driving state of the vehicle. The operation information includes accelerator operation amount and brake operation amount. The driving status information includes vehicle speed, acceleration, steering angle, direction of travel, and the status of the on-board system. The control information added by the vehicle control device 10 includes control values ​​that are autonomously controlled by the vehicle control device 10 when performing driver assistance control.

[0019] The model holding unit 36 holds a learning model that has learned control values for controlling a predetermined vehicle function. The learning unit 34 learns the learning model held in the model holding unit 36 based on the control result of the vehicle and the detection results of the object detection sensor 12 and the vehicle state detection sensor 14. The learning unit 34 uses a neural network method, for example, a deep learning method, to learn the learning model with teacher data.

[0020] The learning unit 34 uses the control value for controlling the traveling device 16 as an input, and learns the learning model using the traveling state information and the environment information, which are the output results, as teacher data. For example, the learning model inputs vehicle state information and environment information and outputs a score indicating the possibility of the vehicle having an accident, and predicts accidents. The score indicating the possibility of the vehicle having an accident is a numerical value between zero and one. If it is zero, it is estimated that an accident will definitely not occur, and if it is one, an accident will definitely occur. Accidents of the vehicle include collisions and rollovers.

[0021] The learning unit 34 learns the learning model using past accident video data and vehicle operation data at the time of the accident, past near-miss video data and vehicle operation data at the time of the near-miss, and past safe driving video data and vehicle operation data. The learning unit 34 creates teacher data with the output of data where an accident occurs set to 1 and teacher data with the output of data where safe driving is performed set to zero. Thereby, the weighting of the learning model can be pre-learned. The data to be learned may also include control values in driving support control.

[0022] FIG. 2 is a diagram for explaining the learning model 40. The learning model 40 outputs output data 48 by taking the first input data 42, the second input data 44, and the third input data 46 as inputs.

[0023] For example, the first input data 42 is environmental information, the second input data 44 is driving state information, and the third input data 46 is a control value for the running gear 16. The control value for the running gear 16 includes control values ​​corresponding to the driver's operation and control values ​​from driver assistance control. For example, the output data 48 is a score indicating the likelihood of the vehicle causing an accident.

[0024] Returning to Figure 1, the derivation unit 26 derives a first control value for executing a predetermined in-vehicle function based on vehicle state information and environmental information. The first control value may consist only of control values ​​from driver assistance control, or it may be the sum of control values ​​corresponding to the driver's operation and control values ​​from driver assistance control. The first control value includes acceleration in the longitudinal direction of the vehicle, steering angle, and steering angular velocity. In addition, in automatic driving control, the derivation unit 26 may use a deep learning method to input vehicle state information and environmental information and derive the first control value.

[0025] If the probability of the vehicle occurring in an accident is greater than a predetermined value using the learned model held in the model holding unit 36, the derivation unit 26 derives the vehicle's acceleration, steering angle, and steering angular velocity as control values ​​to be used in driver assistance control. These are control values ​​that are executed in addition to the driver's operations.

[0026] The threshold output unit 28 inputs vehicle state information and environmental information into the learning model held by the model holding unit 36 ​​and outputs a threshold. The learning model held by the model holding unit 36 ​​normally outputs the probability that the vehicle will have an accident, but the threshold output unit 28 performs calculations in the reverse direction 50 as shown in Figure 2 and outputs third input data 46. For example, the score is set to 0.9 or higher, which is estimated to indicate a high probability of an accident, and the control value of the running device 16 is output as the third input data 46. The threshold output unit 28 inputs vehicle state information, environmental information, and the score which would normally be the output result into the learning model, performs calculations in the reverse direction 50 and outputs third input data 46, and outputs a threshold corresponding to the third input data 46.

[0027] Figure 3 is a diagram illustrating the third input data 46 output by the threshold output unit 28. In Figure 3, the vertical axis represents the degree of steering, and the horizontal axis represents the acceleration in the longitudinal direction of the vehicle. The degree of steering is calculated from the steering angle and the steering angular velocity, and is calculated by adding the steering angle to a proportional value obtained by multiplying the steering angular velocity by a predetermined coefficient.

[0028] The threshold output unit 28 outputs third input data 46, which is shown as a boundary 52 in Figure 3, by performing calculations on the learning model in the reverse direction 50. The outer region 54 of the elliptical boundary 52 represents control values ​​that are estimated to cause an accident, and the inner region 56 of the boundary 52 represents control values ​​that are estimated not to have a high probability of causing an accident. When the control values ​​of the running gear 16 fall within the boundary 52, stable driving is achieved. The threshold output unit 28 sets a threshold for the control values ​​of the running gear 16 so that they fall within the inner region 56 of the boundary 52. ​​In other words, the boundary 52 is a threshold determined by acceleration, steering angle, and steering angular velocity.

[0029] By calculating thresholds based on machine learning models, advanced driver assistance systems capable of handling complex environments can be realized. For example, it can suppress the output of incorrect control values ​​in driving conditions that are not similar to the training data. This reduces the likelihood of omissions in limiting the first control value.

[0030] Returning to Figure 1, the limiting unit 30 limits the first control value derived by the derivation unit 26 by a threshold value output by the threshold output unit 28 to derive a second control value. The limiting unit 30 derives a second control value in which the first control value, which is the vehicle's acceleration, steering angle, and steering angular velocity, is limited so as not to exceed the threshold value.

[0031] The control unit 32 receives a second control value from the limiting unit 30 and executes in-vehicle functions such as driver assistance control based on the second control value. When executing a driver assistance function, the control unit 32 controls the running gear 16 using the second control value.

[0032] The threshold holding unit 38 holds a preset threshold value. The preset threshold value is set through experiments or other means, and the first control value is limited by the set threshold value obtained empirically and the set threshold value determined through design, which is effective in situations anticipated by the designers.

[0033] The limiting unit 30 derives a second control value by limiting the first control value using a preset threshold value and a threshold value output by the threshold output unit 28. This limits the first control value in parallel. Limiting with the preset threshold value ensures stable driving in expected situations and reduces discomfort to the driver. Limiting with the threshold value from the threshold output unit 28 reduces the possibility of an incorrect control value being output that would cause discomfort to the driver in unexpected situations.

[0034] Figure 4 is a flowchart of the vehicle control method performed by the vehicle control system 1 of the embodiment. The vehicle status information acquisition unit 24 acquires vehicle status information from the vehicle status detection sensor 14 (S10). The environmental information acquisition unit 22 acquires environmental information around the vehicle from the object detection sensor 12 (S12).

[0035] The derivation unit 26 derives a first control value for executing a predetermined in-vehicle function based on the vehicle state information and environmental information (S14). The threshold output unit 28 inputs the vehicle state information, environmental information, and a score greater than or equal to a predetermined value into the learning model and outputs a threshold (S16).

[0036] The limiting unit 30 limits the first control value using a plurality of set thresholds held by the threshold holding unit 38 and thresholds output by the threshold output unit 28, and outputs a second control value that does not exceed those thresholds (S18). The control unit 32 controls the traveling device 16 using the second control value as the target value.

[0037] The present disclosure has been explained above based on the examples described. The present disclosure is not limited to the examples described above, and various modifications such as design changes can be made based on the knowledge of those skilled in the art. [Explanation of Symbols]

[0038] 1 Vehicle control system, 10 Vehicle control device, 12 Object detection sensor, 14 Vehicle state detection sensor, 16 Running gear, 20 Communication unit, 22 Environmental information acquisition unit, 24 Vehicle state information acquisition unit, 26 Derivation unit, 28 Threshold output unit, 30 Limiting unit, 32 Control unit, 34 Learning unit, 36 Model holding unit, 38 Threshold holding unit, 40 Learning model, 42 First input data, 44 Second input data, 46 Third input data, 48 Output data, 50 Reverse direction, 52 Boundary.

Claims

1. A model holding unit that holds a learning model that has learned control values ​​for controlling a predetermined in-vehicle function, An acquisition unit that acquires vehicle status information and environmental information around the vehicle, A derivation unit that derives a first control value for executing the in-vehicle function based on vehicle status information and environmental information, The learning model is provided with a threshold output unit that inputs vehicle state information and environmental information and outputs a threshold value, A limiting unit that derives a second control value obtained by limiting the first control value by the threshold, A vehicle control device comprising a control unit that executes the in-vehicle function based on the second control value.

2. The vehicle control device according to claim 1, characterized in that the limiting unit limits the first control value by a preset threshold value and the threshold value output by the threshold output unit to derive the second control value.

3. The control unit performs a driver assistance function as an in-vehicle function to assist the driver in driving, A derivation unit that derives the vehicle's acceleration, steering angle, and steering angular velocity as first control values ​​for executing the aforementioned driving assistance function based on vehicle status information and environmental information, The vehicle control device according to claim 1 or 2, characterized in that the limiting unit limits the acceleration, steering angle, and steering angular velocity of the vehicle by the threshold value.

4. The learning model takes vehicle condition information and environmental information as input and outputs a score indicating the likelihood of the vehicle causing an accident. The vehicle control device according to claim 3, characterized in that the threshold output unit inputs vehicle state information, environmental information, and a score to the learning model and outputs the threshold.

5. A vehicle control method in which each step is performed by a computer, A step of maintaining a learning model that has learned control values ​​for controlling a predetermined in-vehicle function, Steps include acquiring vehicle status information and environmental information around the vehicle, A step of deriving a first control value for executing the in-vehicle function based on vehicle status information and environmental information, The steps include inputting vehicle state information and environmental information into the learning model and outputting a threshold value, The steps include: deriving a second control value obtained by limiting the first control value by the threshold value, A vehicle control method characterized by including the step of executing the in-vehicle function based on the second control value.

Images