Method for generating movement amount estimation model for work vehicle and device for generating movement amount estimation model for work vehicle

Abstract

This movement amount estimation model is a learning model, and is generated such that the model is trained to output the movement amount of a work vehicle upon inputting, as a training dataset, vehicle body data including at least a value related to the power of a travelling body of the work vehicle, and data indicating the movement amount of the work vehicle.

Description

Method for generating a work vehicle movement amount estimation model and device for generating a work vehicle movement amount estimation model

[0001] This application claims priority to Japanese Patent Application No. 2023-141725, filed on August 31, 2023, the contents of which are incorporated herein by reference.

[0002] Patent Document 1 discloses a work vehicle that causes the cutting edge of a blade to follow a design surface. According to the technology disclosed in Patent Document 1, a control device identifies the position of the work vehicle using a Global Navigation Satellite System (GNSS), and determines a target height for the blade based on the identified position.

[0003] International Publication No. 2015 / 083469

[0004] Control using GNSS, such as the technology disclosed in Patent Document 1, may not be able to receive GNSS radio waves depending on the environment of the construction site, making it impossible to determine the position of the work vehicle. Therefore, it is conceivable to determine the position of the work vehicle using autonomous navigation instead of GNSS.

[0005] A work vehicle may travel forward while performing work. The work vehicle travels while receiving a reaction force from the work object. The influence of this reaction force may cause an error between the position of the work vehicle obtained by autonomous navigation and the actual position. An example of an objective of the present disclosure is to provide a method for generating a movement amount estimation model of a work vehicle and a generation device for a movement amount estimation model of a work vehicle for estimating the movement amount of the work vehicle while suppressing the influence of the reaction force from the work object.

[0006] According to one aspect of the present disclosure, a method for generating a movement amount estimation model for a work vehicle includes a step of generating a movement amount estimation model that is a learning model that is trained to output the movement amount when vehicle body data including at least a value related to the power of the running body of the work vehicle and data indicating the movement amount of the work vehicle are used as a learning dataset.

[0007] According to the above aspect, as one example, the measurement system for a work vehicle can generate a movement amount estimation model for the work vehicle for estimating the movement amount of the work vehicle while suppressing the influence of reaction forces from the work object.

[0008] FIG. 1 is a side view of a work vehicle according to a first embodiment. FIG. 2 is a diagram showing the internal configuration of a driver's cab according to the first embodiment. FIG. 3 is a schematic diagram showing a power system of a work vehicle according to the first embodiment. FIG. 4 is a diagram showing the configuration of a measurement system and a control device of a work vehicle according to the first embodiment. FIG. 5 is a flowchart showing a position output method according to the first embodiment. FIG. 6 is a flowchart showing a method of relearning a motion state model according to the first embodiment. FIG. 7 is a diagram showing the configuration of a measurement system and a control device of a work vehicle according to a third embodiment.

[0009] <First embodiment> Hereinafter, an embodiment will be described in detail with reference to the drawings. Fig. 1 is a side view of a work vehicle according to a first embodiment. A work vehicle 100 according to the first embodiment is, for example, a bulldozer. The work vehicle 100 includes a vehicle body 110, a traveling device 120, a work implement 130, and a driver's cab 140.

[0010] The traveling device 120 is provided on the lower part of the vehicle body 110. The traveling device 120 has a pair of crawlers 121 and sprockets 122. The pair of crawlers 121 is provided on the left and right sides of the vehicle body 110. The crawlers 121 are rotated by the drive of the sprockets 122, causing the work vehicle 100 to travel.

[0011] The work machine 130 is used to excavate and transport excavation targets such as earth and sand. The work machine 130 includes a lift frame 131, a blade 132, a blade lift cylinder 133, a ripper 134, and a ripper lift cylinder 135. The blade 132 is disposed in front of the vehicle body 110. The ripper 134 is disposed in the rear of the vehicle body 110.

[0012] The base end of the lift frame 131 is attached to the side of the vehicle body 110 via a pin extending in the vehicle width direction. The tip end of the lift frame 131 is attached to the back surface of the blade 132 via a ball joint. This supports the blade 132 so that it can move up and down relative to the vehicle body 110. A cutting edge is provided at the lower end of the blade 132. The blade lift cylinder 133 is a hydraulic cylinder. The base end of the blade lift cylinder 133 is attached to the side of the vehicle body 110. The tip end of the blade lift cylinder 133 is attached to the lift frame 131. The blade lift cylinder 133 extends and contracts using hydraulic oil, thereby driving the lift frame 131 and the blade 132 in the raising and lowering directions.

[0013] A base end of the ripper lift cylinder 135 is attached to the vehicle body 110. A tip end of the ripper lift cylinder 135 is rotatably attached to the ripper 134. The ripper lift cylinder 135 is extended and contracted by hydraulic oil, thereby driving the ripper 134 in a raising or lowering direction.

[0014] The operator's cab 140 is a space where an operator sits and operates the work vehicle 100. The operator's cab 140 is provided on top of the vehicle body 110. Figure 2 is a diagram showing the internal configuration of the operator's cab 140 according to the first embodiment. Inside the operator's cab 140, there are provided a seat 141, a console 142, a blade operation lever 143, a ripper operation lever 144, a travel operation lever 145, a brake pedal 146, and a deceleration pedal 147. The brake pedal 146 and the deceleration pedal 147 may be configured as a single pedal.

[0015] An operation panel, instruments, and switches are attached to the console 142. The operator can visually check the status of the work vehicle 100 by checking the console 142. The blade operation lever 143 is operated to set the amount of movement for raising or lowering the blade 132. The blade operation lever 143 receives a lowering operation by tilting it forward, and a raising operation by tilting it backward. The blade operation lever 143 may also receive a tilt operation to rotate the blade 132 about an axis in the front-to-rear direction, and an angle operation to rotate the blade 132 about an axis in the up-to-down direction. The ripper operation lever 144 is operated to set the amount of movement for raising or lowering the ripper 134. The ripper operation lever 144 receives a lowering operation by tilting it forward, and a raising operation by tilting it backward. The travel operation lever 145 is operated to set the traveling direction of the travel device 120. The travel control lever 145 is tilted forward to accept a forward movement operation, and tilted backward to accept a reverse movement operation. The travel control lever 145 is also tilted left to accept a left turn operation, and tilted right to accept a right turn operation. The travel control lever 145 is also provided with a gear change button for inputting a speed range. The brake pedal 146 is operated to brake the travel device 120. The deceleration pedal 147 is operated to reduce the rotation speed of the travel device 120.

[0016] 3 is a schematic diagram showing the power system of the work vehicle according to Embodiment 1. The work vehicle 100 is equipped with an engine 210, a PTO (Power Take Off) 220, a pair of HSTs (Hydro Static Transmissions) 230, a hydraulic pump 250, and a proportional control valve 260.

[0017] The engine 210 is, for example, a diesel engine. The PTO 220 transmits a portion of the driving force of the engine 210 to the hydraulic pump 250. In other words, the PTO 220 distributes the driving force of the engine 210 to the HST 230 and the hydraulic pump 250. The HST 230 changes the speed of the driving force input to its input shaft and outputs it from its output shaft. The HST 230 includes a hydraulic pump driven by the rotation of the input shaft and a hydraulic motor that rotates the output shaft. The HST 230 controls the rotation speed of the output shaft by controlling the discharge flow rate of the hydraulic pump. The input shaft of the HST 230 is connected to the PTO 220, and the output shaft is connected to the sprocket 122. In other words, the HST 230 transmits the driving force of the engine 210 distributed by the PTO 220 to the sprocket 122. The output shafts of the pair of HSTs 230 are connected to the left sprocket 122 and the right sprocket 122, respectively. The hydraulic pump 250 is driven by driving force from the engine 210. The hydraulic oil discharged from the hydraulic pump 250 is supplied to the blade lift cylinder 133 via a proportional control valve 260. The proportional control valve 260 controls the flow rate of the hydraulic oil discharged from the hydraulic pump 250. In addition to the proportional control valve 260, the hydraulic pump 250 may supply hydraulic oil to other destinations such as a steering clutch (not shown).

[0018] <<Measurement System>> Figure 4 is a diagram showing the configuration of the measurement system and control device of work vehicle 100 according to the first embodiment. The measurement system of work vehicle 100 acquires vehicle body data that indicates the state of work vehicle 100. Work vehicle 100 is equipped with a rotation sensor 310, a dynamometer 320, an oil pressure sensor 330, an IMU (Inertial Measurement Unit) 340, a blade stroke sensor 350, a ripper stroke sensor 360, and a GNSS sensor 370.

[0019] A rotation sensor 310 is provided on each of the left and right sprockets 122. The rotation sensor 310 calculates the number of rotations of the sprocket 122.

[0020] The dynamometer 320 is provided on the engine 210. The dynamometer 320 measures the power generated by the engine 210.

[0021] An oil pressure sensor 330 is provided on each of the hydraulic motors of the left and right HSTs 230. The oil pressure sensor 330 measures the oil pressure on the outlet side of the hydraulic motor.

[0022] The IMU 340 is provided on the vehicle body 110. The IMU 340 measures the tilt angles of the vehicle body 110 in the roll and pitch directions, and the angular displacement in the yaw direction. The vehicle body coordinate system is, for example, an orthogonal coordinate system whose origin is the center of the traveling device 120 and is represented by an X axis extending in the longitudinal direction of the vehicle body, a Y axis extending in the lateral direction of the vehicle body, and a Z axis extending in the vertical direction of the vehicle body. The direction of rotation of the vehicle body 110 about the X axis is defined as the roll direction, the direction of rotation of the vehicle body 110 about the Y axis is defined as the pitch direction, and the direction of rotation of the vehicle body 110 about the Z axis is defined as the yaw direction.

[0023] The blade stroke sensor 350 is provided on the blade lift cylinder 133. The blade stroke sensor 350 measures the stroke amount of the blade lift cylinder 133. The stroke amount measured by the blade stroke sensor 350 can be converted into the position of the cutting edge of the blade 132 in the vehicle body coordinate system. Specifically, the rotation angle of the lift frame 131 is calculated based on the stroke amount of the blade lift cylinder 133. Because the shapes of the lift frame 131 and the blade 132 are known, the position of the cutting edge of the blade 132 can be identified from the rotation angle of the lift frame 131. Note that work vehicles 100 according to other embodiments may detect the rotation angle of the lift frame 131 using other sensors, such as an encoder.

[0024] The ripper stroke sensor 360 is provided on the ripper lift cylinder 135. The ripper stroke sensor 360 measures the stroke amount of the ripper lift cylinder 135. The stroke amount measured by the ripper stroke sensor 360 can be converted into the position of the ripper cutting edge in the vehicle body coordinate system. Note that work vehicles 100 according to other embodiments may detect the rotation angle of the ripper 134 using another sensor such as an encoder. Note that work vehicles 100 according to other embodiments do not need to be equipped with a sensor that detects the state of the ripper 134. In this case, for example, the control device 400 may determine whether the ripper 134 is raised or lowered based on the amount of operation of the ripper operating lever 144.

[0025] The GNSS sensor 370 is provided on the vehicle body 110. The GNSS sensor 370 measures the position and orientation of the vehicle body 110 in a global coordinate system based on signals from GNSS satellites. The GNSS sensor 370 may perform real-time kinematic (RTK) positioning. In this case, the GNSS sensor 370 communicates with a fixed station provided at the work site and corrects the measurement data. The GNSS sensor 370 outputs accuracy information indicating the positioning accuracy in addition to the position and orientation as measurement data. The accuracy information may indicate, for example, whether correction by the fixed station was successful or the number of satellites that were able to measure. The accuracy information may also be a numerical value indicating the positioning accuracy. The GNSS sensor 370 is an example of a position estimation unit that estimates the position and attitude of the work vehicle 100 based on measurement data.

[0026] <<Control Device>> The work vehicle 100 is equipped with a control device 400 for controlling the work vehicle 100. The control device 400 outputs control signals to the fuel injection device of the engine 210, the HST 230, and the proportional control valve 260 in accordance with the amount of operation of each operation device in the driver's cab 140 (console 142, blade operation lever 143, ripper operation lever 144, travel operation lever 145, brake pedal 146, and deceleration pedal 147). The control device 400 also measures the position of the work vehicle 100 based on measurement data from the measurement system, and displays this on the console 142. The control device 400 may autonomously control the power system based on the position of the work vehicle 100 measured based on the measurement data from the measurement system.

[0027] The control device 400 is a computer that includes a processor 410 , a main memory 430 , a storage 450 , and an interface 470 .

[0028] Storage 450 is a non-transitory tangible storage medium. Examples of storage 450 include a magnetic disk, a magneto-optical disk, and a semiconductor memory. Storage 450 may be an internal medium directly connected to the bus of control device 400, or an external medium connected to control device 400 via interface 470 or a communication line. Storage 450 stores a program for controlling work vehicle 100. Storage 450 also stores a motion state model 451, which is a machine learning model for calculating the velocity and angular velocity, as well as the tilt angle, of vehicle body 110 that have been learned in advance. The motion state model 451 will be described later.

[0029] In another embodiment, the control device 400 may include a custom large-scale integrated circuit (LSI) such as a programmable logic device (PLD) in addition to or instead of the above configuration. Examples of PLDs include programmable array logic (PAL), generic array logic (GAL), complex programmable logic device (CPLD), and field programmable gate array (FPGA). In this case, some or all of the functions implemented by the processor 410 may be implemented by the integrated circuit.

[0030] By executing a program, the processor 410 is provided with an operation amount acquisition unit 411, a measurement data acquisition unit 412, a work implement height calculation unit 413, a movement amount estimation unit 414, a position estimation unit 415, an output unit 416, a data set generation unit 417, and a learning unit 418. Measurement data from the measurement system is input to the processor 410 via an interface 470.

[0031] The operation amount acquisition unit 411 acquires operation amounts from the blade operation lever 143 , the ripper operation lever 144 , the travel operation lever 145 , the brake pedal 146 , and the deceleration pedal 147 .

[0032] The measurement data acquisition unit 412 acquires measurement data from the rotation sensor 310, the dynamometer 320, the oil pressure sensor 330, the IMU 340, the blade stroke sensor 350, the ripper stroke sensor 360, and the GNSS sensor 370. That is, the measurement data acquisition unit 412 acquires measurement data on the yaw angle of the vehicle body 110, the roll angle of the vehicle body 110, the pitch angle of the vehicle body 110, the rotation speeds of the left and right sprockets 122, the stroke amount of the blade lift cylinder 133, and the stroke amount of the ripper lift cylinder 135.

[0033] The work implement height calculation unit 413 calculates the height of the cutting edge of the blade 132 relative to the vehicle body, based on the measurement data of the stroke amount of the blade lift cylinder 133 acquired by the measurement data acquisition unit 412. In other words, the work implement height calculation unit 413 calculates the height of the cutting edge of the blade 132 in the vehicle body coordinate system from pre-stored dimensional information of the vehicle body 110 and the measurement data of the stroke amount of the blade lift cylinder 133. The work implement height calculation unit 413 also calculates the height of the cutting edge of the ripper 134 relative to the vehicle body, based on the measurement data of the stroke amount of the ripper lift cylinder 135 acquired by the measurement data acquisition unit 412.

[0034] The movement amount estimation unit 414 estimates the speed, angular velocity, and tilt angle of the work vehicle 100 by inputting the operation amount acquired by the measurement data acquisition unit 412 and the measurement data acquired by the measurement data acquisition unit 412 into the motion state model 451.

[0035] The position estimation unit 415 estimates the current position, orientation, and inclination angle of the work vehicle 100 based on the previous estimation results of the position, orientation, and inclination angle of the work vehicle 100, and the speed, angular velocity, and inclination angle of the work vehicle 100 estimated by the movement amount estimation unit 414. The position estimation unit 415 may be realized by, for example, a Bayes filter. In other words, the position estimation unit 415 may be a Bayes filter that estimates the position, orientation, and inclination angle at time t using the estimation results of the position and inclination angle at time t-1 as a "state (posterior belief)," the speed and angular velocity of the work vehicle 100 at time t estimated by the movement amount estimation unit 414 as a "control value," and the inclination angle of the work vehicle 100 at time t estimated by the movement amount estimation unit 414 as an "observation value."

[0036] The output unit 416 outputs the position, azimuth, and tilt angle estimated by the position estimation unit 415 to the console 142 .

[0037] The dataset generation unit 417 generates a learning dataset for the motion state model 451 using the data acquired by the operation amount acquisition unit 411, the measurement data acquisition unit 412, and the work machine height calculation unit 413. The learning unit 418 re-learns the motion state model 451 using the learning dataset generated by the dataset generation unit 417. Note that re-learning the motion state model 451 is one aspect of generating the motion state model 451. In other words, re-learning the motion state model 451 means generating a new motion state model 451 using the old motion state model 451 and the learning dataset.

[0038] <Motion State Model> The motion state model 451 according to the first embodiment receives input of the operation amount of the travel control lever 145, the rotation speeds of the left and right sprockets 122, measurement data from the IMU 340, measurement data of the pressure of the HST 230 (hydraulic motor), measurement data from the dynamometer 320, the speed gear, the height of the blade 132, and the height of the ripper 134, and outputs the velocity in the X-axis direction, the Y-axis direction, the Z-axis direction, the angular velocity around the X-axis (roll angular velocity), the angular velocity around the Y-axis (pitch angular velocity), the angular velocity around the Z-axis (swing angular velocity), the roll angle, and the pitch angle in the vehicle body coordinate system. The motion state model 451 is a machine learning model such as a neural network. The motion state model 451 may be configured as a CNN or an RNN. In particular, when the motion state model 451 is configured as a neural network that references information from past steps, such as an LSTM (Long Short Term Memory) or a Transformer, the motion state of the work vehicle 100 can be appropriately estimated in accordance with changes in the state over time. The Transformer is a neural network that can refer to information from past steps using Attention. The motion state model 451 according to other embodiments does not necessarily need to use all of the operation amount of the travel operation lever 145, the rotation speeds of the left and right sprockets 122, the measurement data of the IMU 340, the measurement data of the pressure of the HST 230 (hydraulic motor), the measurement data of the dynamometer 320, the speed gear, the height of the blade 132, and the height of the ripper 134 to estimate the motion state. As described below, the motion state model 451 may calculate the movement amount of the work vehicle 100 using at least one of the height of the blade 132, the height of the ripper 134, the operation amount of the travel operation lever 145, the speed gear, the output of the engine 210, and the pressure of the hydraulic motor of the HST 230.

[0039] The motion state model 451 has been trained using a training data set consisting of a combination of the following input samples and output samples. The input samples are the operation amount of the travel control lever 145, the rotation speeds of the left and right sprockets 122, the measurement data of the IMU 340, the pressure of the HST 230, the power of the engine 210, the speed gear, the height of the blade 132, and the height of the ripper 134. The output samples are the velocity in the X-axis direction, the Y-axis direction, the Z-axis direction, the angular velocity around the X-axis, the angular velocity around the Y-axis, the angular velocity around the Z-axis, the roll angle, and the pitch angle of the vehicle body coordinate system. The velocity of the work vehicle 100 related to the output sample may be a value calculated from the measurement data of the GNSS sensor 370. The angular velocity and attitude angle of the work vehicle 100 related to the output sample may be values ​​identified from the measurement data of the IMU 340.

[0040] The rotation speed of the sprocket 122 has a strong correlation with the movement distance of the work vehicle 100. In typical autonomous navigation, the movement distance is often calculated from the rotation speed. However, when slippage occurs in the crawler 121, the movement distance calculated from the rotation speed of the sprocket 122 may differ from the actual movement distance.

[0041] The measurement data of IMU 340 has a strong correlation with the angular velocity and tilt angle of work vehicle 100. In typical autonomous navigation, the direction of travel of the vehicle body is often calculated from the angular velocity and acceleration of IMU 340. However, when work vehicle 100 is traveling while performing work, vehicle body 110 vibrates, and there is a possibility that noise due to the vibrations may be added to the measurement data of IMU 340.

[0042] The height of the blade 132 and the height of the ripper 134 relate to the reaction force that the work vehicle 100 receives from the work object. The lower the blade 132 and the ripper 134 are, the more likely it is that the blade 132 and the ripper 134 will be in contact with the work object and performing work. When the work vehicle 100 is performing work, the work vehicle 100 will receive a reaction force from the work object. The height of the blade 132 and the height of the ripper 134 are examples of values ​​related to the state of the work implement 130.

[0043] The combination of the amount of operation of the travel control lever 145 and the rotation speed of the sprocket 122 relates to the reaction force that the work vehicle 100 receives from the work object. Furthermore, measurement data related to the power of the travel device 120, such as the output of the engine 210 and the pressure of the hydraulic motor of the HST 230, relates to the reaction force that the work vehicle 100 receives from the work object and the resulting slippage.

[0044] The speed gear is related to slippage that occurs due to a reaction force that the work vehicle 100 receives from a work object. The speed gear determines the maximum speed and torque of the work vehicle 100. Therefore, the lower the speed gear, the less likely slippage occurs, and the higher the speed gear, the more likely slippage occurs.

[0045] When work vehicle 100 receives a strong reaction force from the work object, the rotation speed of sprocket 122 decreases due to static friction force acting on traveling device 120. As the rotation speed of sprocket 122 decreases, the rotation speed of the hydraulic motor of HST 230, which rotates sprocket 122, decreases. As the rotation speed of the hydraulic motor of HST 230 decreases, the pressure between the hydraulic pump and hydraulic motor of HST 230 increases. As the pressure of HST 230 increases, the rotation speed of the input shaft of HST 230 decreases, and the rotation speed of engine 210 decreases.

[0046] Subsequently, when slippage occurs due to a reaction force from the work object, the frictional force acting on the traveling device 120 changes from static frictional force to kinetic frictional force, and the rotation speed of the hydraulic motor of the HST 230 and the sprocket 122 increases. As the rotation speed of the hydraulic motor of the HST 230 increases, the pressure between the hydraulic pump and hydraulic motor of the HST 230 decreases. The decrease in pressure of the HST 230 increases the rotation speed of the input shaft of the HST 230, and the rotation speed of the engine 210 increases. For this reason, if the movement amount of the work vehicle 100 were calculated directly from the rotation speed of the sprocket 122, an error in the movement amount would be large. On the other hand, the motion state model 451 calculates the movement amount of the work vehicle 100 using at least one of the height of the blade 132, the height of the ripper 134, the operation amount of the travel operation lever 145, the speed gear, the output of the engine 210, and the pressure of the hydraulic motor of the HST 230, thereby making it possible to calculate the movement amount while suppressing the effects of slippage.

[0047] Furthermore, when the work vehicle 100 receives a reaction force from the work object, the traveling gear 120 may lift due to the reaction force. For example, if the traveling gear 120 moves forward while the blade 132 receives a reaction force, the vehicle body 110 may move forward while the front of the traveling gear 120 lifts. In this case, the traveling direction of the work vehicle 100 does not match the inclination angle of the vehicle body 110 measured by the IMU 340, and therefore a deviation occurs in the direction of movement in typical autonomous navigation. On the other hand, the motion state model 451 calculates the amount of movement of the work vehicle 100 using at least one of the height of the blade 132, the height of the ripper 134, the operation amount of the travel operation lever 145, the output of the engine 210, and the pressure of the hydraulic motor of the HST 230, thereby making it possible to estimate the traveling direction of the work vehicle 100 even if the traveling gear 120 lifts. Note that the inclination angle output by the motion state model 451 may be the inclination angle of the traveling direction of the work vehicle 100, rather than the inclination angle of the vehicle body 110.

[0048] When the work vehicle 100 is shipped, a motion state model 451 that has been pre-trained in a factory or the like is already stored in the storage 450 .

[0049] <<Position Output Method>> Next, a position output method for work vehicle 100 according to the first embodiment will be described. Fig. 5 is a flowchart showing the position output method according to the first embodiment. The control device 400 repeatedly executes the flowchart shown in Fig. 5 while work vehicle 100 is operating.

[0050] The operation amount acquisition unit 411 acquires the operation amount and speed gear of the travel operation lever 145 (step S1). The measurement data acquisition unit 412 acquires measurement data from the rotation sensor 310, the dynamometer 320, the oil pressure sensor 330, the IMU 340, the blade stroke sensor 350, the ripper stroke sensor 360, and the GNSS sensor 370 (step S2). Next, the work implement height calculation unit 413 calculates the heights of the cutting edges of the blade 132 and the ripper 134 based on the measurement data of the stroke amount of the blade lift cylinder 133 and the stroke amount of the ripper lift cylinder 135 acquired by the measurement data acquisition unit 412 (step S3).

[0051] The output unit 416 reads accuracy information indicating the accuracy of positioning from the measurement data of the GNSS sensor 370 acquired by the measurement data acquisition unit 412, and determines whether or not the measurement data of the GNSS sensor 370 can be used as the position and orientation of the work vehicle 100 (step S4). For example, the output unit 416 determines that the measurement data of the GNSS sensor 370 cannot be used if the accuracy information indicates that correction by the fixed station has failed or if the number of satellites that could be measured is less than a predetermined number.

[0052] If the measurement data of the GNSS sensor 370 can be used (step S4: YES), the output unit 416 outputs the position and orientation of the work vehicle 100 indicated by the measurement data of the GNSS sensor 370 and the tilt angles (roll angle and pitch angle) indicated by the measurement data of the IMU 340 to the console 142 (step S5). Next, the dataset generation unit 417 calculates the speed and angular velocity of the work vehicle 100 based on the measurement data of the GNSS sensor 370 and the previous position and orientation of the work vehicle 100 recorded in the main memory 430 (step S6). In other words, the dataset generation unit 417 calculates the speed of the work vehicle 100 based on the difference between the position indicated by the measurement data of the GNSS sensor 370 and the previous position of the work vehicle 100. The dataset generation unit 417 also calculates the angular velocity of the work vehicle 100 based on the difference between the orientation indicated by the measurement data of the GNSS sensor 370 and the previous orientation of the work vehicle 100. The data set generator 417 records the operation amount and speed gear of the travel control lever 145 acquired in step S1, the rotation speeds of the left and right sprockets 122 acquired in step S2, the measurement data of the IMU 340, the measurement data of the GNSS sensor 370, the pressure measurement data of the HST 230 and the measurement data of the dynamometer 320, the height of the blade 132 and the height of the ripper 134 calculated in step S3, and the speed and angular velocity calculated in step S6 in the storage 450 as a training data set (step S7). Of the training data set, the operation amount of the travel control lever 145, the speed gear, the rotation speed of the sprocket 122, the tilt angle, the pressure of the HST 230, the power generated by the engine 210, the height of the blade 132, and the height of the ripper 134 are input samples. Of the training data set, the speed, angular velocity, and tilt angle are output samples. The output unit 416 records the position and orientation output in step S5 in the main memory 430 (step S8).

[0053] On the other hand, if the measurement data of the GNSS sensor 370 cannot be used (step S4: NO), the movement amount estimator 414 inputs the operation amount and speed stage of the travel operation lever 145 acquired in step S1, the rotation speeds of the left and right sprockets 122 acquired in step S2, the measurement data of the IMU 340, the pressure measurement data of the HST 230, and the measurement data of the dynamometer 320, as well as the height of the blade 132 and the height of the ripper 134 calculated in step S3, into the motion state model 451. As a result, the movement amount estimator 414 estimates the speed in the X-axis direction, the speed in the Y-axis direction, the speed in the Z-axis direction, the angular velocity about the X-axis, the angular velocity about the Y-axis, the angular velocity about the Z-axis, and the tilt angle (roll angle and pitch angle) of the work vehicle 100 (step S9).

[0054] The position estimation unit 415 estimates the current position, orientation and inclination angle of the work vehicle 100 based on the previous position, orientation and inclination angle of the work vehicle 100 recorded in the main memory 430 and the speed, angular velocity and inclination angle of the work vehicle 100 estimated in step S4 (step S10).

[0055] The output unit 416 outputs the position, orientation, and tilt angle estimated by the position estimation unit 415 to the console 142 (step S11). The output unit 416 also records the output position, orientation, and tilt angle in the main memory 430 (step S8).

[0056] <<Relearning Method>> Next, a method for relearning the motion state model 451 in the work vehicle 100 according to the first embodiment will be described. Fig. 6 is a flowchart showing the method for relearning the motion state model 451 according to the first embodiment. The console 142 displays a button for issuing an instruction to update the motion state model 451 on a menu that can be operated by the operator, for example. The control device 400 accepts an instruction to relearn the motion state model 451 via the console 142. When the control device 400 accepts the relearning instruction, it executes the flowchart shown in Fig. 6.

[0057] Upon receiving the re-learning instruction, the learning unit 418 copies the current kinetic state model 451 and records it in the storage 450 as the kinetic state model 451 to be learned (step S31). The learning unit 418 reads the learning dataset recorded in the storage 450 and performs relearning of the kinetic state model 451 to be learned using the learning dataset (step S32). For example, the learning unit 418 inputs input samples of the learning dataset to the kinetic state model 451 to be learned, and updates the parameters of the kinetic state model 451 to be learned by performing backpropagation based on the difference between the calculated value and the output sample of the learning dataset.

[0058] Next, the learning unit 418 calculates an evaluation function of the motion state model 451 to be learned (step S33). The evaluation function may be, for example, a mean absolute error or a mean square error. A smaller evaluation function indicates a higher reliability of the motion state model 451. The learning unit 418 determines whether the value of the evaluation function is smaller than a predetermined threshold (step S34). If the value of the evaluation function is equal to or greater than the threshold (step S34: NO), the process returns to step S32 and continues re-learning. If a new training dataset has been added to the storage 450 during this time, re-learning may be performed using the new training dataset.

[0059] If the value of the evaluation function is smaller than the threshold value (step S34: YES), the learning unit 418 overwrites and updates the current motion state model 451 stored in the storage 450 with the motion state model 451 to be learned (step S35). As a result, the re-learned motion state model 451 is used for subsequent estimation of the position, orientation, and tilt angle.

[0060] <<Actions and Effects>> In this way, the control device 400 according to the first embodiment uses vehicle body data including at least a value related to the power of the running body of the work vehicle 100 and data indicating the movement amount of the work vehicle 100 as a learning data set, and generates a motion state model 451, which is a learning model that is trained to output the movement amount when vehicle body data is input. Because the reaction force acting on the work vehicle 100 is reflected in the value related to the power of the running body of the work vehicle 100, the control device 400 can estimate the movement amount of the work vehicle by using the motion state model 451 while suppressing the influence of the reaction force from the work object. Note that the motion state model 451 according to the first embodiment outputs the tilt angle in addition to the movement amount of the work vehicle 100, but the motion state model 451 according to other embodiments does not need to output the tilt angle.

[0061] Furthermore, the control device 400 according to the first embodiment re-learns the motion state model 451, which was learned before shipping from the factory, using vehicle data acquired at the work site, thereby optimizing the motion state model 451 for each work site.

[0062] Second Embodiment A control device 400 according to a second embodiment outputs the uncertainty of each value in addition to the velocity, angular velocity, and tilt angle as estimation results of the motion state model 451. The motion state model 451 further outputs a variance-covariance matrix having dimensions equal to the number of items for estimating the uncertainty. The movement amount estimation unit 414 is trained to output the mean value and standard deviation of a Gaussian distribution as estimation results.

[0063] The control device 400 according to the first embodiment starts re-learning in response to an operation by an operator or the like as a trigger, whereas the control device 400 according to the second embodiment automatically starts re-learning when the uncertainty output as a variance-covariance matrix by the movement amount estimation unit 414 falls below a predetermined threshold.

[0064] Furthermore, the control device 400 according to the second embodiment may estimate the amount of movement using each of the current motion state model 451 and the re-learned motion state model 451, and when the uncertainty of the re-learned motion state model 451 exceeds the uncertainty of the current motion state model 451, may overwrite and update the current motion state model 451 stored in the storage 450 with the motion state model 451 to be learned. In this case, the uncertainty of the current motion state model 451 is an example of a threshold for overwriting and updating.

[0065] <Third Embodiment> Figure 7 is a diagram showing the configuration of a measurement system and control device of a work vehicle 100 according to a third embodiment. The work vehicle 100 according to the third embodiment further includes, as a measurement system, an imaging device 380 that captures an image of a work target of the work vehicle 100. Examples of the imaging device 380 include a camera, LiDAR (Light Detection and Ranging), and a laser scanner. The measurement data of the imaging device 380 is a two-dimensional image or a range image.

[0066] The movement amount estimator 414 according to the third embodiment inputs measurement data from the imaging device 380 into a motion state model 451 to estimate the position and orientation of the work vehicle 100. The motion state model 451 according to the third embodiment receives the measurement data from the imaging device 380 as input, in addition to the operation amount of the travel control lever 145, the rotation speeds of the left and right sprockets 122, measurement data from the IMU 340, measurement data of the pressure of the HST 230 (hydraulic motor), measurement data from the dynamometer 320, the speed gear, the height of the blade 132, and the height of the ripper 134. Instead of the measurement data itself from the imaging device 380, feature amount data extracted from the measurement data may be input to the motion state model 451. The feature amount of the measurement data from the imaging device 380 may be obtained, for example, from the intermediate layer of a convolutional autoencoder.

[0067] The measurement data of the imaging device 380 relates to slippage that occurs due to the reaction force that the work vehicle 100 receives from the work object. For example, the likelihood of the work vehicle 100 slipping varies depending on the work object and the properties of the ground. If the ground is muddy, the work vehicle 100 is more likely to slip. Therefore, the images and their feature quantities, which are the measurement data of the imaging device 380, may reflect the properties of the work object and the ground.

[0068] Furthermore, when the motion state model 451 is configured as a regression neural network, the measurement data of the image capture device 380 correlates with the movement amount and angular velocity of the work vehicle 100. As the work vehicle 100 travels, the scenery captured in the image, which is the measurement data of the image capture device 380, changes. The magnitude of the change in scenery over time differs depending on the movement amount and angular velocity of the work vehicle 100. Furthermore, when the image capture device 380 is a LiDAR, the measurement data indicates distance. The change in distance correlates with the movement amount of the work vehicle 100.

[0069] Furthermore, the dataset generation unit 417 records the operation amount of the travel operation lever 145, the rotation speeds of the left and right sprockets 122, the measurement data of the IMU 340, the measurement data of the pressure of the HST 230 (hydraulic motor), the measurement data of the dynamometer 320, the speed gear, the height of the blade 132, and the height of the ripper 134, as well as the measurement data of the imaging device 380 or its feature quantities, as input samples of the learning dataset in the storage 450. The learning unit 418 re-learns the motion state model 451 using the learning dataset generated by the dataset generation unit 417.

[0070] As a result, the movement amount estimating unit 414 according to the third embodiment can improve reliability by estimating the position and orientation of the work vehicle 100 using measurement data from the imaging device 380 .

[0071] Fourth Embodiment A work vehicle 100 according to a fourth embodiment stores, at the time of factory shipment, multiple motion state models 451 that have been trained using different training data sets. The movement amount estimator 414 of the control device 400 uses each of the multiple motion state models 451 to determine the speed, angular velocity, and tilt angle of the work vehicle 100, and the position estimator 415 calculates the position, orientation, and tilt angle using the obtained speed, angular velocity, and tilt angle. The movement amount estimator 414 then selects, from the multiple motion state models 451, one that has a small error between the estimated position and orientation and the position and orientation acquired by the GNSS sensor 370, as the model to be used to estimate the movement amount. In this case, the control device 400 does not necessarily need to perform re-learning.

[0072] Fifth Embodiment A work vehicle 100 according to a fifth embodiment transmits vehicle data to an external learning device and receives a re-trained motion state model 451 from the learning device. The learning device may be, for example, a cloud server.

[0073] The control device 400 according to the fifth embodiment does not need to include the learning unit 418. The dataset generation unit 417 according to the fifth embodiment transmits a learning dataset generated according to the flowchart shown in FIG. 5 to the learning device. The learning device collects learning datasets from multiple work vehicles 100. The learning device uses the collected learning datasets to learn a motion state model 451. When learning of the motion state model 451 is complete, the learning device transmits the learned motion state model 451 to each of the multiple work vehicles 100. The control devices 400 of the multiple work vehicles 100 receive the motion state model 451 from the learning device and record it in the storage 450. At this time, the control device 400 may overwrite and update the current motion state model 451 with the received motion state model 451.

[0074] <Other Embodiments> Although one embodiment has been described in detail above with reference to the drawings, the specific configuration is not limited to that described above, and various design modifications and the like are possible. That is, in other embodiments, the order of the above-described processes may be changed as appropriate. Furthermore, some processes may be executed in parallel. The control device 400 according to the above-described embodiment may be configured by a single computer, or the configuration of the control device 400 may be divided into multiple computers, and the multiple computers may function as the control device 400 by cooperating with each other. In this case, some of the computers configuring the control device 400 may be mounted inside the work vehicle 100, and other computers may be provided outside the work vehicle 100.

[0075] Although the work vehicle 100 according to the above-described embodiment includes the HST 230 as the power transmission device, this is not limited thereto. For example, the power transmission device of the work vehicle 100 according to another embodiment may be a hydromechanical continuously variable transmission (HMT) (which performs gear shift control by combining the HST 230 and a planetary gear mechanism). Furthermore, the power transmission device of the work vehicle 100 according to another embodiment may be a clutch. Even when the power transmission device is an HMT or a clutch, the output of the engine 210 is related to the reaction force that the work vehicle 100 receives from the work object and the resulting slippage. This is because, when the engine 210 and the power transmission device are connected, the reaction force affects the engine 210. Therefore, by inputting the measurement data of the dynamometer 320 into the motion state model 451, the control device 400 can estimate the movement amount of the work vehicle 100 while suppressing the influence of the reaction force from the work object. Furthermore, for example, the power transmission device of the work vehicle 100 according to another embodiment may be an electric mechanical transmission (EMT) having a generator and an electric motor instead of a hydraulic pump and a hydraulic motor. In this case, the control device 400 may estimate the amount of movement of the work vehicle 100 by inputting measurement data of the electric motor current value into the motion state model 451. The measurement data of the electric motor current value is an example of a value related to the power of the traveling body.

[0076] Furthermore, while the work vehicle 100 according to the embodiment described above is a bulldozer, the work vehicle 100 according to other embodiments is not limited to this. For example, the work vehicle 100 according to other embodiments may be a wheel loader, a motor grader, or the like. In other words, the running body of the work vehicle 100 according to other embodiments does not necessarily have to be a crawler, and may be, for example, tires or legs. Furthermore, the work vehicle 100 according to other embodiments may be an electric vehicle equipped with a battery and an electric motor instead of the engine 210.

[0077] Furthermore, although the control device 400 according to the embodiment described above displays the estimated position, orientation, and tilt angle of the work vehicle 100 on the console 142, this is not limited to this. For example, the control device 400 according to other embodiments may perform autonomous control that controls the drivetrain based on the estimated position, orientation, and tilt angle of the work vehicle 100.

[0078] Furthermore, although the control device 400 according to the embodiment described above uses GNSS measurement data to calculate the speed and angular velocity of the work vehicle 100 associated with the training data set, this is not limited to this. For example, the control device 400 according to another embodiment may calculate the speed and angular velocity of the work vehicle 100 associated with the training data set based on the position and orientation measured using a total station.

[0079] According to the present disclosure, as an example, a measurement system for a work vehicle can generate a movement amount estimation model for the work vehicle to estimate the movement amount of the work vehicle while suppressing the influence of reaction forces from the work object.

[0080] 100...Work vehicle 110...Vehicle body 120...Traveling device 121...Crawler 122...Sprocket 123...Travel motor 130...Work machine 131...Lift frame 132...Blade 133...Blade lift cylinder 134...Ripper 135...Ripper lift cylinder 140...Operator's cab 141...Seat 142...Console 143...Blade operation lever 144...Ripper operation lever 145...Travel operation lever 146...Brake pedal 147...Deceleration pedal 210...Engine 220...PTO 230...HST 250...Hydraulic pump 260...Proportional control valve 310...Rotation sensor 320...Dynamometer 330...Hydraulic pressure sensor 340...IMU 350...Blade stroke sensor 360...Ripper stroke sensor 370...GNSS sensor 380...Imaging device 400: Control device 410: Processor 411: Operation amount acquisition unit 412: Measurement data acquisition unit 413: Work machine height calculation unit 414: Movement amount estimation unit 415: Position estimation unit 416: Output unit 417: Data set generation unit 418: Learning unit 430: Main memory 450: Storage 451: Motion state model 470: Interface

Claims

1. A method for generating a movement amount estimation model for a work vehicle, comprising: a step of generating a movement amount estimation model which is a learning model trained to output the movement amount when vehicle body data including at least a value related to the power of the running body of the work vehicle and data indicating the movement amount of the work vehicle are used as a learning data set, and when the vehicle body data is input, the movement amount estimation model is generated.

2. The method for generating a work vehicle movement amount estimation model according to claim 1, wherein the data indicating the movement amount of the work vehicle is generated based on measurement data of the position of the work vehicle.

3. A method for generating a movement amount estimation model for a work vehicle as described in claim 2, comprising a step of estimating a position of the work vehicle using a Global Navigation Satellite System (GNSS), and in the step of generating the movement amount estimation model, re-training the movement amount estimation model using data indicating the movement amount of the work vehicle as a new learning data set, the data being generated based on the measurement data acquired when the measurement accuracy of the measurement data is higher than a predetermined threshold value.

4. A method for generating a movement amount estimation model for a work vehicle as described in claim 1, comprising: a step of estimating a value relating to the movement amount of the work vehicle based on the movement amount estimation model and vehicle body data including at least a value relating to the power of the running body of the work vehicle; and a step of re-learning the movement amount estimation model using a new learning dataset when, in the step of generating the movement amount estimation model, the accuracy of the estimation of the value relating to the movement amount is lower than a predetermined threshold.

5. The method for generating a movement amount estimation model for a work vehicle according to claim 1, wherein, in the step of generating the movement amount estimation model, re-learning of the movement amount estimation model is started when an update instruction is received.

6. A method for generating a movement amount estimation model for a work vehicle as described in claim 1, comprising: a step of estimating a value relating to the movement amount of the work vehicle based on the movement amount estimation model and vehicle body data including at least a value relating to the power of the running body of the work vehicle; and if, in the step of estimating the value relating to the movement amount of the work vehicle, the reliability of the re-learned movement amount estimation model exceeds a predetermined threshold, the value relating to the movement amount of the work vehicle is estimated using the re-learned movement amount estimation model.

7. A method for generating a movement amount estimation model for a work vehicle as described in claim 1, comprising the steps of: estimating the movement amount of the work vehicle using a plurality of movement amount estimation models; measuring the movement amount of the work vehicle; and selecting one movement amount estimation model from the plurality of movement amount estimation models based on a deviation between the estimated movement amount and the measured movement amount.

8. A method for generating a movement amount estimation model for a work vehicle as described in claim 1, wherein in the step of generating the movement amount estimation model, the movement amount estimation model is generated using the vehicle body data acquired from a plurality of work vehicles and data indicating the movement amount.

9. A learning device for a movement amount estimation model of a work vehicle, comprising: a learning unit that uses vehicle body data including at least a value related to the power of the running body of the work vehicle and data indicating the movement amount of the work vehicle as a learning data set, and generates a movement amount estimation model that is a learning model trained to output the movement amount when the vehicle body data is input.

Images