Behavior estimation system and behavior estimation method of saddle riding vehicle

Abstract

To use a sensor included in a general saddle riding vehicle to estimate whether the saddle riding vehicle has jumped.SOLUTION: A behavior estimation system 30 analyzes a behavior of a motorcycle 10. The motorcycle 10 includes a vehicle body 11, an engine 17, and a rear wheel 13 that runs using a drive force generated by the engine 17. The behavior estimation system 30 includes a receiver 31 and processor 32. The receiver 31 receives an output instruction value of the engine 17 issued by a driver and detected by a sensor, which is included in the motorcycle 10, during travel of the motorcycle 10, and a rotation speed of the engine 17. Based on the output instruction value and rotation speed received by the receiver 31, the processor 32 estimates whether a front wheel 12 and the rear wheel 13 are in contact with a road surface, that is, the motorcycle is grounded or the front wheel 12 and rear wheel 13 are separated from the road surface, that is, the motorcycle have jumped.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present application relates to an apparatus and method for analyzing the behavior of a saddle-type vehicle. [Background technology]

[0002] Patent Document 1 discloses a posture estimation device that estimates the posture of a motorcycle, which is one type of saddle-ride type vehicle. In detail, sensors such as a GPS antenna or an IMU are arranged on each of the frame, front fork, and swing arm of the motorcycle. The posture estimation device detects the posture angles of each part based on the detection results of three sensors provided on the motorcycle, and estimates the posture of the motorcycle. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2019-64469 A Summary of the Invention [Problem to be solved by the invention]

[0004] The posture estimation device of Patent Document 1 targets a motorcycle posture calculated from the posture angles of the frame, front fork, and swing arm. Therefore, Patent Document 1 does not mention detecting whether the motorcycle is jumping or not. In addition, since a typical motorcycle is not provided with three sensors such as a GPS antenna or an IMU, in order to use the posture estimation device of Patent Document 1, it is necessary to provide the motorcycle with a sensor dedicated to posture detection. Furthermore, these problems are not limited to motorcycles, but are also problems that arise in other saddle-type vehicles.

[0005] The present application has been made in consideration of the above circumstances, and a primary object of the application is to provide a behavior estimation device that is capable of estimating whether or not a saddle-ride type vehicle is jumping by using sensors that are equipped on typical saddle-ride type vehicles. [Means for solving the problem]

[0006] The problem to be solved by the present application is as described above. Next, the means for solving this problem and the effects thereof will be described.

[0007] According to a first aspect of the present invention, there is provided a behavior estimation device having the following configuration. That is, the behavior estimation device analyzes the behavior of a saddle-ride type vehicle including a vehicle body, a drive source, and a ground-contacting traveling body that travels using a driving force generated by the drive source. The behavior estimation device includes a receiving device and a processing device. The receiving device receives an output command value of the drive source by a driver and a rotation speed of the drive source, which are detected by a sensor provided in the saddle-ride type vehicle while the saddle-ride type vehicle is traveling. The processing device estimates whether the ground-contacting traveling body is in a ground contact state in which it is in contact with the road surface, or in a jump state in which the ground-contacting traveling body is separated from the road surface, based on the output command value and the rotation speed received by the receiving device.

[0008] According to a second aspect of the present invention, the following behavior estimation method is provided. That is, in the behavior estimation method, the behavior of a saddle-ride type vehicle including a vehicle body, a drive source, and a ground-contacting traveling body that is in contact with a road surface and travels by a driving force generated by the drive source is analyzed. In the behavior estimation method, a behavior instruction value by a driver that is detected by a sensor while the saddle-ride type vehicle is traveling is received. In the behavior estimation method, based on the received behavior instruction value, it is estimated whether the ground-contacting traveling body is in a ground contact state in which it is in contact with the road surface, or in a jump state in which the ground-contacting traveling body is separated from the road surface. Effect of the Invention

[0009] According to the present application, it is possible to estimate whether or not a saddle-ride type vehicle is jumping by using a sensor that is provided in a typical saddle-ride type vehicle. [Brief description of the drawings]

[0010] [Figure 1] FIG. 2 is a block diagram of a motorcycle and a behavior estimation device. [Diagram 2]FIG. [Diagram 3] 13 is a flowchart for creating an estimation model. [Figure 4] 13 is a flowchart of an estimation process using an estimation model. [Diagram 5] FIG. 1 is an explanatory diagram conceptually illustrating an estimation process using an estimation model. [Figure 6] FIG. 13 is a diagram showing an image in which the estimation result is superimposed on a map. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] Next, an embodiment of the present application will be described with reference to the drawings. Fig. 1 is a block diagram of a motorcycle 10 and a behavior estimation device 30.

[0012] The motorcycle 10 is a type of saddle-ride type vehicle. The motorcycle 10 is a type of vehicle that mainly travels on unpaved roads. In particular, the motorcycle 10 is used in races on unpaved roads. In this type of race, there is an uphill slope as shown in FIG. 2, and the rider may make the motorcycle 10 jump at the end of the uphill slope. The rider may also make the motorcycle 10 jump over a section that transitions from a flat road to a downhill slope or over a section that has large irregularities.

[0013] The behavior estimation device 30 estimates the behavior of the motorcycle 10 based on detection values ​​of sensors provided on the motorcycle 10. In detail, the behavior estimation device 30 estimates whether the motorcycle 10 is in a ground contact state or a jump state. The ground contact state is a state in which at least one of the front wheel 12 and the rear wheel 13 of the motorcycle 10 is in contact with the road surface. The jump state is a state in which both the front wheel 12 and the rear wheel 13 of the motorcycle 10 are off the road surface.

[0014] The behavior estimation device 30 in this embodiment is a motorcycle 10 used in races on unpaved roads. However, the behavior estimation device 30 can also be applied to motorcycles that mainly run on paved roads, and motorcycles that can be run not only in races but also on public roads. In addition, the behavior estimation device 30 can also be applied to vehicles with three or more wheels, snowmobiles, etc., without being limited to motorcycles.

[0015] The detailed configurations of the motorcycle 10 and the behavior estimation device 30 will be described below.

[0016] The motorcycle 10 includes a body 11, a front wheel 12, and a rear wheel 13. The body 11 includes a body frame and a cowl. The cowl may be omitted. The front wheel 12 is attached to the body 11 via a front fork 14. The front fork 14 is of a telescopic type and includes a front suspension function. The front wheel 12 is attached to the body 11 via a swing arm 15. The swing arm 15 is capable of swinging relative to the body 11. A rear suspension 16 is provided between the body 11 and the swing arm 15, and vibrations of the rear wheel 13 are reduced by the rear suspension 16 and transmitted to the body 11.

[0017] An engine 17 is attached to the vehicle body 11. The engine 17 generates driving force using fuel such as gasoline, diesel, or gaseous fuel. The driving force generated by the engine 17 is reduced in speed by a transmission provided integrally with the engine 17 and transmitted to the rear wheels 13, which are driving wheels. The engine 17 corresponds to the driving source. However, the driving source is not limited to an engine, and may be an electric motor. Also, the vehicle may be provided with both an engine and an electric motor. In this case, both the engine and the electric motor correspond to the driving source.

[0018] The front wheel 12 and the rear wheel 13 correspond to the ground-contacting body. The ground-contacting body is a part that contacts the road surface and travels using a driving force generated by a driving source such as the engine 17. The rear wheel 13 travels by being rotated and driven by the driving force generated by the engine 17. The front wheel 12 travels by rotating and driven in response to the rear wheel 13 being rotated and driven. In this way, the front wheel 12 and the rear wheel 13 are parts that travel using a driving force. The ground-contacting body is not limited to a wheel, but may be a crawler. For example, if the saddle-type vehicle is a snowmobile, the crawler corresponds to the ground-contacting body. The ground-contacting body is, in other words, a road surface engaging member or a road surface contact member.

[0019] The motorcycle 10 further includes a steering handle 18 and a seat 19. The steering handle 18 is a part that is operated by the rider while the motorcycle 10 is traveling. For example, the rider can turn the motorcycle 10 by rotating the steering handle 18 about a steering axis. The grip on the right side of the steering handle 18 is a throttle grip 18a. The rider can increase the amount of fuel supplied to the engine 17 by rotating the throttle grip 18a in a predetermined direction about the axial direction, thereby accelerating the motorcycle 10. In other words, the throttle grip 18a is an operator for instructing a change in the output of the drive source. Note that an accelerator pedal or an accelerator lever may be provided instead of the throttle grip 18a. The seat 19 is a part on which the rider sits. The seat 19 is narrow, and the rider sits on the seat 19 by straddling it. Therefore, the motorcycle 10 corresponds to a saddle-ride type vehicle.

[0020] As shown in FIG. 1, the motorcycle 10 is provided with a control device 21 and various sensors. The control device 21 includes a calculation device such as a CPU, a nonvolatile memory such as an SSD or a flash memory, a main memory such as a RAM, and a communication device capable of short-distance wireless communication. While the motorcycle 10 is traveling, the control device 21 receives sensor values ​​detected by the sensors. The control device 21 stores the input sensor values ​​in the nonvolatile memory. Then, while the motorcycle 10 is traveling, the control device 21 transmits the sensor values ​​to the behavior estimation device 30 using the communication device. Note that it is not essential to transmit the sensor values ​​to the behavior estimation device 30 while the motorcycle 10 is traveling, and the sensor values ​​stored in the nonvolatile memory may be transmitted to the behavior estimation device 30 after the motorcycle 10 has completed traveling. Specifically, the sensor values ​​may be transmitted from the control device 21 to the behavior estimation device 30 by wired communication or wireless communication, or the nonvolatile memory may be removed from the control device 21 and attached to the behavior estimation device 30 to read out the sensor values.

[0021] The sensors provided on the motorcycle 10 include a GNSS sensor 22, a rotation speed sensor 23, a gear shift sensor 24, a grip sensor 25, and a stroke sensor 26. The GNSS sensor 22 includes a GNSS antenna and a GNSS receiver. The GNSS receiver performs positioning calculations based on the positioning signal received by the GNSS antenna, and calculates position information indicating the absolute position of the vehicle. The rotation speed sensor 23 detects the rotation speed of the engine 17. The gear shift sensor 24 is a sensor that detects a gear position indicating the number of stages of the gear shift device. The grip sensor 25 detects the rotation angle of the throttle grip 18a, in other words, the output instruction value for the engine 17. The stroke sensor 26 detects the stroke of the suspension. The detection target of the stroke sensor 26 may be only the front fork 14, only the rear suspension 16, or both the front fork 14 and the rear suspension 16.

[0022] These sensor values ​​are acquired at a predetermined sampling rate and stored in the control device 21 in association with the acquisition time. That is, time-series data of the sensor values ​​is stored in the control device 21. Time-series data is data associated with information indicating timing such as time. Therefore, for example, a number string or character string indicating timing may be used instead of time. Furthermore, if the measurement start timing and sampling rate are consistent for multiple types of sensor values, data in which multiple sensor values ​​are simply arranged in order also corresponds to time-series data. Furthermore, these sensors are known sensors that are equipped in advance in a typical motorcycle 10. Therefore, explanations of the positions at which the sensors are provided and the principles of detection of sensor values ​​will be omitted.

[0023] Furthermore, the control device 21 calculates the vehicle speed and longitudinal acceleration based on the time-series data of the position information. The distance traveled by the motorcycle 10 in a given time can be determined based on the time-series data of the position information, so that the vehicle speed can be calculated. The vehicle speed calculated in this manner can be measured even when the motorcycle 10 is in a jumping state. The longitudinal acceleration is the acceleration of the motorcycle 10 in the fore-aft direction. The longitudinal acceleration can be calculated by calculating the change in vehicle speed over time. Note that instead of the method of calculating the vehicle speed and longitudinal acceleration using the position information, the vehicle speed and longitudinal acceleration may be calculated using a vehicle speed sensor or an acceleration sensor.

[0024] The behavior estimation device 30 is a computer including a receiving device 31, a processing device 32, a storage device 33, a display device , and a transmitting device .

[0025] The receiving device 31 receives the sensor values ​​from the control device 21. Therefore, when the control device 21 transmits the sensor values ​​using short-range wireless communication, the receiving device 31 is a wireless communication module. When the control device 21 transmits the sensor values ​​via a wire, the receiving device 31 is a wired communication module. When the non-volatile memory of the control device 21 is removed and attached to the behavior estimation device 30, the receiving device 31 is a reading device.

[0026] The processing device 32 is a calculation device such as a CPU. The processing device 32 can execute various processes by reading a program into a main memory and executing it. For example, the processing device 32 executes the above-mentioned process of estimating whether the motorcycle 10 is in a ground contact state or a jump state.

[0027] The storage device 33 is a non-volatile memory such as an HDD, an SSD, or a flash memory, and stores the above-mentioned programs or control data. The sensor values ​​received by the receiving device 31 are stored in the storage device 33.

[0028] The display device 34 is a liquid crystal or organic EL display, and is capable of displaying various information. The display device 34 is not an essential component and can be omitted.

[0029] The transmitting device 35 transmits the estimation result of the processing device 32 to an external device. The external device is, for example, the driver's smartphone, tablet terminal, PC, etc. Alternatively, the external device may be a server managed by a company that operates the behavior estimation device 30. The transmitting device 35 communicates with the external device using the Internet or short-range wireless communication, etc. When the receiving device 31 and the transmitting device 35 use the same communication method, one module serves as both the receiving device 31 and the transmitting device 35. Note that the transmitting device 35 is not an essential component and can be omitted.

[0030] Next, the state of the motorcycle 10 will be described with reference to Fig. 2. As described above, the state of the motorcycle 10 is divided into a jump state and a ground contact state. The definitions of the jump state and the ground contact state are as described above, but they may be defined from another perspective. For example, in the jump state, the suspension is in an extended state because the weight of the vehicle itself is less likely to act on the suspension. Therefore, the jump state may be defined as a case in which the stroke of the suspension detected by the stroke sensor is longer than a predetermined value, and the other states may be defined as ground contact states. The ground contact state is further divided into a pre-jump ground contact state, a landing state, and a normal ground contact state.

[0031] The pre-jump ground contact state is a state during the transition to the jump state. For example, the start timing of the pre-jump ground contact state may be the moment a predetermined number of seconds before the transition to the jump state, the start timing of the pre-jump ground contact state may be the moment a specific operation is performed when transitioning to the jump state, the start timing of the pre-jump ground contact state may be the moment the front wheel 12 leaves the road surface when transitioning to the jump state, or the start timing of the pre-jump ground contact state may be the moment the suspension starts to change when transitioning to the jump state.

[0032] The landing state is a state when landing from a jump state. For example, the landing state is a state from the moment when the jump state transitions to the ground state until a predetermined number of seconds have elapsed. Alternatively, the landing state may be a state from the moment when one wheel contacts the road surface until both wheels contact the road surface. Alternatively, the landing state may be a state until the suspension fluctuation when landing from a jump state falls below a threshold value.

[0033] The normal ground contact state is a ground contact state other than the pre-jump ground contact state and the landing state. It is not essential to further divide the ground contact state into three states. The state division of the motorcycle 10 in this embodiment is an example. Therefore, the jump state may be divided into multiple states. For example, the jump state may be divided into a small jump state and a large jump state depending on the duration or flying distance of the jump state.

[0034] In this embodiment, as a preliminary step, an estimation model is created by performing machine learning on the time-series data of the learning sensor values. Then, during operation, the time-series data of the judgment sensor values ​​is input to the estimation model, and the estimation model outputs an estimation result of the state of the motorcycle 10.

[0035] Here, the basis for being able to estimate the state of the motorcycle 10 using the sensor value of this embodiment will be described. Just before jumping, the driver sets the output command value to a very large value in order to jump. After that, the driver significantly reduces the output command value in order to control the attitude of the motorcycle 10 in the jumping state. Specifically, the driver often reduces the output command value to 0 in the jumping state. Then, when landing, the driver often sets the output command value to a maximum or a very large value in order to reduce impact. In this way, the output command value shows a characteristic change before and after a jump. Therefore, when the output command value shows such a characteristic change, it is determined that the motorcycle is in a jumping state, and other states can be estimated as a ground contact state.

[0036] In this way, it is possible to estimate whether the motorcycle 10 is in a jump state or a ground contact state using only the output command value, but in order to improve the estimation accuracy, it is preferable to use, for example, the engine rotation speed. The output command value and the engine rotation speed are correlated, and the engine rotation speed increases by increasing the output command value. However, when the motorcycle 10 is in a jump state, the load on the engine 17 is smaller than when the motorcycle 10 is in a ground contact state, and therefore a different correlation is shown. In other words, it is possible to estimate whether the motorcycle 10 is in a jump state or a ground contact state by analyzing the correlation between the output command value and the engine rotation speed. As described above, it is possible to improve the estimation accuracy by using the output command value and the engine rotation speed.

[0037] The ideal gear position immediately before a jump may differ from the ideal gear position immediately after landing. In this case, the rider changes the gear position while in a jump state. In a jump state, there is no need to obtain propulsive force from the road surface via the rear wheel 13, so the operation of changing the gear position may differ from the operation of changing the gear position when the motorcycle 10 is in a grounded state. Therefore, by further using the gear position, it may be possible to improve the estimation accuracy.

[0038] Furthermore, when the motorcycle 10 is in a grounded state, since the propulsive force is obtained from the road surface via the rear wheel 13, if the engine speed and the gear position are known, the vehicle speed can be estimated with some degree of accuracy. Furthermore, if the time series data of the engine speed and the gear position are known, the longitudinal acceleration can be estimated with some degree of accuracy. However, when the motorcycle 10 is in a jumping state, since the propulsive force is not obtained via the rear wheel 13, the vehicle speed or longitudinal acceleration estimated from the engine speed and the gear position may differ greatly from the actual vehicle speed or longitudinal acceleration. Therefore, the estimation accuracy can be improved by further using the vehicle speed or longitudinal acceleration. Furthermore, when the motorcycle 10 is in a grounded state, the vehicle speed is likely to decrease due to resistance from the road surface, but when the motorcycle 10 is in a jumping state, the vehicle speed is unlikely to decrease because there is no resistance from the road surface. Therefore, even if this point is taken into consideration, the change in the vehicle speed or longitudinal acceleration differs depending on whether the motorcycle 10 is in a grounded state or in a jumping state. Therefore, even from this viewpoint, the estimation accuracy can be improved by using the vehicle speed or longitudinal acceleration.

[0039] The types of sensor values ​​described above are merely examples, and some types of sensor values ​​may be omitted. For example, it is preferable to include the output command value since it is an important value for estimating the state of the motorcycle 10. It is also preferable to further include, for example, the engine speed in case the estimation accuracy is low using only the output command value. Also, instead of or in addition to the engine speed, the gear position, the vehicle speed, or the longitudinal acceleration may be added.

[0040] In addition, in the jumping state, the rider may control the attitude of the vehicle by applying the brakes to reduce the rotational speed of the front wheel 12 and the rear wheel 13. The brake operation at this time is aimed at attitude control and is significantly different from the brake operation aimed at braking. Therefore, it is possible to detect whether the motorcycle 10 is in the grounded state or in the jumping state based on the brake operation. In this case, it is necessary to provide a sensor that detects the amount of operation of the brake lever or the brake pedal. Also, the sensor value of the sensor of the brake lever or the brake pedal corresponds to the behavior instruction value. Also, the above-mentioned output instruction value corresponds to the behavior instruction value.

[0041] Since the above-mentioned sensor values ​​may show characteristic changes, it is preferable to estimate the state of the motorcycle 10 using time-series data of the sensor values ​​in order to detect them. In other words, in order to estimate the state of the motorcycle 10 at time T1, it is preferable to estimate the state of the motorcycle 10 using the sensor values ​​before time T1 and the sensor values ​​after time T1. In other words, it is preferable to estimate the state of the motorcycle 10 using the change over time of the sensor values. However, it is not essential to use the time-series data of the sensor values ​​or the change over time of the sensor values. For example, in a jump state, a combination of the output command value and the engine rotation speed or a combination of other sensor values ​​may show a characteristic value, and when the characteristic combination occurs, it is estimated that the motorcycle is in a jump state, and other states are estimated as a ground contact state.

[0042] From the above, it has been shown that there is a sufficient correlation between the sensor values ​​of this embodiment and the state of the motorcycle 10, and that the state of the motorcycle 10 can be estimated based on the sensor values.

[0043] Next, a specific estimation process will be described. First, a process for creating an estimation model will be described.

[0044] In this embodiment, the processing device 32 of the behavior estimation device 30 creates an estimation model by performing the processing of the flowchart in Fig. 3. The processing in Fig. 3 may be performed spontaneously by the processing device 32, or may be performed by the processing device 32 upon receiving a command from a human. At least a part of the processing for creating the estimation model may be performed by using another computer.

[0045] First, time series data of sensor values ​​for learning is prepared. Specifically, a rider rides the motorcycle 10 and performs a test run on an unpaved course, and obtains time series data of sensor values ​​detected by the sensor during the test run of the motorcycle 10. The processing device 32 receives the time series data of the sensor values ​​(S101). Furthermore, if the sampling rate differs for each sensor value, the processing device 32 performs down-sampling so as to match the lowest sampling rate or a predetermined sampling rate (S101).

[0046] Next, the processing device 32 creates time-series data of the correct answer data (S102). The correct answer data is data indicating which of the four states the motorcycle 10 is in during the test run. It is called correct answer data because it is data indicating the correct answer when learning. There are various methods for creating the correct answer data. For example, the correct answer data may be created by a person determining which state the motorcycle 10 is in based on a video obtained by shooting the test run of the motorcycle 10 with a camera and inputting the determined state into the processing device 32. The processing device 32 may also determine and create the correct answer data based on the detection results of the stroke sensor 26 or another sensor. In this case, the other sensor is a sensor that detects the behavior of the motorcycle 10, and is, for example, a sensor that is not provided on a mass-produced motorcycle 10 or a sensor that is provided outside the motorcycle 10.

[0047] Next, the processing device 32 creates a feature from the time series data of the sensor value (S103). By using a known library, the feature can be created by converting the time series data of the sensor value. Specifically, libraries such as tsfresh and hctsa can be used. For example, the maximum value, minimum value, or average value of the sensor value in a predetermined time length can be created as a feature, or the number of times an extreme value is taken in a predetermined time length can be created as a feature. Note that the process of creating the feature may use multiple libraries, or may not use a library. Also, the sensor value may be treated as a feature without performing the process of creating the feature. In other words, the process of step S103 is not essential and can be omitted.

[0048] Next, the processing device 32 reduces the features based on the contribution to the estimation to determine the features for learning (S104). A large number of features are created in the process of step S103, and features with low priority are reduced in the process of step S104, making it possible to extract useful features. Specifically, the features are reduced using SHAP. SHAP is a known framework for interpreting predicted values ​​from complex models, and calculates and compares SHAP values ​​indicating the contribution to the estimation. This makes it possible to reduce features with low contribution to the estimation. Note that the features may be reduced using a method other than SHAP. Note that the process of step S104 is not essential and can be omitted.

[0049] As a result, learning data consisting of learning features and answer data can be prepared. The learning data corresponds to teacher data in supervised learning. In this embodiment, a plurality of courses are test driven to prepare time-series data of sensor values. Therefore, the processing device 32 performs the above process for each course to prepare learning data (S105). Note that learning may be performed using only learning data for one course.

[0050] Here, during the test run, the time that the motorcycle 10 is in a jump state is often significantly shorter than the time that the motorcycle 10 is in a ground contact state. Therefore, there is a bias in the learning data. Therefore, in order to reduce the influence of the bias, the processing device 32 weights the learning data in the jump state greater than the weight of the learning data in the ground contact state (S106). Alternatively, the processing device 32 may reduce the influence of the bias by thinning out the learning data in the ground contact state. In other words, the processing device 32 performs processing such that the learning data in the jump state has a higher learning priority than the learning data in the ground contact state.

[0051] Next, the processing device 32 performs ensemble learning using the learning data for each course to create an estimation model including a plurality of learned models (S107). That is, one learned model is created using the learning data obtained from one course. As the machine learning, a GBDT-based machine learning, for example, Catboost, can be used, but other methods may also be used. By using the GBDT-based machine learning, high classification accuracy can be expected, but even if other methods are used, a learned model with sufficient estimation accuracy can be created. In addition, in this embodiment, the input of the learned model is time-series data of the feature amount, and the output of the learned model is the confidence level for each state of the motorcycle 10. The confidence level is the degree to which it is estimated that the motorcycle belongs to each state, and the higher the confidence level, the higher the probability that the motorcycle belongs to that state. Note that the ensemble learning is one example. Therefore, the processing device 32 may create an estimation model composed of one learned model.

[0052] Next, a process for estimating the state of the motorcycle 10 using the estimation model will be described.

[0053] In this embodiment, the processing device 32 of the behavior estimation device 30 estimates the state of the motorcycle 10 by performing the process of the flowchart in Fig. 4. Note that Fig. 5 is a diagram showing this process from the viewpoint of data flow.

[0054] When making the estimation, first, time series data of the sensor value to be estimated is prepared. Specifically, the rider rides the motorcycle 10 on a course of an unpaved road, and time series data of the sensor value detected by the sensor while the motorcycle 10 is traveling is acquired. The course traveled here may be the same as the course when the estimation model was created, or may be different. This is because, in this embodiment, the state of the motorcycle 10 is estimated based on the sensor value, not the position or shape of the course, and therefore the state of the motorcycle 10 can be estimated regardless of the position or shape of the course. The processing device 32 receives the time series data of the sensor value to be estimated (S201). Also, as in the case of creating the estimation model, when the sampling rate differs for each sensor value, the processing device 32 performs downsampling so as to match the lowest sampling rate or a predetermined sampling rate (S201).

[0055] Next, the processing device 32 creates time-series data of features based on the time-series data of the sensor values ​​(S202). The features created here are the same as the features after reduction in step S104, in other words, the features used in machine learning. Note that, when creating an estimation model without creating features, the process of step S202 can be omitted.

[0056] Next, the processing device 32 inputs the time series data of the feature amounts to a plurality of trained models and acquires time series data of the certainty of each state (S203). For example, it is assumed that the trained models include a first trained model and a second trained model. The first trained model outputs time series data of the certainty of the state of the motorcycle 10 in association with the acquisition time of the sensor value. For example, as shown in FIG. 5, the certainty of each of the normal ground contact state, the ground contact state before the jump, the jump state, and the landing state is output according to time.

[0057] Next, the processing device 32 estimates the state at each time based on the average value of the confidence levels of the multiple trained models, and outputs time-series data of the state estimation results (S204). In this embodiment, multiple trained models output their respective confidence levels by ensemble learning. Therefore, for example, even if the estimation accuracy of one trained model is low, it can be compensated for by other trained models. In this embodiment, the average value of the confidence levels is calculated, and the state with the highest average value is set as the estimation result. Alternatively, without using the average value, the state with the highest confidence level may be identified for each trained model, and the estimation result may be determined by majority vote. Also, if ensemble learning is not performed, the process of step S204 is not necessary, and the state with the highest confidence level output by one trained model is set as the estimation result.

[0058] Next, the processing device 32 superimposes the estimation result on a map based on the time series data of the estimation result and the time series data of the location information, and displays it on the display device 34 (S205). Specifically, by making the time series data of the estimation result correspond to the time series data of the location information using time, data indicating the correspondence between the estimation result and the location information can be created. Also, it is assumed that the behavior estimation device 30 stores the location information of the map. As a result of the above, the processing device 32 can superimpose the location information of the estimation result on the map, as shown in FIG. 6. Note that, although the processing device 32 displays it on the display device 34 in this embodiment, it may also be displayed on an external device via the transmission device 35.

[0059] (Feature 1) As described above, the behavior estimation device 30 of this embodiment analyzes the behavior of the motorcycle 10. The motorcycle 10 includes the body 11, the engine 17, and the rear wheel 13 that runs using the driving force generated by the engine 17. The behavior estimation device 30 includes a receiving device 31 and a processing device 32. The receiving device 31 receives an output command value for the engine 17 by the rider and the rotation speed of the engine 17, which are detected by a sensor provided on the motorcycle 10 while the motorcycle 10 is running. The processing device 32 estimates, based on the output command value and the rotation speed received by the receiving device 31, whether the front wheel 12 and the rear wheel 13 are in a ground contact state in which they are in contact with the road surface, or in a jump state in which the front wheel 12 and the rear wheel 13 are off the road surface.

[0060] Since it is possible to estimate whether the motorcycle 10 is in a ground contact state or a jump state, it is possible to grasp detailed behavior of the motorcycle 10 while it is traveling. In particular, since a typical motorcycle 10 has sensors that detect the output command value and the rotation speed, it is possible to grasp the behavior of the motorcycle 10 in principle without providing a dedicated sensor.

[0061] (Feature 2) In the behavior estimation device 30 of this embodiment, the processing device 32 receives time series data of the output command value and the rotational speed from the receiving device 31, and outputs time series data of the estimation result indicating whether the motorcycle 10 is in a ground contact state or a jump state.

[0062] Since the behavior of the motorcycle 10 can be estimated using not only the output command value and rotation speed at a specific point in time but also the time-dependent changes in the output command value and rotation speed, the estimation accuracy can be improved. Also, the time-dependent changes in the behavior of the motorcycle 10 can be grasped.

[0063] (Feature 3) In the behavior estimation device 30 of this embodiment, the ground contact state includes a pre-jump ground contact state, a landing state, and a normal ground contact state. The pre-jump ground contact state is a state in which the front wheel 12 and the rear wheel 13 are on the ground, and indicates a state in the middle of transitioning to a jump state. The landing state is a state in which the front wheel 12 and the rear wheel 13 are on the ground, and indicates a state when landing from a jump state. The normal ground contact state is a state in which the front wheel 12 and the rear wheel 13 are on the ground, and is a state other than the pre-jump ground contact state and the landing state. The processing device 32 estimates which of the pre-jump ground contact state, the landing state, the normal ground contact state, and the jump state it corresponds to.

[0064] The behavior of the motorcycle 10 can be understood in more detail.

[0065] (Feature 4) In the behavior estimation device 30 of this embodiment, the processing device 32 further estimates whether the motorcycle 10 is in a ground contact state or a jump state, based on the vehicle speed and longitudinal acceleration detected by sensors provided on the motorcycle 10 while the motorcycle 10 is traveling.

[0066] This increases the accuracy of estimating the behavior of the motorcycle 10.

[0067] (Feature 5) In the behavior estimation device 30 of this embodiment, the processing device 32 calculates the vehicle speed and longitudinal acceleration of the motorcycle 10 based on time-series data of the position information of the motorcycle 10 .

[0068] By simply using the time series data of the position information, the vehicle speed and longitudinal acceleration during a jump can be easily calculated.

[0069] (Feature 6) In the behavior estimation device 30 of this embodiment, the receiving device 31 receives time-series data of position information of the motorcycle 10 from the GNSS sensor 22 serving as a sensor.

[0070] The GNSS sensor 22 is a commonly used sensor, and this type of sensor can be used to calculate position information, vehicle speed, and longitudinal acceleration.

[0071] (Feature 7) In the behavior estimation device 30 of this embodiment, the processing device 32 displays, on the display device 34, information in which the estimation result is superimposed on a map, based on the position information of the motorcycle 10 and the estimation result as to whether the motorcycle 10 is in a ground contact state or a jump state.

[0072] The behavior of the motorcycle 10 according to the position can be understood at a glance.

[0073] (Feature 8) In the behavior estimation device 30 of this embodiment, the processing device 32 uses an estimation model that receives an output command value and a rotation speed as input and outputs an estimation result indicating whether the state is in a ground contact state or a jump state. The estimation model is created by learning a feature value based on the output command value, a feature value based on the rotation speed, and ground truth data indicating whether the state is in a ground contact state or a jump state for a combination of the feature values ​​as learning data.

[0074] Since the estimation result is obtained based on the tendency indicating how the output command value and the rotation speed affect the behavior of the motorcycle 10, it is possible to obtain a highly accurate estimation result.

[0075] (Feature 9) In the behavior estimation device 30 of this embodiment, the processing device 32 receives time-series data of the output command value and the rotation speed from the receiving device 31. The processing device 32 performs processing to match the sampling rates of the output command value and the rotation speed by down-sampling at least one of the time-series data of the output command value and the rotation speed, and then inputs the data to the estimation model.

[0076] Downsampling allows the data to be handled at the same sampling rate.

[0077] (Feature 10) In the behavior estimation device 30 of this embodiment, the learning data includes data detected when the motorcycle 10 travels along a plurality of travel routes.

[0078] By using data from multiple driving routes, a highly accurate estimation model can be created.

[0079] (Feature 11) In the behavior estimation device 30 of this embodiment, the motorcycle 10 serving as a saddle-ride type vehicle travels on unpaved ground.

[0080] The motorcycle 10 traveling on unpaved ground frequently jumps, and therefore the effects of the present invention can be effectively utilized.

[0081] The above-mentioned features 1 to 11 can be appropriately combined as long as no contradiction occurs. For example, feature N (N=1, 2, . . . , 11) can be appropriately combined with at least one of features 1 to N-1.

[0082] Although the preferred embodiment of the present application has been described above, the above configuration can be modified, for example, as follows.

[0083] The estimation using the estimation model is not essential, and the state of the motorcycle 10 may be estimated without using the estimation model. In this case, the processing device 32 estimates the state of the motorcycle 10 based on the estimation rule. The estimation rule is a rule-based condition for estimating the state of the motorcycle 10. As described above, the sensor values, particularly the output command value and the engine speed, show characteristic values ​​and changes in the ground contact state and the jump state. Therefore, these values ​​and changes can be captured using the threshold value of the sensor value and the threshold value of the differential value of the sensor value. This makes it possible to estimate the state of the motorcycle 10 without using the estimation model. Furthermore, since the output command value shows characteristic values ​​and changes in the pre-jump state and the landing state, it is also possible to estimate which of the four states the motorcycle 10 is in by creating an estimation rule using a similar method.

[0084] The flowcharts shown in the above embodiments are merely examples, and as described above, some processes may be omitted, the contents of some processes may be changed, or new processes may be added.

[0085] The functions of the elements disclosed herein can be performed using circuits or processing circuits, including general purpose processors, special purpose processors, integrated circuits, Application Specific Integrated Circuits (ASICs), conventional circuits, and / or combinations thereof, configured or programmed to perform the disclosed functions. Processors are considered processing circuits or circuits because they include transistors and other circuits. In this disclosure, a circuit, unit, or means is hardware that performs the recited functions or hardware that is programmed to perform the recited functions. The hardware may be hardware disclosed herein or other known hardware that is programmed or configured to perform the recited functions. Where the hardware is a processor, which is considered a type of circuit, the circuit, means, or unit is a combination of hardware and software, and the software is used to configure the hardware and / or the processor. [Explanation of symbols]

[0086] 10 Motorcycles (saddle-type vehicles) 30 Behavior Estimation Device 31 Receiving device 32 Processing equipment

Claims

1. A behavior estimation device for analyzing behavior of a saddle-type vehicle including a vehicle body, a drive source, and a ground-mounted traveling body that travels using a driving force generated by the drive source, comprising: a receiving device that receives an output command value of a drive source input by a driver and a rotation speed of the drive source, the output command value being detected by a sensor provided in the saddle riding type vehicle while the saddle riding type vehicle is traveling; a processing device that estimates whether the ground-contacting traveling body is in a ground contact state where the ground-contacting traveling body is in contact with a road surface, or in a jump state where the ground-contacting traveling body is separated from the road surface, based on the output indication value and the rotation speed received by the receiving device; A behavior estimation device comprising:

2. The behavior estimation device according to claim 1 , The processing device receives time series data of the output instruction value and the rotational speed from the receiving device, and outputs time series data of the estimation result indicating whether the saddle-type vehicle is in the ground contact state or the jump state.

3. The behavior estimation device according to claim 1 or 2, The grounded state includes: a pre-jump ground contact state in which the ground-contact running body is in a ground contact state and is in the middle of transitioning to the jump state; a landing state in which the ground-contact running body is in a state where it is touching the ground and is in a state where it is landing from the jump state; a normal ground contact state in which the ground-contact running body is in a state where the ground-contact running body is in a state other than the pre-jump ground contact state and the landing state; Includes: The processing device is a behavior estimation device that estimates which of the pre-jump ground contact state, the landing state, the normal ground contact state, and the jump state applies.

4. The behavior estimation device according to claim 1 or 2, The processing device further estimates whether the saddle-type vehicle is in the ground contact state or the jump state based on the vehicle speed and longitudinal acceleration detected by sensors provided on the saddle-type vehicle while the saddle-type vehicle is traveling.

5. The behavior estimation device according to claim 4, The processing device is a behavior estimating device that calculates a vehicle speed and a longitudinal acceleration of the saddle riding type vehicle based on time series data of position information of the saddle riding type vehicle.

6. The behavior estimation device according to claim 5, The receiving device receives time-series data of position information of the saddle-type vehicle from a GNSS sensor serving as the sensor, in the behavior estimation device.

7. The behavior estimation device according to claim 1 or 2, The processing device is a behavior estimation device that displays, on a display device, information in which the estimation result is superimposed on a map based on position information of the saddle-ride type vehicle and an estimation result as to whether the saddle-ride type vehicle is in the ground contact state or the jumping state.

8. The behavior estimation device according to claim 1 or 2, the processing device uses an estimation model that receives the output command value and the rotation speed as inputs and outputs an estimation result indicating whether the state is the ground contact state or the jump state, the estimation model is created by learning, as learning data, feature amounts based on the output instruction value, feature amounts based on the rotational speed, and ground truth data that indicates whether the state is in the ground contact state or the jump state for a combination of the feature amounts.

9. The behavior estimation device according to claim 8, the processing device receives the time series data of the output indication value and the rotation speed from the receiving device; The processing device downsamples the time series data of at least one of the output command value and the rotation speed to match sampling rates of the output command value and the rotation speed, and then inputs the data to the estimation model.

10. The behavior estimation device according to claim 8, The learning data includes data detected when the saddle-type vehicle travels a plurality of travel routes.

11. The behavior estimation device according to claim 1 or 2, The saddle-type vehicle is a motorcycle that travels on unpaved ground.

12. A behavior estimation method for analyzing behavior of a saddle-type vehicle including a vehicle body, a drive source, and a ground-mounted traveling body that is in contact with a road surface and travels by a driving force generated by the drive source, comprising: receiving a behavior instruction value from a driver detected by a sensor while the saddle-ride type vehicle is traveling; A behavior estimation method for estimating whether the ground-contacting traveling body is in a ground contact state in which it is in contact with a road surface, or in a jump state in which the ground-contacting traveling body is separated from the road surface, based on the received behavior indication value.

Images