Program, device, training model, and method

Abstract

The present invention accurately controls a direction indicator of a motorcycle. The program executed in a computer, involves: acquiring, from at least one sensor, traveling data of a motorcycle continuously measured across a prescribed period; and determining a corresponding driving operation, which corresponds to the traveling data, on the basis of the traveling data and a plurality of pieces of reference traveling data that are continued across the prescribed period and are respectively associated with a plurality of driving operations of the motorcycle.

Description

Program, device, learning model, and method

[0001] The present invention relates to a program, an apparatus, a learning model, and a method.

[0002] A motorcycle driver uses a turn signal (blinker) when turning right or left (referred to as a right turn, etc.) or when changing lanes. Turn signals are used, for example, to indicate the driver's driving intentions to surrounding drivers and pedestrians and ensure safe driving. The turn signal of a motorcycle must first be turned on (e.g., flashing) by the driver, and then turned off (e.g., extinguished) after completing a right turn, etc. At this time, the driver may forget to operate the turn signal, resulting in the turn signal not being turned off or forgotten to be turned off. Patent Document 1 discloses a turn signal canceling device to prevent forgetting to turn off the turn signal.

[0003] JP 62-53256

[0004] The turn signal cancellation device described in Patent Document 1 controls the turning off of turn signals based on the difference in rotation speed between the front and rear wheels that occurs when turning right or left. More specifically, the difference in rotation speed at a given point in time is acquired each time, and the turning off of turn signals is controlled based on the change in the difference. When using a value at a given point in time (instantaneous value), it is necessary to determine whether the value exceeds a threshold value. However, in the case of motorcycles, the body tilts depending on the speed when turning right or left, making it necessary to change the threshold value depending on the speed, which makes the determination difficult. Furthermore, because motorcycles have a large degree of freedom in the position that the body can assume relative to the width of the road, there is a problem in that inadvertent movements are sometimes detected as turning right or left.

[0005] The present invention has been made in view of the above circumstances, and has as its object to provide a program or the like for controlling the turn indicators of a motorcycle with high precision.

[0006] A program according to one aspect of the present invention causes a computer to acquire, from at least one sensor, driving data of a motorcycle measured continuously over a predetermined period of time, and determine a corresponding driving action that corresponds to the driving data based on the driving data and a plurality of reference driving data that are continuous over the predetermined period of time and to which each of a plurality of driving actions of the motorcycle is associated.

[0007] Another aspect of the present invention provides an apparatus that includes an acquisition unit that acquires motorcycle driving data measured continuously over a predetermined period from at least one sensor, and a determination unit that determines a corresponding driving action that corresponds to the driving data based on the driving data and a plurality of reference driving data that are continuous over a predetermined period and to which each of a plurality of motorcycle driving actions is associated.

[0008] A learning model according to another aspect of the present invention is trained using teacher data that takes as input a plurality of reference driving data of a motorcycle measured continuously over a predetermined period of time and outputs driving actions associated with each of the plurality of reference driving data, and causes a computer to function such that, in response to input of motorcycle driving data measured continuously over a predetermined period of time, it outputs corresponding driving actions that correspond to the driving data.

[0009] A method according to another aspect of the present invention includes a computer acquiring motorcycle driving data measured continuously over a predetermined period from at least one sensor, and determining corresponding driving actions corresponding to the driving data based on the driving data and a plurality of reference driving data measured continuously over the predetermined period and associated with each of a plurality of driving actions of the motorcycle.

[0010] According to the present invention, it is possible to provide a program or the like that controls the turn indicators of a motorcycle with high precision.

[0011] 1 is a block diagram of a motorcycle according to a first embodiment. FIG. 2 is a diagram illustrating an overview of control in the motorcycle according to the first embodiment. FIG. 3 is a diagram illustrating an overview of control in the motorcycle according to the first embodiment. FIG. 4 is a flowchart illustrating determination of a driving action in the first embodiment. FIG. 5 is a diagram illustrating a learning model according to the first embodiment. FIG. 6 is a diagram illustrating determination of a driving action in the first embodiment. FIG. 7 is a flowchart illustrating processing for determining the end of a driving action in the first embodiment. FIG. 8 is a diagram illustrating determination of a driving action and determination of the end of a driving action in the first embodiment. FIG. 9 is a block diagram of a motorcycle according to a second embodiment. FIG. 10 is a diagram illustrating an overview of control in the motorcycle according to the second embodiment. FIG. 11 is a diagram illustrating an example of classification probability in the motorcycle according to the second embodiment. FIG. 12 is a flowchart illustrating determination of a driving action and determination of the end of a driving action in the second embodiment. FIG. 13 is a block diagram of a motorcycle control system according to a third embodiment. FIG. 14 is another block diagram of the motorcycle control system according to the third embodiment. FIG. 15 is a graph showing an example of sensor values ​​in the motorcycle control systems according to the first to third embodiments. FIG. 16 is a graph showing an example of classification probability in the motorcycle control systems according to the first to third embodiments. FIG. 17 is a graph showing another example of sensor values ​​in the motorcycle control systems according to the first to third embodiments. 10 is a graph showing an example of a case where a sensor value is binarized in the motorcycle control system according to the first to third embodiments.

[0012] An embodiment of the present invention will be described below, firstly, a reference example to be compared with the embodiment of the present invention will be described.

[0013] Regarding the determination of right or left turns, in general, in the case of automobiles (four-wheeled vehicles), right or left turns are made by rotating the steering wheel, so that the angle of the steering wheel is detected and a right or left turn can be determined based on a threshold value set for the angle of the steering wheel. On the other hand, in the case of motorcycles (two-wheeled vehicles), particularly when traveling at high speeds, the body is tilted when turning to prevent the body from tipping over due to centrifugal force. At this time, the change in the steering wheel angle becomes smaller depending on the inclination of the body. For this reason, it is difficult to determine right or left turns based on the steering wheel angle alone. Therefore, it becomes necessary to determine right or left turns based on multiple parameters, including speed, inclination of the body, and the like, in addition to the steering wheel angle. However, as the number of parameters used for determination increases, there is a problem in that it becomes more difficult to set the parameters and the threshold value for determination.

[0014] In such cases, the use of ensemble trees in the field of machine learning can be considered as a method for determining whether a turn has been made by taking into account the correlation between multiple parameters. While the use of ensemble trees makes it possible to set the probability of a turn from the instantaneous values ​​of multiple parameters, it still uses instantaneous values, and noise in the instantaneous values ​​can affect the determination results.

[0015] Furthermore, in the case of two-wheeled vehicles, the vehicle width is small compared to the width of the road, allowing for a degree of freedom in the vehicle's position on the road. Therefore, there is a possibility that a right or left turn has been determined based on irregular movements, such as moving left and right on the road. To prevent this phenomenon, a method for determining right or left turns based on continuous data from multiple sensors, rather than the above-mentioned method using instantaneous values, can be considered. For example, using self-location estimation technology, the vehicle's movement trajectory can be estimated from the integrated sensor values. This makes it possible to estimate the cumulative movement distance and angle, which cannot be determined from instantaneous values, and may therefore be less susceptible to noise in the instantaneous values. However, in this case, steady-state errors in the sensor values ​​are also integrated, which can result in problems with the accuracy of the determination.

[0016] In view of the above, the present invention proposes the following embodiment to solve the problem.

[0017] A first embodiment will now be described. Fig. 1 shows a block diagram of a motorcycle 101 according to the first embodiment.

[0018] The motorcycle 101 includes a body section 201 , an operating section 202 , a direction indicator 203 , a driving section 204 , a sensor section 205 , and a control section 206 .

[0019] The body section 201 is the body of the motorcycle 101, and the motorcycle 101 is configured by attaching an operating section 202, a direction indicator 203, a driving section 204, a sensor section 205, and a control section 206 to the body section 201.

[0020] The operation unit 202 is a device that allows the rider to operate the motorcycle 101. The operation unit 202 includes, for example, a handle that the rider holds while driving, an accelerator for accelerating, a brake for decelerating, an operation unit for switching on and off the direction indicator 203, and the like.

[0021] The turn indicator 203 is a device for indicating to drivers of other vehicles and pedestrians when the motorcycle 101 is making a right turn, a left turn, a lane change, or other driving action (driving action). For example, the turn indicator 203 is a flashing lamp attached to the motorcycle 101.

[0022] The drive unit 204 is an engine or a motor, wheels, etc. The drive unit 204 drives the motorcycle 101 based on the operation of the driver via the operation unit 202 or the control from the control unit 206.

[0023] The sensor unit 205 includes various devices that detect driving data of the motorcycle 101. The driving data includes, for example, the speed, acceleration, tilt, wheel rotation speed, accelerator opening, brake input, gear position, etc. of the motorcycle 101. The sensor unit 205 includes a speed sensor, an acceleration sensor, a gravity sensor (tilt sensor), a tachometer (rotation speed sensor), etc., and detects various driving data. While sensor data is typically one-dimensional data, two-dimensional data may also be acquired using a camera or radar. The sensor unit 205 does not need to acquire all of the above driving data; it is sufficient to acquire at least one of the driving data. Furthermore, the types of driving data are not limited to the above examples.

[0024] The control unit 206 performs signal processing and information processing in the motorcycle 101. The control unit 206 has a communication unit 2061, a storage unit 2062, a traveling data acquisition unit 2063, a conversion unit 2064, a determination unit 2065, and an operation control unit 2066. The control unit 206 is a computer that performs predetermined processing by having a processor execute a program stored in the storage unit 2062.

[0025] The communication unit 2061 controls information communication in the motorcycle 101. For example, the communication unit 2061 transmits and receives data to and from the sensor unit 205 by performing wired or wireless communication with the sensor unit 205. Alternatively, the communication unit 2061 may control communication between the motorcycle 101 and an external update terminal. The update terminal transmits, for example, information to the control unit 206 for updating a program that operates the control unit 206, thereby updating the program in the control unit 206.

[0026] The memory unit 2062 is, for example, a semiconductor memory. The memory unit 2062 stores a learning model 20621. The learning model 20621 receives input data based on the driving data and outputs a driving action corresponding to the input data (corresponding driving action). The learning model 20621 also outputs a classification probability that the input data will be classified as the corresponding driving action. Details of the learning model 20621 will be described later.

[0027] The traveling data acquisition unit 2063 acquires traveling data from the sensor unit 205. The traveling data acquisition unit 2063 acquires traveling data that has been continuously measured over a predetermined period of time. Here, the predetermined period is, for example, a period determined based on the distance traveled by the motorcycle 101. Specifically, it is the period required for the motorcycle 101 to travel a certain distance. When the motorcycle 101 has traveled a certain distance, the traveling data acquisition unit 2063 performs sampling at intervals according to the rotation of the wheels, for example, at equal distance intervals, and acquires traveling data covering the range from the start to the end of the movement as one set of traveling data.

[0028] The converter 2064 converts the driving data into input data for the determination unit 2065. For example, the converter 2064 may plot the driving data information against, for example, travel distance on the horizontal axis to generate two-dimensional data (input two-dimensional data). In this case, the converter 2064 generates input data for the determination unit 2065. For example, the converter 2064 generates a graph with travel distance on the horizontal axis, which is generated by converting the driving data acquired by the driving data acquisition unit 2063 so that travel distance is used as a reference based on the number of wheel rotations. The input data is generated as data in which multiple variables, such as speed, inclination, and steering angle, are plotted together on a single graph. Furthermore, if the driving data is one-dimensional, the converter 2064 may treat the graph as a two-dimensional image, and may convert the driving data into two-dimensional data if, for example, the data has frequency characteristics. Methods such as Fourier transform and wavelet transform are known as methods for converting one-dimensional data into two-dimensional data, but other methods that can ultimately classify the movement may also be used as the conversion method.

[0029] The determination unit 2065 determines a driving action corresponding to the driving data based on the driving data. Specifically, the determination unit 2065 inputs the input data converted by the conversion unit 2064 to the learning model 20621, and obtains the corresponding driving action and classification probability from the learning model 20621.

[0030] The operation control unit 2066 controls the operation of the motorcycle 101 based on the classification probability of the corresponding driving action. Specifically, the operation control unit 2066 determines the completion of the driving action of turning right or left based on a change in the classification probability. When the operation control unit 2066 determines that the motorcycle 101 has completed turning right or left, it switches the turn indicator 203 from the on state to the off state.

[0031] An outline of the processing in the motorcycle 101 will be described with reference to FIG.

[0032] 2 shows a schematic diagram of the operation of the motorcycle 101 when looking down at the intersection. When the signal S indicates that a right turn and a left turn are permitted, the motorcycle 101 turns left along a route P1 and continues straight along a route P2. Alternatively, the motorcycle 101 turns right along a route P3 and continues straight along a route P4.

[0033] When turning right or left, the driver of the motorcycle 101 operates the operation unit 202 to turn on the turn indicator 203 .

[0034] Here, an example will be described in which the motorcycle 101 makes a left turn. In step (a), the driving data acquisition unit 2063 acquires driving data of the motorcycle 101. In step (b), the conversion unit 2064 generates an input image based on the driving data.

[0035] The processing of steps (a) and (b) is repeated while the motorcycle 101 is traveling. A plurality of input images are generated at different times, such as input images I11, I12, and I13, for each travel distance.

[0036] In step (c), the determination unit 2065 inputs an input image (e.g., input image I11) to the learning model. In step (d), the learning model 20612 calculates the classification probability of the driving action and determines the corresponding driving action of the motorcycle based on the classification probability. The processing up to step (d) is performed for each of the multiple input images.

[0037] In step (e), the end timing of the left turn is estimated based on the change in classification probability. In step (f), the operation of the motorcycle 101 is controlled based on the end timing, and the turn indicator 203 is turned off. Up to this point, the case of a left turn has been described, but the same processing is also performed in the case of a right turn.

[0038] The motorcycle 101 determines the corresponding driving action based on the input image generated for each travel distance. This reduces the impact of waiting time, which occurs when the motorcycle 101 waits to turn right or left depending on the state of the traffic light S, or when the motorcycle 101 needs to stop temporarily at an intersection due to the presence of an oncoming vehicle B3. This is because the travel distance does not change during a temporary stop. This makes it possible to determine whether to turn right or left based on the travel data for the entire route P1, P3 at the time of the right or left turn, without being affected by the waiting time, even when the motorcycle 101 needs to stop temporarily. This therefore increases the accuracy of determining the corresponding driving state.

[0039] Furthermore, turning right, etc., on a motorcycle is performed by tilting the body when traveling at high speeds, and by operating the handlebars to change the direction of the wheels when traveling at low speeds, so it is necessary to select variables according to the traveling state and determine whether to turn right, etc., based on the selected variables. Even in such cases, the motorcycle 101 can improve the accuracy of determining the corresponding driving behavior by determining whether to turn right, etc., using an input image generated based on multiple variables.

[0040] The learning model 20621 may be trained to distinguish between right and left lane changes, in addition to right and left turns. For example, as shown in FIG. 3 , a motorcycle 101 in a lane near the center of the road may blink the turn signal 203 indicating the left direction and change lanes to the left. In this case, by distinguishing between left turns and left lane changes, a left lane change can be prevented from being determined as a left turn, thereby improving the accuracy of the determination. The same applies to distinguishing between right turns and right lane changes.

[0041] The learning model 20621 may be trained to determine a single driving action, "the motorcycle 101 turns," without distinguishing between right turns and left turns. The action control unit 2066 may turn off the turn indicator 203 when it is determined that the corresponding driving action is "the motorcycle 101 turns" and that the motorcycle 101 has finished turning.

[0042] 4 to 6, the processing and learning model in the motorcycle 101 will be described in detail. In the motorcycle 101, when an operation is performed to turn on the direction indicator 203, a determination is made using the learning model 20621.

[0043] FIG. 4 shows a flowchart of the processing in the control unit 206.

[0044] In step S401, the driving data acquisition unit 2063 acquires driving data of the motorcycle 101 from the sensor unit 205. In step S402, the conversion unit 2064 generates an input image based on the driving data.

[0045] In step S403 , the determination unit 2065 inputs the input image to the learning model 20621 .

[0046] The learning model 20621 performs the processing described with reference to FIG. 5 . The learning model 20621 is a learning model trained using teacher data that receives a plurality of reference driving data as input and outputs driving actions associated with each of the plurality of reference driving data. The learning model 20621 is trained using teacher data TD in which driving actions are associated with teacher image groups TI1, TI2, TI3, TI4, and TI5, which are groups of teacher images generated based on the reference driving data. The learning model 20621 receives a plurality of teacher image groups TI1, TI2, TI3, TI4, and TI5 as input and trains to output the correct driving actions corresponding to each of the teacher image groups.

[0047] The reference driving data is, for example, data measured including a predetermined period from when the motorcycle turns on the turn indicator 203 to make a right turn until the turn indicator 203 is turned off after the right turn is completed. The reference driving data is data based on, for example, manual operation of the turn indicator 203, rather than control of the turn indicator 203 by the operation control unit 2066. The same applies to left turns and other driving operations. Note that when a distinction is not made between right turns and left turns, the reference driving data is data measured including a predetermined period from when the motorcycle turns on the turn indicator 203 to make a right turn or left turn until the turn indicator 203 is turned off after the right turn or left turn is completed.

[0048] The teacher image group TI1 is associated with the driving action of turning right. The teacher image group TI2 is associated with the driving action of turning left. The teacher image group TI3 is associated with the driving action of changing lanes to the right. The teacher image group TI4 is associated with the driving action of changing lanes to the left. The teacher image group TI5 is associated with other driving actions (for example, going straight). Note that when there is no distinction between turning right and turning left, the teacher image groups TI1 and TI2 are associated with the driving action of "the motorcycle 101 turning."

[0049] The learning model 20621 is trained by a general machine learning method such as a method using a neural network. More specifically, the learning model 20621 is a learning model that uses a two-dimensional convolutional neural network. The learning model 20621 may also be a learning model that uses a one-dimensional convolutional neural network. Using one-dimensional convolution enables calculations to be performed at high speed. The learning model 20621 may also be configured as a single learning model, or as a learning model that combines multiple learning models.

[0050] When an input image is input, the learning model 20621 extracts the feature amount of the input image by convolving the input image with a filter (kernel) in step S501.

[0051] In step S502, the learning model 20621 classifies the input image based on the feature amount to generate a classification result. The classification result is shown as a classification probability, which is the probability that the input image is classified into each driving action, as shown in FIG. 6 , for example.

[0052] In step S503, the learning model 20621 determines and outputs a corresponding driving action based on the classification result. The corresponding driving action is, for example, the driving action with the highest classification probability. In the example of FIG. 6, the corresponding driving action is a right turn. In step S504, the determination unit 2065 obtains the corresponding driving action and the classification probability of the corresponding driving action.

[0053] The above processing allows for the determination of a corrective driving behavior based on a single piece of driving data. It is also possible to determine the end of a corrective driving behavior based on a plurality of individually determined corrective driving behaviors. The determination of the end of a corrective driving behavior will be described with reference to FIGS. 7 and 8.

[0054] 7 is a flowchart of the processing in the control unit 206. In step S701, the determination unit 2065 acquires a corresponding driving behavior and a first classification probability based on the first driving data at a first timing.

[0055] Here, an example of the first timing and the second and third timings used in the following description will be described with reference to FIG.

[0056] 8A is a schematic diagram of a motorcycle 101 making a right turn along a route P3. The motorcycle 101 makes a right turn by moving the length (L3) of the route P3, with the travel distance at the start of the right turn being used as a reference (MD=0). The time at which the right turn starts is defined as T=T0.

[0057] The conversion unit 2064 generates an input image based on driving data detected over a certain travel distance. Here, it is assumed that the conversion unit 2064 generates the input image based on driving data detected when the motorcycle 101 travels a distance a. More specifically, as shown in FIG. 8( b), the input image is generated by extracting driving data using a window of distance a. Driving data is added as the vehicle travels, and old data is removed or deleted from the window of distance a.

[0058] In the example of FIG. 8( a), an input image is generated at times T=T1, T2, and T3, when the travel distance MD reaches L1, L2, and L3, respectively. Specifically, at T=T1 (first timing), an input image I21, as shown in FIG. 8( b), is generated based on travel data (first travel data) corresponding to travel distances MD=0 to L1. Once the input image I21 is generated, the determination unit 2065 inputs the input image I21 into the learning model 20621 and obtains the corresponding driving action and classification probability (first classification probability) shown in FIG. 8( c) from the learning model 20621. Based on the input image I21, the corresponding driving action is a "right turn" and the classification probability is 0.40. In this manner, the processing of step S701 is performed.

[0059] Thereafter, the motorcycle 101 moves, and at time T=T2 (second timing) when the motorcycle 101 has traveled a distance L2, an input image I22 is generated based on travel data (second travel data) corresponding to the travel distance up to the distance L2. Based on the input image I22, a corresponding driving action and a classification probability (second classification probability) are also acquired from the learning model 20621. Based on the input image I22, the corresponding driving action is "turn right," and the classification probability is "0.60." In this way, in step S702, the determination unit 2065 acquires a corresponding driving action and a second classification probability based on the second travel data at the second timing.

[0060] At time T=T3 (third timing) when the motorcycle 101 has traveled a distance L3, an input image I23 is generated based on travel data (third travel data) corresponding to the travel distance up to L3. Based on the input image I23, a corresponding driving action and classification probability (third classification probability) are also acquired from the learning model 20621. Based on the input image I23, the corresponding driving action is "turn right," and the classification probability is "0.40." In this way, in step S703, the determination unit 2065 acquires a corresponding driving action and a third classification probability based on the third travel data at the third timing.

[0061] In step S704, the operation control unit 2066 determines whether the classification probability at the intermediate timing among the three timings, i.e., the second classification probability, is greater than the first classification probability and the third classification probability. If the determination in step S704 is negative, the processes from step S701 to S704 are repeated.

[0062] If a positive determination is made in step S704, in step S705, the operation control unit 2066 determines that the corresponding driving operation has ended.

[0063] The reason why the above determination is possible in step S705 will be explained. When a right turn such as that shown in FIG. 7A is performed, the driving data near the start of the right turn (e.g., from MD=0 to L1) is somewhat similar to the reference driving data used to learn the right turn. Since the degree of match (correlation) between the driving data and the reference driving data is somewhat small, the corresponding driving action is a right turn, but its classification probability is relatively small in the classification result.

[0064] Next, the driving data during the right turn (e.g., from MD=L1 to L2) is closer to and matches better with the reference driving data used to learn the right turn. Therefore, in this case, the corresponding driving action is a right turn, and its classification probability is relatively large in the classification result.

[0065] Finally, the driving data near the end of the right turn (e.g., from MD=L2 to L3) is somewhat similar to the reference driving data used to learn about right turns. Because the degree of match between the driving data and the reference driving data is somewhat smaller, the corresponding driving action is a right turn, but its classification probability is relatively small in the classification result.

[0066] In other words, when the classification probability of a right turn for different travel distances changes to reach a peak value, it can be determined that the motorcycle 101 has already performed an action that is closest to a right turn. The fact that the motorcycle 101 has performed an action that is closest to a right turn means that the right turn has been completed. Therefore, the action control unit 2066 can determine that the corresponding driving action (here, a right turn) has been completed.

[0067] In step S706, the operation control unit 2066 turns off the direction indicator 203. In the example of Fig. 8, the direction indicator 203 is turned off at time T = T3.

[0068] In this way, by turning off the turn signal 203 based on a change in the classification probability in a determination using motorcycle driving data measured continuously over a predetermined period, it is possible to turn off the turn signal 203 at a natural timing after the motorcycle 101 turns right or left. By handling data extracted over a predetermined period instead of instantaneous values, noise within the period is averaged, making it possible to reduce the possibility of the turn signal being turned off in the middle of a right or left turn.

[0069] A second embodiment will be described. From the second embodiment onward, descriptions of matters common to the first embodiment will be omitted, and only differences will be described. In particular, similar effects resulting from similar configurations will not be mentioned in each embodiment.

[0070] FIG. 9 shows a block diagram of a motorcycle 101A according to the second embodiment. The motorcycle 101A according to the second embodiment differs from the motorcycle 101 according to the first embodiment in that the learning model 20621A includes a learning model 20622 for right-turn determination (first learning model) and a learning model 20623 for left-turn determination (second learning model). The learning model 20622 for right-turn determination is, for example, a model specialized for determining right turns, right lane changes, and other driving operations of the motorcycle 101A. Similarly, the learning model 20623 for left-turn determination is trained to determine left turns, left lane changes, and other driving operations of the motorcycle 101A.

[0071] First, an example in which the motorcycle 101A turns left will be described. The processes of step (a1) and step (b1) are the same as those of steps (a) and (b) in the first embodiment. In step (c1), the determination unit 2065 inputs an input image (e.g., input image I11) to the left-turn determination learning model 20623. In step (d1), the left-turn determination learning model 20623 calculates a classification probability of the driving behavior and determines the driving behavior (left turn) of the motorcycle based on the classification probability. An example of the classification result for a left turn is shown in FIG. 11. In step (e1), the end timing of the left turn is estimated based on the change in the classification probability. In step (f1), the operation of the motorcycle 101A is controlled based on the end timing, and the turn indicator 203 is turned off. Similarly, when the motorcycle 101A turns right, the processes from step (a2) to step (f2) are repeated, and the direction indicator 203 is finally turned off based on the input images I31, I32, and I33.

[0072] The processing of the control unit 206A in the second embodiment will be described with reference to Fig. 12. In the motorcycle 101A, when an operation is performed to turn on the turn indicator 203, a determination is made using the learning model 20621A. In the control unit 206A, when the turn indicator 203 indicates a right turn, the processing from steps S12011 to S12014 is performed. At this time, in step S12011, the control unit 206A obtains an input indicating a right turn from the operation unit 202 or the turn indicator 203.

[0073] In step S12012, the determination unit 2065 and the operation control unit 2066 determine the end of the right turn using the right turn determination learning model 20622. The determination unit 2065 and the operation control unit 2066 determine the corresponding driving operation (here, a right turn) and the end of the right turn, similar to the method described in the first embodiment. In step S12013, the determination unit 2065 and the operation control unit 2066 determine whether the right turn has been completed. If a negative determination is made in step S12013, the process returns to step S12012. If a positive determination is made in step S12013, in step S12014, the operation control unit 2066 turns off the direction indicator 203 indicating a right turn.

[0074] Similarly, in the control unit 206A, when the direction indicator 203 indicates a left turn, the processing from steps S12021 to S12024 is performed. At this time, in step S12021, the control unit 206A acquires an input indicating a left turn from the operation unit 202 or the direction indicator 203. In step S12022, the determination unit 2065 and the operation control unit 2066 determine the completion of the left turn using the left turn determination learning model 20623. The determination unit 2065 and the operation control unit 2066 determine the corresponding driving action (left turn) and the completion of the left turn, similar to the method described in the first embodiment. In step S12023, the determination unit 2065 and the operation control unit 2066 determine whether the left turn has been completed. If a negative determination is made in step S12024, the processing returns to step S12023. If an affirmative determination is made in step S12023, in step S12024, the operation control unit 2066 turns off the direction indicator 203 indicating a left turn.

[0075] The travel distance when turning left and the travel distance when turning right are different, for example, as shown in routes P1 and P3. When driving on the left side of the road, the travel distance when turning left is shorter than the travel distance when turning right, and when driving on the right side of the road, the travel distance when turning right is often shorter than the travel distance when turning left. When the travel distances required to complete a right turn and a left turn differ, the accuracy of the determination can be improved by determining right turns and left turns using separate learning models, as in the case of motorcycle 101A. In addition, unnecessary movements in the opposite direction (such as a right turn when turning left) can be excluded from the determination.

[0076] The third embodiment will be described with reference to Fig. 13. Fig. 13 shows a block diagram of a motorcycle control system 1300 according to the third embodiment. The motorcycle control system 1300 includes a motorcycle 101B and a determination device 1301.

[0077] In the motorcycle control system 1300, the conversion unit 2064, determination unit 2065, and learning model 20621 of the motorcycle 101 are realized in a determination device 1301 external to the motorcycle 101B. The determination device 1301 has a communication unit 13011 for communicating with the outside, a memory unit 13012 for storing various information, a learning model 13013, a sensor unit 13014, a conversion unit 13015, and a determination unit 13016. The functions of the learning model 13013, the conversion unit 13015, and the determination unit 13016 are similar to the functions of the learning model 20621, the conversion unit 2064, and the determination unit 2065 described in the first embodiment. The determination device 1301 is, for example, an information processing device such as a smartphone capable of communicating with the motorcycle 101B. The memory unit 13012 may store dedicated applications for realizing the functions of the learning model 13013, the sensor unit 13014, the conversion unit 13015, and the determination unit 13016.

[0078] In this case, the operation control unit 2066 transmits the driving data acquired by the driving data acquisition unit 2063 to the determination device 1301 and acquires information indicating an instruction to turn off the turn indicators from the determination device 1301. At this time, the determination unit 13016 may also use data acquired by the sensor unit 13014. Thereafter, the operation control unit 2066 controls the turn indicators 203 in the same manner as in the first embodiment. By adopting a format in which the determination device 1301 is used to determine the corresponding driving behavior and information indicating an instruction to turn off the turn indicators is transmitted to the motorcycle 101, it is not necessary to store the learning model 20621 in the motorcycle 101. This makes it possible to easily provide the functions of determining the corresponding driving behavior and automatically turning off the turn indicators 203 as described in the first embodiment. In addition, it is also possible to compensate for missing sensors with sensors from the determination device.

[0079] Furthermore, as in the motorcycle control system 1400 shown in FIG. 14 , the driving data acquisition unit 2063 may acquire sensor information from an external sensor 1401 outside the motorcycle 101B in addition to information from the sensor unit 205. For example, if the sensor unit 205 does not have the function of detecting the inclination of the motorcycle 101B, the motorcycle 101B may acquire information indicating the inclination of the motorcycle 101B from an external sensor 1401 connected to the motorcycle 101B by wire. Alternatively, the external sensor 1401 may be a camera or a radar device. By acquiring information necessary for determination from outside the motorcycle 101B, it is possible to provide the functions of determining a corresponding driving behavior or automatically turning off the turn signal 203 even when the configuration of the sensor unit 205 is insufficient to provide these functions. Furthermore, the external sensor 1401 may transmit sensor information or driving data directly to the determination device 1301 via wireless communication.

[0080] [Correction based on Rule 91, 13.09.2023] A supplementary note is provided regarding the motorcycles 101, 101A, and 101B according to the first to third embodiments. While the following description uses the motorcycle 101 according to the first embodiment as an example, the following description is also applicable to the motorcycle 101A according to the second embodiment and the motorcycle 101B according to the third embodiment. For the motorcycle 101, the reference driving data is described as data measured over a predetermined period, for example, from when the motorcycle turns on the turn signal 203 to make a right turn until the turn signal 203 is turned off after the right turn is completed. For the motorcycle 101, straight-line travel occurs for a certain period after the turn signal 203 is turned on, and data related to straight-line travel is not necessary. The specified period may be, for example, a fixed period including the period during which the motorcycle 101 made a right or left turn, going back from the time when the turn indicator 203 was turned off to a certain time after the time when the turn indicator 203 was turned on.

[0081] In the first embodiment, each of the teacher image groups TI1, TI2, TI3, TI4, and TI5 has been described as being associated with a respective driving action. Here, when a predetermined period is set based on the operation of the turn signal 203 and data is acquired, since straight driving does not normally involve the operation of the turn signal 203, it is not possible to acquire actual measured data. Therefore, since the actual measured values ​​of the various sensors are approximately constant when driving straight, these values ​​may be simulated by an information processing device, and the simulated data may be used as a substitute for the teacher image group.

[0082] Furthermore, the learning model 20621 may be a learning model that uses a Vision Transformer.

[0083] In the first embodiment, a case was described in which a corresponding driving action (here, a right turn) is determined to be completed when the classification probability of a right turn changes to a peak value. Here, the explanation will be supplemented by an example of a left turn. FIG. 15 shows the sensor values ​​of each axis (x-axis: direction of travel, y-axis: right hand side of the direction of travel, z-axis: perpendicular to the x-axis and y-axis toward the ground) and speed of the gyro sensor when two left turns are performed. FIG. 15 shows the x-axis gyro sensor value G1, the y-axis gyro sensor value G2, the z-axis gyro sensor value G3, and the speed V1. FIG. 15 shows a left turn action that tilts the vehicle body as a fluctuation in the gyro sensor value G3. FIG. 16 shows the classification probability calculated by the determination unit 2065 when the movement shown in FIG. 15 is performed.

[0084] Here, the determination unit 2065 may calculate the classification probability using the image data shown in Figure 15 as a single image data, or may calculate the classification probability using multiple images plotted based on the values ​​of each axis of the gyro sensor and an image plotted based on the speed as inputs.

[0085] FIG. 16 shows the classification probability PS of driving actions other than right and left turns, including going straight, the classification probability PL of left turns, and the classification probability PR of right turns. The classification probability PL of left turns increases as the gyro sensor value G3 in FIG. 15 increases. After the gyro sensor value G3 decreases, the classification probability PL of left turns reaches a peak value at the position where the correlation between the input image and the pre-learned image is maximized, that is, the position where the turn indicator 203 is turned off. In FIG. 16, the peak occurs at a distance of approximately 40 m, then decreases, and occurs again at a distance of approximately 70 m. In other words, it is shown that two left turns have occurred.

[0086] The determination unit 2065 determines driving behavior, such as going straight or turning right or left, based on such changes in classification probability. The operation of turning off the turn signal 203 is controlled based on the peak value of the left-turn classification probability PL. Furthermore, the operation of turning off the turn signal 203 may be controlled not at the maximum point of the left-turn or right-turn classification probability, but at the point at which the straight-running classification probability becomes the largest among the classification probabilities after the left-turn or right-turn classification probability has reached its maximum. In this way, fluctuations in the maximum value, which is susceptible to road noise, due to road noise can be suppressed, thereby suppressing fluctuations in the timing of the turn-off determination. Such a determination is made possible by setting the predetermined period as the acquisition range of the reference driving data used for learning not back from the point at which the turn signal 203 was turned off, but back from a position prior to that point.

[0087] In the above example, the gyro sensor value and speed were used as parameters. By taking speed into consideration, it becomes possible to distinguish, for example, between a normal lane change that is made without slowing down and movement at a low speed at an intersection. Therefore, it becomes possible to turn off the turn indicator 203 only when traveling at high speed and changing lanes, without distinguishing between a right or left turn and a lane change.

[0088] Furthermore, in the first embodiment, the control unit 206 including the traveling data acquisition unit 2063 and the conversion unit 2064 may be provided in the handle unit (cockpit unit) where meters and electronic boards for control, etc. are provided. Since the processing by the control unit 206 is less susceptible to the influence of vibrations in the handle unit, it becomes possible to install a sensor unit including a gyro sensor, etc. in the handle unit.

[0089] Furthermore, in the first embodiment, the conversion unit 2064 may generate an image in which information is binarized according to the sensor values, and use the image for determination by the determination unit 2065. In Fig. 17, when a certain driving behavior is performed, the sensor values ​​of each axis (x-axis: direction of travel, y-axis: to the right of the direction of travel, z-axis: perpendicular to the x-axis and y-axis toward the ground) of the gyro sensor and the speed are shown as the x-axis gyro sensor value G1, the y-axis gyro sensor value G2, the z-axis gyro sensor value G3, and the speed V1, respectively, similar to Fig. 15 .

[0090] The converter 2064 may binarize the sensor values, for example, as shown in FIG. 18 , which illustrates an example of converting the x-axis gyro sensor value G1. In FIG. 18 , darker areas are represented by 1 and other areas by 0. In this manner, graph information can be generated as an image having a predetermined number of pixels. The converter 2064 may generate multiple images based on each gyro sensor value and speed, and the determination unit 2065 may use the generated images to determine driving behaviors such as going straight or turning right or left. Note that at this time, the reference driving data is also binarized and learned. Furthermore, by converting the image of the binarized sensor values ​​into an image of integrated values ​​using a cumulative function, the effects of fluctuations due to road noise, etc., can be offset by the cumulative function, thereby suppressing the occurrence of noise in the image and a decrease in determination accuracy. Furthermore, the converter 2064 may use an image of integrated sensor values ​​using a cumulative function based on the sensor values ​​as shown in FIG. 17 , rather than a binarized image, to determine driving behaviors such as going straight or turning right or left.

[0091] The above describes an exemplary embodiment of the present invention. The motorcycle 101 stores a program that causes the motorcycle 101 to acquire motorcycle driving data measured continuously over a predetermined period from at least one sensor, and to determine a corresponding driving action that corresponds to the driving data, based on the driving data and a plurality of reference driving data that are continuous over the predetermined period and to which each of a plurality of motorcycle driving actions is associated.

[0092] By determining the driving behavior of a motorcycle using motorcycle driving data measured continuously over a predetermined period of time, it is possible to avoid the influence of waiting time even when the motorcycle makes a temporary stop, and also to mitigate the influence of temporary noise, thereby improving the accuracy of determining the corresponding driving state.

[0093] In addition, in the case of the motorcycle 101, the multiple driving actions include turning right and turning left of the motorcycle, and the action control unit 2066 controls the motorcycle's turn indicator to switch from an on state to an off state based on the corresponding driving action of either turning right or turning left.

[0094] In addition, the motorcycle 101 calculates a classification probability, which is the probability that the driving data will be classified into at least one driving action, based on the driving data and multiple reference driving data, and determines the corresponding driving action based on the classification probability.

[0095] This makes it possible to determine a corresponding driving action that is more suitable for the driving data by, for example, setting a condition for the classification probability of at least one driving action. Here, the condition means determining the corresponding driving action as the driving action associated with the largest classification probability among multiple classification probabilities, or determining the corresponding driving action as the driving action associated with a classification probability that exceeds a predetermined threshold.

[0096] In the motorcycle 101, the determination unit 2065 inputs an input image as driving data into a learning model trained using teacher data that receives a plurality of pieces of reference driving data as input and outputs driving actions associated with each piece of reference driving data, and obtains corresponding driving actions as output from the learning model, thereby determining corresponding driving actions. This makes it possible to increase the accuracy of determining corresponding driving states.

[0097] In the motorcycle 101A, the learning model 20621A includes a learning model 20622 for right turn determination that determines a right turn and a learning model 20623 for left turn determination that determines a left turn, and the driving data acquisition unit 2063 acquires information indicating the operation when making the motorcycle turn right or left, and the judgment unit 2065 judges the corresponding driving action using the first learning model if the operation corresponds to a right turn, and judges the corresponding driving action using the second learning model if the operation corresponds to a left turn.

[0098] This allows the system to determine right and left turns using different learning models, improving the accuracy of the determination, even if the distances required to complete a right turn and a left turn are different.

[0099] For the motorcycle 101, the multiple driving actions further include right lane changes and left lane changes of the motorcycle, and the learning model 20621 is trained to be able to distinguish between right turns, left turns, right lane changes, and left lane changes. It is possible to distinguish between right lane changes, in which the turn signal 203 is operated in the same way as a right turn, and left lane changes, in which the turn signal 203 is operated in the same way as a left turn. This improves the accuracy of determining the corresponding driving action.

[0100] In the motorcycle 101, the driving data includes first driving data acquired over a predetermined period from a first timing, second driving data acquired over a predetermined period from a second timing after the first timing, and third driving data acquired over a predetermined period from a third timing after the second timing, and the classification probability of the corresponding driving action includes a first classification probability based on the first driving data, a second classification probability based on the second driving data, and a third classification probability based on the third driving data.

[0101] In addition, in the motorcycle 101, the operation control unit 2066 acquires the classification probability, and if the second classification probability is greater than the first classification probability and the third classification probability, controls the operation of the motorcycle based on the corresponding driving operation at a time after the third timing.

[0102] This makes it possible to determine the end of the corrective driving behavior based on the peak of the second classification probability, which is located at an intermediate timing. For example, the classification probability of a right turn (first classification probability) at the start of a right turn (first timing) is smaller than the classification probability of a right turn (second classification probability) during the right turn (second timing). The classification probability of a right turn (third classification probability) at the end of the right turn (third timing) is also smaller than the second classification probability. In this way, since the learning data is data for a certain period from the light-off timing, determining the end of the corrective driving behavior based on changes in the classification probability, which is the degree of similarity, makes it possible, for example, to turn off the turn signal 203 at a timing that is more natural to the driver.

[0103] In the motorcycle 101, the conversion unit 2064 generates input two-dimensional data based on the driving data, and the determination unit 2065 uses the input two-dimensional data as driving data and determines a corresponding driving action, which is a driving action corresponding to the driving data, based on the input two-dimensional data and multiple reference driving data. This makes it possible to capture correlations between variables by converting the driving data into, for example, a single image, even when the driving data includes multiple variables. By making a determination using multiple variables, the accuracy of determining the corresponding driving action is improved.

[0104] In the motorcycle 101, if the driving data is one-dimensional data, the conversion unit 2064 converts the driving data into two-dimensional data, and the input image is converted input data based on the driving data converted into two-dimensional data. This makes it possible to determine a corresponding driving behavior using two-dimensional data even if the sensor can acquire only one variable.

[0105] For the motorcycle 101, the predetermined period is the time it takes for the motorcycle to travel a predetermined distance. By basing the period on distance rather than time, it is possible to reduce the impact of driving actions that affect time but do not contribute to the progression of the motorcycle 101, such as stopping the motorcycle 101.

[0106] The motorcycle 101 further includes a communication unit 2061 that acquires information used to update the learning model 20621 from a terminal external to the motorcycle 101 and updates the learning model 20621. This makes it possible to easily update the learning model.

[0107] In addition, the motorcycle 101 has a learning model stored in the memory unit 2062 that is learned using teacher data that takes as input a plurality of reference driving data of the motorcycle that are continuous over a predetermined period of time and outputs driving actions associated with each of the plurality of reference driving data, and causes the control unit 206 to function such that, in response to input of motorcycle driving data that has been continuously measured over a predetermined period of time, corresponding driving actions are driving actions that correspond to the driving data.

[0108] In the case of the motorcycle 101, the plurality of reference driving data may include data acquired over a period going back from when the turn signal of the motorcycle was switched from an on state to an off state.

[0109] The plurality of reference driving data may include data acquired over a period going back from a point in time before the turn signal of the motorcycle was switched from an on state to an off state. In this case, the operation control unit 2066 may acquire classification probabilities, and if the first classification probability indicates a right turn or a left turn, the second classification probability indicates going straight, and the second classification probability is greater than the first classification probability, control the operation of the motorcycle based on the corresponding driving action at a time on or after the second timing. In this way, when the plurality of reference driving data is acquired over a period going back from a point in time before the turn signal was switched from an on state to an off state, fluctuations in the maximum classification probability, which is susceptible to road noise, due to road noise can be suppressed, and fluctuations in the timing to determine whether the turn signal is to be turned off can be suppressed.

[0110] In the motorcycle 101, the conversion unit 2064 may convert the driving data into an integrated value using a cumulative function to generate input two-dimensional data. This allows the cumulative function to offset the influence of fluctuations due to road noise and the like, thereby suppressing the occurrence of noise in the image and a decrease in determination accuracy.

[0111] In the motorcycle 101, the reference driving data may include data generated by an information processing device, rather than data obtained by actual driving. This allows data to be generated and used for machine learning even when it is not possible to obtain data by actual measurement.

[0112] In the motorcycle 101, the driving data acquisition unit 2063 and the determination unit 2065 may be provided in the handlebars of the motorcycle 101. The driving data and the reference driving data may be data acquired from a gyro sensor and a speed sensor provided in the handlebars.

[0113] The above-described embodiments are intended to facilitate understanding of the present invention and are not intended to limit the scope of the present invention. The present invention may be modified or improved without departing from its spirit, and equivalents are also included within the scope of the present invention. In other words, designs modified appropriately by those skilled in the art are also encompassed within the scope of the present invention as long as they incorporate the features of the present invention. For example, the elements of the embodiments, their arrangement, conditions, shape, size, etc., are not limited to those illustrated and may be modified as appropriate. Furthermore, the programs described in the above embodiments can be stored on a storage medium. The storage medium storing the programs may be a non-transitory computer-readable medium. The non-transitory storage medium is not particularly limited, and may be, for example, a USB memory stick, a CD-ROM, or other storage medium. Furthermore, the embodiments are merely examples, and it goes without saying that partial substitutions or combinations of the configurations shown in different embodiments are possible. These are also encompassed within the scope of the present invention as long as they incorporate the features of the present invention.

[0114] <1> A program that causes a computer to acquire, from at least one sensor, driving data of a motorcycle that has been continuously measured over a predetermined period of time, and determine a corresponding driving action that is a driving action that corresponds to the driving data, based on the driving data and a plurality of reference driving data that is continuous over the predetermined period and to which each of a plurality of driving actions of the motorcycle is associated. <2> The program described in <1>, wherein the plurality of driving actions include a right turn and a left turn of the motorcycle, and the program further causes the computer to control switching a turn signal of the motorcycle from an on state to an off state, based on the corresponding driving action of either the right turn or the left turn. <3> The program described in <1> or <2>, wherein the determining includes calculating a classification probability that the driving data will be classified into each of the at least one driving action, based on the driving data and the plurality of reference driving data, and determining the corresponding driving action based on the classification probability. <4> The program according to any one of <1> to <3>, wherein the determining includes determining the corresponding driving action by inputting the driving data into a learning model that has been trained to be able to distinguish between the right turn and the left turn using teacher data that has the plurality of reference driving data as input and has driving actions associated with each of the plurality of reference driving data as output, and obtaining the corresponding driving action as output from the learning model. <5> The program according to <4>, wherein the learning model includes a first learning model that determines the right turn and a second learning model that determines the left turn, and further causes the computer to obtain information indicating an operation for causing the motorcycle to make the right turn or the left turn, and the determining includes determining the corresponding driving action using the first learning model if the operation corresponds to the right turn, and determining the corresponding driving action using the second learning model if the operation corresponds to the left turn.<6> The program according to <4> or <5>, wherein the plurality of driving actions further include a right lane change and a left lane change of the motorcycle, and the learning model is trained to be able to distinguish between the right turn, the left turn, the right lane change, and the left lane change. <7> The program according to <3>, wherein the driving data includes first driving data acquired over the predetermined period from a first timing, second driving data acquired over the predetermined period from a second timing after the first timing, and third driving data acquired over the predetermined period from a third timing after the second timing, and the classification probabilities of the corresponding driving behavior include a first classification probability based on the first driving data, a second classification probability based on the second driving data, and a third classification probability based on the third driving data, and the controlling further includes acquiring the classification probabilities, and, if the second classification probability is greater than the first classification probability and the third classification probability, controlling the operation of the motorcycle based on the corresponding driving behavior at a time after the third timing. <8> The program according to <1> to <7>, further causing the computer to generate input two-dimensional data based on the driving data, and the determining includes determining a corresponding driving action that is a driving action corresponding to the driving data based on the input two-dimensional data and the plurality of reference driving data, using the input two-dimensional data as the driving data, and determining a corresponding driving action that is a driving action that is a driving action that corresponds to the driving data based on the input two-dimensional data and the plurality of reference driving data, using the input two-dimensional data as the driving data. <9> The program according to <8>, further causing the computer to convert the driving data into two-dimensional data if the driving data is one-dimensional data, and the input two-dimensional data is data based on the driving data converted into two-dimensional data. <10> The program according to any of <1> to <9>, wherein the predetermined period is a range required for the motorcycle to travel a predetermined distance.<11> A device comprising: an acquisition unit that acquires motorcycle driving data measured continuously over a predetermined period from at least one sensor; and a determination unit that determines a corresponding driving action corresponding to the driving data based on the driving data and a plurality of reference driving data that are continuous over the predetermined period and are associated with each of a plurality of driving actions of the motorcycle. <12> The device described in <11>, wherein the determination unit inputs the driving data to a learning model trained using teacher data that takes the plurality of reference driving data as input and outputs driving actions associated with each of the plurality of reference driving data, and obtains the corresponding driving action as output from the learning model, thereby determining the corresponding driving action, and further comprising a communication unit that acquires information used to update the learning model from a terminal external to the device and updates the learning model. <13> The device described in <11> or <12>, wherein the acquisition unit acquires at least a portion of the driving data from a terminal that is capable of communicating with the acquisition unit. <14> The device according to any one of <11> to <13>, wherein the acquisition unit acquires at least a portion of the driving data from a radar or a camera. <15> The device according to any one of <11> to <14>, wherein the plurality of driving actions include a right turn and a left turn of the motorcycle, and further comprising an operation control unit that switches a turn signal of the motorcycle from an on state to an off state based on the corresponding driving action of either the right turn or the left turn. <16> The device according to <15>, wherein the determination unit calculates a classification probability, which is a probability that the driving data will be classified into each of the plurality of driving actions, based on the driving data and the plurality of reference driving data, and determines the corresponding driving action based on the classification probability.<17> The device according to <15> or <16>, wherein the determination unit determines the corresponding driving action by inputting the driving data into a learning model that has been trained to be able to distinguish between a right turn and a left turn of the motorcycle using teacher data that has the plurality of reference driving data as input and outputs driving actions associated with each of the plurality of reference driving data, and obtaining the corresponding driving action as an output from the learning model. <18> The device according to <17>, wherein the learning models include a first learning model that determines the right turn and a second learning model that determines the left turn, and further comprising an acquisition unit that acquires information indicative of an operation to cause the motorcycle to make the right turn or the left turn, wherein the determination unit determines the corresponding driving action using the first learning model when the operation corresponds to the right turn, and determines the corresponding driving action using the second learning model when the operation corresponds to the left turn. <19> The device according to <17>, wherein the plurality of driving actions further include a right lane change and a left lane change of the motorcycle, and the learning model is trained to be able to distinguish between the right turn, the left turn, the right lane change, and the left lane change. <20> The device according to any one of <15> to <19>, wherein the plurality of reference driving data includes data acquired over a period going back from a point in time when the turn indicator of the motorcycle was switched from the on state to the off state.<21> The device described in <20>, wherein the driving data includes first driving data acquired over the predetermined period from a first timing, second driving data acquired over the predetermined period from a second timing after the first timing, and third driving data acquired over the predetermined period from a third timing after the second timing, and the classification probabilities of the corresponding driving action include a first classification probability based on the first driving data, a second classification probability based on the second driving data, and a third classification probability based on the third driving data, and an operation control unit that switches a turn signal of the motorcycle from an on state to an off state based on the corresponding driving action of either turning the motorcycle right or turning the motorcycle left acquires the classification probabilities, and if the second classification probability is greater than the first classification probability and the third classification probability, controls the operation of the motorcycle based on the corresponding driving action at a time after the third timing. <22> The device according to any one of <15> to <19>, wherein the plurality of reference driving data includes data acquired over a period going back from a point in time before a turn signal of the motorcycle was switched from an on state to an off state. <23> The device described in <22>, wherein the driving data includes first driving data acquired over the predetermined period from a first timing, and second driving data acquired over the predetermined period from a second timing after the first timing, and the classification probabilities of the corresponding driving action include a first classification probability based on the first driving data and a second classification probability based on the second driving data, and an operation control unit that switches a turn signal of the motorcycle from an on state to an off state based on the corresponding driving action of either a right turn of the motorcycle or a left turn of the motorcycle acquires the classification probabilities, and, if the first classification probability indicates the right turn or the left turn, the second classification probability indicates going straight, and the second classification probability is greater than the first classification probability, further controls the operation of the motorcycle based on the corresponding driving action at a time after the second timing.<24> The device according to any one of <11> to <23>, further comprising a conversion unit that generates input two-dimensional data based on the driving data, wherein the determination unit uses the input two-dimensional data as the driving data and determines a corresponding driving action that is a driving action corresponding to the driving data based on the input two-dimensional data and the plurality of reference driving data. <25> The device according to <24>, wherein, if the driving data is one-dimensional data, the conversion unit converts the driving data into two-dimensional data, and the input two-dimensional data is data based on the driving data converted into the two-dimensional data. <26> The device according to <24> or <25>, wherein the conversion unit converts the driving data into an integrated value using a cumulative function to generate the input two-dimensional data. <27> The device according to any one of <11> to <26>, wherein the predetermined period is a time range required for the motorcycle to travel a predetermined distance. <28> The device according to any one of <11> to <27>, wherein the reference driving data is different from data obtained by actual driving and includes data generated by an information processing device. <29> The device according to any one of <11> to <28>, wherein the acquisition unit and the determination unit are provided on a handlebar of the motorcycle. <30> The device according to <29>, wherein the driving data and the reference driving data are data obtained from a gyro sensor and a speed sensor provided on the handlebar. <31> A learning model that learns using teacher data that receives as input a plurality of reference driving data of a motorcycle that are continuous over a predetermined period and outputs driving actions associated with each of the plurality of reference driving data, and causes a computer to function in response to input of driving data of the motorcycle that is continuously measured over the predetermined period, to output corresponding driving actions that are driving actions corresponding to the driving data. <32> The learning model according to <31>, wherein the predetermined period is the time range required for the motorcycle to travel a predetermined distance.<33> A method including, by a computer, acquiring motorcycle driving data measured continuously over a predetermined period from at least one sensor, and determining a corresponding driving action corresponding to the driving data based on the driving data and a plurality of reference driving data that are continuous over the predetermined period and to which each of a plurality of driving actions of the motorcycle is associated. <34> The method according to <33>, wherein the predetermined period is a time range required for the motorcycle to travel a predetermined distance.

[0115] 101, 101A, 101B...motorcycle, 203...direction indicator, 205...sensor unit, 206, 206A...control unit, 1100...motorcycle control system, 1101...determination device, 1102...sensing terminal, 2063...traveling data acquisition unit, 2064...conversion unit, 2065...determination unit, 2066...operation control unit, 20612...learning model

Claims

1. On the computer, acquiring motorcycle driving data continuously measured over a predetermined period of time from at least one sensor; and determining a corresponding driving action, which is a driving action that corresponds to the driving data, based on the driving data and a plurality of reference driving data that continues over the specified period and to which each of a plurality of driving actions of the motorcycle is associated.

2. The program according to claim 1, the plurality of driving actions include right turns and left turns of the motorcycle; The computer includes: The program further causes the computer to execute control to switch a direction indicator of the motorcycle from an on state to an off state based on the corresponding driving action of either the right turn or the left turn.

3. 3. The program according to claim 1 or 2, The determining step comprises: A program comprising: calculating a classification probability, which is the probability that the driving data is classified into each of the plurality of driving actions based on the driving data and the plurality of reference driving data; and determining the corresponding driving action based on the classification probability.

4. The program according to claim 1, The determining step comprises: a program for determining the corresponding driving action by inputting the plurality of reference driving data into a learning model that has been trained to distinguish between a right turn and a left turn of the motorcycle using teacher data that has as input the plurality of reference driving data and outputs driving actions associated with each of the plurality of reference driving data, and obtaining the corresponding driving action as an output from the learning model.

5. The program according to claim 4, The learning model includes a first learning model for determining the right turn and a second learning model for determining the left turn, The computer includes: acquiring information indicating an operation for making the motorcycle turn right or left; The determining step comprises: If the operation corresponds to the right turn, determining the corresponding driving action using the first learning model; If the operation corresponds to the left turn, the program determines the corresponding driving action using the second learning model.

6. The program according to claim 4 or 5, the plurality of driving actions further include a right lane change and a left lane change of the motorcycle; The learning model is trained to be able to distinguish between the right turn, the left turn, the right lane change, and the left lane change.

7. The program according to claim 3, The running data is first running data acquired over the predetermined period from a first timing, second running data acquired over the predetermined period from a second timing after the first timing, and third running data acquired over the predetermined period from a third timing after the second timing, The classification probability of the corresponding driving action is a first classification probability based on the first traveling data, a second classification probability based on the second traveling data, and a third classification probability based on the third traveling data, The program further causes the computer to execute control to switch a turn indicator of the motorcycle from an on state to an off state based on the corresponding driving action, either a right turn of the motorcycle or a left turn of the motorcycle, and to acquire the classification probability, and, if the second classification probability is greater than the first classification probability and the third classification probability, to control the operation of the motorcycle based on the corresponding driving action at a time after the third timing.

8. The program according to claim 1, The computer includes: generating input two-dimensional data based on the running data; The determining step comprises: A program that uses the input two-dimensional data as the driving data and determines a corresponding driving action that is a driving action that corresponds to the driving data based on the input two-dimensional data and the multiple reference driving data.

9. The program according to claim 8, The computer includes: If the traveling data is one-dimensional data, converting the traveling data into two-dimensional data; The input two-dimensional data is data based on the traveling data converted into the two-dimensional data.

10. The program according to claim 1, The predetermined period is a time range required for the motorcycle to travel a predetermined distance.

11. an acquisition unit that acquires motorcycle driving data continuously measured over a predetermined period from at least one sensor; and a determination unit that determines a corresponding driving action corresponding to the driving data based on the driving data and a plurality of reference driving data that are continuous over the specified period and to which each of a plurality of driving actions of the motorcycle is associated.

12. 12. The apparatus of claim 11, The determination unit is inputting the plurality of reference driving data into a learning model trained using teacher data having driving actions associated with each of the plurality of reference driving data as an input, and obtaining the corresponding driving action as an output from the learning model, thereby determining the corresponding driving action; The apparatus further includes a communication unit that acquires information used to update the learning model from a terminal external to the apparatus and updates the learning model.

13. 13. An apparatus according to claim 11 or 12, comprising: The acquisition unit acquires at least a portion of the driving data from a terminal capable of communicating with the acquisition unit.

14. 12. The apparatus of claim 11, The acquisition unit acquires at least a portion of the driving data from a radar or a camera.

15. 12. The apparatus of claim 11, the plurality of driving actions include right turns and left turns of the motorcycle; The apparatus further includes an operation control unit that switches a turn indicator of the motorcycle from an on state to an off state based on the corresponding driving operation of either the right turn or the left turn.

16. 16. The apparatus of claim 15, The determination unit is The device calculates a classification probability, which is the probability that the driving data will be classified into each of the multiple driving actions, based on the driving data and the multiple reference driving data, and determines the corresponding driving action based on the classification probability.

17. 17. Apparatus according to claim 15 or 16, comprising: The determination unit is The device inputs the multiple reference driving data and outputs driving actions associated with each of the multiple reference driving data, and inputs the driving data into a learning model that has been trained to distinguish between a right turn and a left turn of the motorcycle using teacher data.The device determines the corresponding driving action by obtaining the corresponding driving action as an output from the learning model.

18. 18. The apparatus of claim 17, The learning model includes a first learning model for determining the right turn and a second learning model for determining the left turn, an acquisition unit that acquires information indicating an operation for making the motorcycle turn right or left, The determination unit is If the operation corresponds to the right turn, determining the corresponding driving action using the first learning model; If the maneuver corresponds to the left turn, the apparatus determines the corresponding driving action using the second learning model.

19. 18. The apparatus of claim 17, the plurality of driving actions further include a right lane change and a left lane change of the motorcycle; The learning model is trained to be able to distinguish between the right turn, the left turn, the right lane change, and the left lane change.

20. 16. The apparatus of claim 15, The plurality of reference driving data includes data acquired over a period going back from when the turn signal of the motorcycle was switched from the on state to the off state.

21. 21. The apparatus of claim 20, The running data is first running data acquired over the predetermined period from a first timing, second running data acquired over the predetermined period from a second timing after the first timing, and third running data acquired over the predetermined period from a third timing after the second timing, The classification probability of the corresponding driving action is a first classification probability based on the first traveling data, a second classification probability based on the second traveling data, and a third classification probability based on the third traveling data, An operation control unit that switches a turn signal of the motorcycle from an on state to an off state based on the corresponding driving action of either a right turn of the motorcycle or a left turn of the motorcycle acquires the classification probability, and, if the second classification probability is greater than the first classification probability and the third classification probability, controls the operation of the motorcycle based on the corresponding driving action at a time after the third timing.

22. 16. The apparatus of claim 15, The plurality of reference driving data includes data acquired over a period of time prior to a time when a turn signal of the motorcycle was switched from an on state to an off state.

23. 23. The apparatus of claim 22, The running data is First running data is acquired over the predetermined period from a first timing, and second running data is acquired over the predetermined period from a second timing that is later than the first timing, The classification probability of the corresponding driving action is a first classification probability based on the first running data; and a second classification probability based on the second running data, The device further executes an operation control unit that switches a turn indicator of the motorcycle from an on state to an off state based on the corresponding driving action of either a right turn of the motorcycle or a left turn of the motorcycle, and acquires the classification probability, and when the first classification probability indicates the right turn or the left turn, the second classification probability indicates going straight, and the second classification probability is greater than the first classification probability, controls the operation of the motorcycle based on the corresponding driving action at a time after the second timing.

24. 12. The apparatus of claim 11, A conversion unit that generates input two-dimensional data based on the running data, The determination unit is The device determines a corresponding driving action that corresponds to the driving data based on the input two-dimensional data and the plurality of reference driving data, using the input two-dimensional data as the driving data.

25. 25. The apparatus of claim 24, The conversion unit converts the traveling data into two-dimensional data when the traveling data is one-dimensional data, The input two-dimensional data is data based on the driving data converted into the two-dimensional data.

26. 26. Apparatus according to claim 24 or 25, comprising: The conversion unit converts the driving data into an integrated value using a cumulative function to generate the input two-dimensional data.

27. 12. The apparatus of claim 11, The predetermined period of time is a range of time required for the motorcycle to travel a predetermined distance.

28. 12. The apparatus of claim 11, The reference driving data is different from data obtained by actual driving and includes data generated by an information processing device.

29. 12. The apparatus of claim 11, The acquisition unit and the determination unit are provided in a handle portion of the motorcycle.

30. 30. The apparatus of claim 29, wherein the driving data and the reference driving data are data obtained from a gyro sensor and a speed sensor provided in the handle portion.

31. A plurality of reference driving data of a motorcycle, which are consecutive over a predetermined period of time, is input, and training is performed using teacher data, which outputs driving actions associated with each of the plurality of reference driving data; Computer, A learning model that functions to output, in response to input of driving data of the motorcycle measured continuously over the specified period, a corresponding driving action that is a driving action corresponding to the driving data.

32. 32. A learning model according to claim 31, comprising: The predetermined period is a time range required for the motorcycle to travel a predetermined distance.

33. The computer acquiring motorcycle driving data continuously measured over a predetermined period of time from at least one sensor; determining a corresponding driving action corresponding to the driving data based on the driving data and a plurality of reference driving data that are continuous over the predetermined period and to which each of a plurality of driving actions of the motorcycle is associated.

34. 34. The method of claim 33, The method, wherein the predetermined period of time is a range of time it takes the motorcycle to travel a predetermined distance.