Assist Device

Abstract

This assisting device 10 comprises: a first mount tool 11 mounted on at least the waist section BW of a user U; an assisting arm 13 which is disposed along a thigh section BF of the user U and can rotate with respect to the first mount tool 11; a motor 40 which generates torque for rotating the assisting arm 13; a second mount tool 12 which is provided to the assisting arm 13 and mounted on the thigh section BF; an acceleration sensor 15 provided to the first mount tool 11; a rotation detector 41 which detects the rotation state of the assisting arm 13; and a control device 16 which controls the motor 40. The control device 16 comprises a processing unit 45 which performs: an estimation process 45b for acquiring, on the basis of an output of the acceleration sensor 15 and an output of the rotation detector 41, an estimated weight of luggage lifted by the user U; and a control process 45a for controlling the motor 40 on the basis of the estimated weight.

Description

[Technical Field]

[0001] The present invention relates to an assist device. [Background technology]

[0002] In recent years, various assist devices have been proposed that are worn on the body of a user to assist the user in performing tasks (see, for example, Patent Document 1). Such an assist device is configured to transmit the output of an actuator (motor) to the user's thighs via an arm, and assist the rotational movement of the thighs relative to the waist (flexion and extension of the hip joint).

[0003] The assist torque given to the user does not need to be very large if the weight of the load the user is lifting is relatively small, and it is preferable that it can be adjusted according to the weight of the load. For this reason, for example, the assist device disclosed in Patent Document 1 is configured to provide load sensors on the soles of the user's feet or hands, and to detect the weight of the luggage detected by the load sensors and adjust the assist torque according to the weight of the luggage. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Publication No. 2020-93375 Summary of the Invention [Problem to be solved by the invention]

[0005] In the above-mentioned conventional assisting devices, a dedicated sensor for detecting the weight of luggage needs to be installed on the soles of the feet or elsewhere, away from the attachment attached to the waist or thighs, which not only increases costs but also makes it cumbersome for the user to wear the assisting device.

[0006] Therefore, a method is desired that can control the assist torque according to the weight of the luggage without providing a dedicated sensor for detecting the weight of the luggage. [Means for solving the problem]

[0007] An assist device according to an embodiment includes a first attachment to be worn at least around the waist of a user, an arm disposed along the user's thigh and rotatable relative to the first attachment, a motor that generates torque to rotate the arm, a second attachment attached to the arm and worn on the thigh, an acceleration sensor attached to the first attachment, a rotation detector that detects the rotation state of the arm, and a control device that controls the motor. The control device includes a processing unit that executes an estimation process to calculate an estimated weight of a load to be lifted by the user based on the output of the acceleration sensor and the output of the rotation detector, and a control process to control the motor based on the estimated weight. [Effects of the Invention]

[0008] According to the present disclosure, it is possible to control the assist torque according to the weight of the luggage without providing a dedicated sensor for detecting the weight of the luggage. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is a rear view of the assist device worn by a user. [Figure 2] FIG. 2 is a side view of the assist device. [Figure 3] FIG. 3 is a diagram showing the configuration of the actuator. [Figure 4] FIG. 4 is a block diagram showing an example of the configuration of the control device. [Figure 5] Figure 5 shows the posture of the user, where (a) in Figure 5 is a side view of the user in an upright position, and (b) in Figure 5 is a side view of the user in a crouching position holding luggage on the luggage platform. [Figure 6]FIG. 6 is a flowchart showing the manner of motor control performed by the processing unit. [Figure 7] FIG. 7 is a flowchart illustrating an example of the estimation process. [Figure 8] FIG. 8 is a diagram showing the contents of the first time series data and the second time series data obtained from the discrete value data. [Figure 9] FIG. 9 is a diagram for explaining the actions of lifting and unloading luggage by a user. [Figure 10] FIG. 10 is a graph showing an example of the arm angular velocity when a user continuously performs actions of lifting and putting down a load. [Figure 11] FIG. 11 is a graph showing an example of the f1 score when the estimated weight is calculated by changing the threshold value and the time length of the past period. [Figure 12] FIG. 12 is a block diagram showing the configuration of a processing unit of an assist device according to the second embodiment. [Figure 13] FIG. 13 is a block diagram showing an example of the configuration of a control device according to the third embodiment. [Figure 14] FIG. 14 is a flowchart showing the estimation process performed by the processing unit of the third embodiment. [Figure 15] FIG. 15 is a diagram showing a manner in which the first down-sampled data and the second down-sampled data are generated from the first time-series data and the second time-series data. [Figure 16] FIG. 16 is a diagram for explaining downsampling performed by the processing unit of the third embodiment. [Figure 17] FIG. 17 is a diagram for explaining the movement pattern when a user lifts a load. [Figure 18] FIG. 18 is a diagram for explaining the movement pattern when a user lifts a load. [Figure 19] FIG. 19 is a flowchart showing the manner of motor control performed by the processing unit of the third embodiment. [Figure 20] FIG. 20 is a diagram illustrating a storage unit of a control device according to the fourth embodiment. [Figure 21] FIG. 21 is a diagram for explaining the movement pattern when a user lifts a load. [Figure 22] FIG. 22 is a flowchart showing the estimation process performed by the processing unit of the fourth embodiment. [Figure 23] FIG. 23 is a diagram showing a manner in which the first down-sampled data and the second down-sampled data are generated from the first time-series data and the second time-series data in the fourth embodiment. [Figure 24] FIG. 24 is a flowchart showing the process of determining movement and calculating estimated weight in the fourth embodiment. [Figure 25] FIG. 25 is a diagram showing an assisting device according to a modified example. DETAILED DESCRIPTION OF THE INVENTION

[0010] First, the contents of the embodiment will be listed and explained. [Outline of the embodiment] (1) An embodiment of an assist device includes a first attachment device attached to at least the waist of a user, an arm positioned along the user's thigh and rotatable relative to the first attachment device, a motor that generates torque to rotate the arm, a second attachment device attached to the arm and attached to the thigh, an acceleration sensor attached to the first attachment device, a rotation detector that detects the rotation state of the arm, and a control device that controls the motor, and the control device includes a processing unit that executes an estimation process that determines an estimated weight of the luggage being lifted by the user based on the output of the acceleration sensor and the output of the rotation detector, and a control process that controls the motor based on the estimated weight.

[0011] According to the above configuration, an estimation process is performed to determine the estimated weight of the luggage being lifted by the user based on the output of the rotation detector and the output of the acceleration sensor, so that the torque (assist torque) can be controlled according to the weight of the luggage without providing a dedicated sensor for detecting the weight of the luggage.

[0012] (2) In the above-described assist device, it is preferable that the estimation process includes an acquisition process for acquiring the output of the rotation detector and the output of the acceleration sensor over time, a determination process for determining whether the output of the rotation detector satisfies a predetermined condition, and, when it is determined that the output of the rotation detector satisfies the predetermined condition, a weight estimation process for calculating the estimated weight based on first time series data based on a plurality of outputs of the rotation detector acquired in a past period going back a predetermined time from the time when the output of the rotation detector that satisfied the predetermined condition was acquired, and second time series data based on a plurality of outputs of the acceleration sensor acquired in the past period.

[0013] There is a correlation between the output of the rotation detector and the output of the acceleration sensor immediately after the user starts lifting the luggage, and the weight of the luggage being lifted by the user. The output of the rotation detector indicates the rotation state of the thigh relative to the waist. Therefore, if a predetermined condition is set so that the determination process is performed immediately after the thigh starts to rotate, the first time series data and the second time series data can be obtained immediately after the user starts to lift the load. Therefore, by calculating the estimated weight using the first time-series data and the second time-series data immediately after the thigh starts to rotate, the estimation accuracy of the estimated weight can be improved.

[0014] (3) In the above assisting device, the predetermined condition may be that the angular velocity of the arm obtained from the output of the rotation detector when the arm rotates in a direction in which the user's hip joint extends is greater than a preset threshold value. In this case, the threshold value can be used to determine when the thigh starts to rotate, and the first time-series data and the second time-series data can be appropriately acquired immediately after the user starts to lift the load.

[0015] (4) In the above assisting device, it is preferable that the first time series data includes time series data of the angle of the arm relative to the first attachment and time series data of the angular velocity of the arm, and the second time series data includes time series data of the acceleration in the up-down direction in the acceleration sensor and time series data of the acceleration in the forward-backward direction of the user in the acceleration sensor. By using these time series data, it is possible to accurately estimate the weight.

[0016] (5) In the above assisting device, it is preferable that in the weight estimation process, the estimated weight is determined using a trained model that has learned the relationship between the first time series data and the second time series data and the weight of the luggage. In this case, the accuracy of the estimated weight can be further improved.

[0017] (6) In the above assisting device, the processing unit may further execute a process of receiving training data indicating the relationship between the first time series data and the second time series data and the weight of the luggage, and a process of relearning the trained model based on the training data. In this case, the trained model is retrained using the relationship between the first and second time series data and the weight of the luggage when the user actually lifts the luggage as training data, so the trained model can be optimized according to the user and the accuracy of the estimated weight can be improved.

[0018] (7) In the above assisting device, the estimation process may include an acquisition process for acquiring the output of the rotation detector and the output of the acceleration sensor over time; a muscle torque estimation process for calculating an estimated muscle torque for rotating the thigh exerted by the user's muscle force based on the inclination angle and angular velocity of the user's upper body obtained from the output of the acceleration sensor and the angle and angular velocity of the arm obtained from the output of the rotation detector; a determination process for determining whether the output of the rotation detector satisfies a predetermined condition; and a weight estimation process for, if it is determined that the output of the rotation detector satisfies the predetermined condition, calculating the estimated weight based on a plurality of estimated muscle torques obtained based on a plurality of rotation detector outputs and a plurality of acceleration sensor outputs acquired in a past period up to a predetermined time going back from the time when the output of the rotation detector that satisfied the predetermined condition was acquired. In this case, the estimated weight can be calculated based on a plurality of estimated muscle torques calculated in the past period, and the estimated weight can be calculated with high accuracy.

[0019] (8) In the above-described assisting device, it is preferable that the estimation process includes an acquisition process for acquiring the output of the rotation detector and the output of the acceleration sensor over time; a judgment process for determining whether the output of the rotation detector satisfies a predetermined condition; a movement judgment process for, when it is determined that the output of the rotation detector satisfies the predetermined condition, judging the movement of the user when lifting the luggage based on first time series data based on a plurality of outputs of the rotation detector acquired in a past period from the time when the output of the rotation detector that satisfied the predetermined condition was acquired to a time going back a predetermined time; and a weight estimation process for calculating the estimated weight based on the judgment result of the movement judgment, the first time series data, and the second time series data. In this case, the estimated weight is calculated based on the result of the user's motion determination, so that the accuracy of the estimated weight can be further improved.

[0020] (9) When the motion determination process determines which of a plurality of pre-set motion patterns the user's motion when lifting the luggage corresponds to, it is preferable that the weight estimation process obtains the estimated weight by selectively using a plurality of trained models that have been trained for each of the plurality of motion patterns to determine the relationship between the first time series data and the second time series data and the weight of the luggage. In this case, a trained model according to the user's movement pattern can be used, further improving the accuracy of the estimated weight.

[0021] (10) Preferably, the estimation process includes a generation process of generating first downsampled data by downsampling the first time series data and the second time series data at a first sampling rate and second downsampled data by downsampling the first time series data and the second time series data at a second sampling rate when it is determined that the output of the rotation detector satisfies the predetermined condition. In this case, the motion determination process performs the motion determination based on the first downsampled data, and the weight estimation process calculates the estimated weight based on a result of the motion determination and the second downsampled data. This makes it possible to reduce the amount of data to be processed in the motion determination process and weight estimation process, thereby reducing the processing load on the processing unit.

[0022] (11) When the processing unit further performs a correction process to correct the estimated weight obtained by the estimation process using a plurality of sigmoid functions corresponding to each of the plurality of operation patterns, it is preferable that the control process uses the corrected weight obtained by the correction process as the estimated weight. In this case, the estimated weight can be corrected nonlinearly, and the assist torque can be adjusted to an appropriate value for the user.

[0023] (12) In the above assist device, if it further includes another arm arranged along the other thigh of the user and rotatable relative to the first attachment, another motor that generates torque to rotate the other arm, another second attachment provided on the other arm and attached to the other thigh, and another rotation detector that detects the rotational state of the other arm, it is preferable that the acquisition process acquires the output of the other rotation detector over time in addition to the output of the rotation detector and the output of the acceleration sensor, and the first time series data includes time series data of a first value based on the output of the rotation detector, time series data of a second value based on the output of the other rotation detector, and time series data of an average value of the first value and the second value. In this case, by using each type of time-series data appropriately, it becomes possible to determine the behavior of the user U in more detail, and the number of behavior patterns at the time of determination can be increased.

[0024] [Details of the embodiment] Preferred embodiments will now be described with reference to the drawings. [Regarding the first embodiment] FIG. 1 is a rear view of the assisting device worn by a user, and FIG. 2 is a side view of the assisting device. The assisting device 10 according to the first embodiment is a device that assists the rotation of the thighs BF of the user U relative to the waist BW (flexion and extension of the hip joint) when the user U, for example, lifts or lowers a load, or assists the rotation of the thighs BF of the user U relative to the waist BW when the subject walks. The movement that the assisting device 10 assists with respect to the body of the user U is referred to as an "assisting movement."

[0025] In each figure, the X, Y, and Z directions are perpendicular to each other, and the Z direction is parallel to the vertical direction. The Y direction is the front-to-back direction of the user U wearing the assist device 10 and standing upright. The X direction is the left-to-right direction of the user U wearing the assist device 10 and standing upright. Regarding the assisting operation, the assistance of the rotation of the thighs BF relative to the waist BW as described above is the same as the assistance of the rotation of the waist BW relative to the thighs BF. The assisting operation is an operation that applies a torque to the user U about a virtual axis Li that is parallel to the X direction and passes through the waist BW of the user U. This torque is also referred to as an "assist torque."

[0026] 1 includes a first attachment 11, a pair of second attachments 12, and a pair of assist arms 13. The first attachment 11 is attached to the upper body BU including the waist BW of the user U. The pair of second attachments 12 are attached to the left and right thighs BF of the user U.

[0027] The first wearing tool 11 has a lumbar support part 21, a jacket part 22, a frame pipe 39, a backpack part 24, and a pair of rotation mechanisms 25. The lumbar support part 21 is worn around the lumbar region BW of the user U. The lumbar support part 21 includes a front belt 21a, a pair of rear belts 21b, and a pair of lumbar side pads 21c. The front belt 21a and the pair of rear belts 21b secure the pair of lumbar side pads 21c to both sides of the lumbar region BW of the user U.

[0028] The pair of pivoting mechanisms 25 are fixed to the pair of lumbar side pads 21c. One of the pair of pivoting mechanisms 25 is fixed to the right lumbar side pad 21c, and the other is fixed to the left lumbar side pad 21c. A pair of assist arms 13 are rotatably fixed to the pair of pivoting mechanisms 25. Each rotation mechanism 25 includes a case 25a fixed to the lumbar side pad 21c and a driven pulley (not shown) housed inside the case 25a and rotatable relative to the case 25a. The driven pulley has a shaft 25b that rotates integrally with the driven pulley. The shaft 25b protrudes from the surface of the case 25a opposite to the surface facing the user U. The assist arms 13 are fixed to the shaft portion 25b. Therefore, the driven pulley and the pair of assist arms 13 can rotate integrally. As a result, the pair of assist arms 13 can rotate relative to the first attachment 11.

[0029] The jacket portion 22 is worn around the shoulders BS and chest BB of the user U. The jacket portion 22 has a pair of shoulder belts 22a and a chest belt 22b. The pair of shoulder belts 22a are connected to the frame pipe 39. The frame pipe 39 is fixed to the back of the user U by the pair of shoulder belts 22a. The chest belt 22b connects the pair of shoulder belts 22a in front of the chest BB of the user U. The frame pipe 39 is more firmly fixed to the back of the user U by the chest belt 22b.

[0030] The frame pipe 39 is provided in a U-shape. The frame pipe 39 passes between the back of the user U and the backpack unit 24, and connects the rotation mechanisms 25 arranged on the left and right of the user U to each other. The backpack section 24 is fixed to the frame pipe 39 . The pair of rotation mechanisms 25 are fixed to both ends of the frame pipe 39 and also fixed to the pair of lumbar side pads 21c. At this time, the rotation center of the pair of assist arms 13 coincides with an imaginary axis Li in the left-right direction of the user U that passes through the lumbar region BW of the user U.

[0031] The pair of second wearing devices 12 are worn around the left and right thighs BF of the user U. The second wearing devices 12 are belt-like members made of resin, leather, cloth, or the like, and are wrapped around the thighs BF and attached and fixed to the thighs BF. The tip of an assist arm 13 is attached to each second attachment 12. The pair of assist arms 13 extend from the pair of pivot mechanisms 25 along both sides of the user U. Thus, the pair of assist arms 13 connect the pair of second attachments 12 and the pair of first attachments 11. The pair of second attachments 12 fix the tip ends of the pair of assist arms 13 to both thighs BF of the user U. As a result, the pair of assist arms 13 are arranged along both thighs BF of the user U and rotate together with both thighs BF.

[0032] The assist device 10 further includes a pair of actuators 14 , an acceleration sensor 15 , and a control device 16 . The pair of actuators 14 , the acceleration sensor 15 , and the control device 16 are housed in the backpack portion 24 . In addition to these, the backpack section 24 also accommodates a battery (not shown) for supplying the necessary power to each section.

[0033] Each actuator 14 has a function of generating an assist torque that rotates the assist arm 13 relative to the first attachment 11 . Fig. 3 is a diagram showing the configuration of the actuator 14. Note that Fig. 3 shows a simplified configuration of the actuator 14. The actuator 14 includes a motor 40 , a spiral spring 40 b , a rotation detector 41 , a reducer 42 , and a drive pulley 43 . The motor 40 generates torque that rotates the assist arm 13. The motor 40 outputs a rotational force from an output shaft 40a. The motor 40 is controlled by the control device 16. One end of the spiral spring 40b is connected to the output shaft 40a of the motor 40. The other end of the spiral spring 40b is connected to the input shaft 42a of the reducer 42. As a result, the spiral spring 40b transmits the rotational force of the motor 40 to the reducer 42.

[0034] The reducer 42 has a function of reducing the rotational force of the motor 40. The rotational force of the motor 40 is transmitted to the input shaft 42a via the spiral spring 40b. The reduced rotational force of the motor 40 is transmitted to the drive pulley 43. The rotation detector 41 has a function of detecting the rotation state of the input shaft 42a of the reducer 42, and detects the rotation state of the assist arm 13. The rotation detector 41 is a rotary encoder, a hall sensor, a resolver, etc. The output of the rotation detector 41 is given to the control device 16.

[0035] The drive pulley 43 is rotationally driven by the rotational force of the motor 40 which has been reduced in speed by the reducer 42 . A wire 44 is fixed to the drive pulley 43 . The wire 44 passes through a protective tube (not shown) extending along the frame pipe 39 and is connected to one of the pair of rotation mechanisms 25 (FIG. 1). One end of the wire 44 is fixed to the drive pulley 43. The other end of the wire 44 is fixed to the driven pulley of the rotation mechanism 25. Therefore, the rotational force of the drive pulley 43 is transmitted to the driven pulley of the rotation mechanism 25 via the wire 44 . The wires 44 are protected by a frame cover 23 arranged along the frame pipe 39 .

[0036] The driven pulley of the rotation mechanism 25 rotates due to the rotational force from the drive pulley 43. This causes the assist arm 13, which can rotate integrally with the driven pulley, to also rotate. In this way, the rotational force of the motor 40 is transmitted to the rotation mechanism 25 via the drive pulley 43, the wire 44, and the driven pulley, and is used as torque to rotate the assist arm 13.

[0037] 1, the pair of actuators 14 are arranged side by side on the left and right sides inside the backpack unit 24. The left actuator 14 of the pair of actuators 14 is connected to the left rotation mechanism 25 of the pair of rotation mechanisms 25, and transmits rotational force to the left rotation mechanism 25. The right actuator 14 of the pair of actuators 14 is connected to the right rotation mechanism 25 of the pair of rotation mechanisms 25 and transmits a rotational force to the right rotation mechanism 25.

[0038] Acceleration sensor 15 is a triaxial acceleration sensor that detects acceleration in each of three mutually perpendicular axes. The acceleration sensor is mounted, for example, on a board of control device 16 and fixed inside backpack unit 24. The output of acceleration sensor 15 is provided to control device 16.

[0039] The control device 16 is fixed and housed within the backpack portion 24. The control device 16 is configured with a computer or the like. FIG. 4 is a block diagram showing an example of the configuration of the control device 16. As shown in FIG. 4, the control device 16 includes a processing unit 45 including a processor and the like, and a storage unit 46 including a memory and a hard disk.

[0040] The storage unit 46 stores computer programs to be executed by the processing unit 45 and necessary information. The processing unit 45 executes a computer program stored in a computer-readable non-transitory recording medium such as the storage unit 46 to realize various processing functions of the processing unit 45. The storage unit 46 also stores a trained model 46a and discrete value data 46b, which will be described later.

[0041] The processing unit 45 executes the above-mentioned computer program to perform a control process 45a, an estimation process 45b, and a relearning process 45c. The estimation process 45b includes an acquisition process 45b1, a determination process 45b2, and a weight estimation process 45b3. These processes will be described later.

[0042] [About acquisition processing] By executing the acquisition process 45b1 (FIG. 4), the processing unit 45 acquires the output of the acceleration sensor 15 and the outputs of both the left and right rotation detectors 41 over time, and obtains values ​​and information required for various processes.

[0043] The processing unit 45 calculates the acceleration in the Y direction and the acceleration in the Z direction of the acceleration sensor 15 based on the output of the acceleration sensor 15 . As described above, the acceleration sensor 15 is a triaxial acceleration sensor that detects acceleration in each of three mutually perpendicular axes. Therefore, the processing unit 45 can determine the acceleration in the Y direction of the acceleration sensor 15 and the acceleration in the Z direction of the acceleration sensor 15 based on the output of the acceleration sensor 15. The acceleration sensor 15 is fixedly housed within the backpack 24. Therefore, the Y-direction acceleration of the acceleration sensor 15 and the Z-direction acceleration of the acceleration sensor 15 indicate the Y-direction acceleration and the Z-direction acceleration of the upper body BU of the user U. In the following description, the acceleration in the Y direction of the acceleration sensor 15 will be simply referred to as the Y-direction acceleration, and the acceleration in the Z direction of the acceleration sensor 15 will be simply referred to as the Z-direction acceleration.

[0044] Furthermore, the processing unit 45 can three-dimensionally determine the tilt angle of the acceleration sensor 15 relative to the vertical direction (the direction of gravitational acceleration) based on the output of the acceleration sensor. Therefore, the processing unit 45 can also determine the tilt angle of the upper body BU relative to the vertical direction.

[0045] Figure 5 shows the posture of the user, where (a) in Figure 5 is a side view of the user U in an upright position, and (b) in Figure 5 is a side view of the user U in a crouching position holding luggage N on the luggage stand D. FIG. 5(b) shows a state in which the user U is in a crouching position with his hip and knee joints bent.

[0046] 5(a) and 5(b), a virtual line Lu extending from the virtual axis Li and along the upper body BU is defined as a line diagram showing the tilt direction of the first wearing device 11 (upper body BU) with respect to the vertical direction. The virtual line Lu is defined so as to be parallel to the vertical direction when the user U is in an upright posture. The virtual line Lu is inclined in accordance with the tilt of the first wearing device 11 (upper body BU). 5(a) and 5(b), a virtual line Lf extending from the virtual axis Li and along the assist arm 13 is defined as a diagram showing the angular position of the assist arm 13 (thigh BF) relative to the first attachment 11. The virtual line Lf is defined so as to be parallel to the vertical direction when the user U is in an upright posture. The virtual line Lf rotates in response to the rotation of the assist arm 13 relative to the first attachment 11.

[0047] When the user U assumes a crouching posture as shown in (b) of Figure 5, the imaginary line Lu (upper body BU) tilts forward, and the imaginary line Lf (thighs BF) tilts backward.

[0048] In FIG. 5(b), the tilt angle α of the upper body BU with respect to the vertical direction is represented by the imaginary line Lu and the vertical line g that passes through the imaginary axis Li and is parallel to the vertical direction. As described above, the processing unit 45 can determine the tilt angle α based on the output of the acceleration sensor 15.

[0049] 5(a) and 5(b), the arm angle β is the relative angle of the assist arm 13 (thigh BF) with respect to the first attachment 11 (upper body BU), and is the angle between the imaginary line Lu and the imaginary line Lf. In other words, the arm angle β indicates the hip joint angle of the user U.

[0050] The processing unit 45 determines the arm angle β and the arm angular velocity ω based on the output of the rotation detector 41. The arm angular velocity ω is the angular velocity of the assist arm 13 when the assist arm 13 rotates relative to the first attachment 11. The processing unit 45 can determine the angle and angular velocity of the input shaft 42a, which are the rotational state of the reducer 42, based on the output of the rotation detector 41. The angle of the input shaft 42a indicates the cumulative angle of the input shaft 42a.

[0051] Here, the rotational force of the motor 40 is transmitted to the assist arm 13 via the reducer 42, the drive pulley 43, the wire 44, and the rotating mechanism 25. Therefore, the rotation of the motor 40 and the reducer 42 corresponds to the rotation of the assist arm 13 at a constant ratio. That is, the angle and angular velocity of the input shaft 42a can be converted into the relative angle and angular velocity of the assist arm 13 with respect to the first attachment 11. As a result, the processing unit 45 can determine the relative angle and angular velocity of the assist arm 13 (thigh BF) with respect to the first attachment 11 based on the output of the rotation detector 41. In addition, the output of the rotation detector 41 includes not only the output when the motor 40 and the reducer 42 are outputting rotational force, but also the output when the input shaft 42a of the reducer 42 rotates due to the rotation of the assist arm 13 as the assist arm 13 rotates relative to the first mounting fixture 11.

[0052] For example, the angular position of the assist arm 13 relative to the first attachment 11 when the user U is in an upright position is considered to be a state in which the hip joint is extended, and this angular position is set as the reference angular position. In FIG. 5(a), the user U is in an upright position, so the virtual line Lf0 indicating the reference angle position and the virtual line Lf overlap each other. In addition, the arm angle β when the angular position of the assist arm 13 (virtual line Lf) is at the reference angular position is set to 180 degrees.

[0053] 5(b), the assist arm 13 is rotated toward the upper body BU, and the imaginary line Lf is rotated toward the upper body BU more than the imaginary line Lf0. The processing unit 45 can obtain the angle γ between the virtual line Lf0, which is the reference angle position, and the virtual line Lf, and subtract the angle γ from 180 degrees to obtain the arm angle β. In this way, the processing unit 45 can determine the arm angle β based on the virtual line Lf0 indicating the reference angle position.

[0054] The arm angular velocity ω when the thigh BF rotates relative to the upper body is obtained by acquiring the arm angle β over time and finding it based on the increment of the arm angle β per unit time.

[0055] In this way, the processing unit 45 obtains the Y-direction acceleration and the Z-direction acceleration based on the acquired output of the acceleration sensor 15. Furthermore, the processing unit 45 determines the arm angle β and the arm angular velocity ω based on the acquired output of the rotation detector 41. The processing unit 45 determines the arm angle β and the arm angular velocity ω for each of the left and right assist arms 13.

[0056] The processing unit 45 acquires the output of the acceleration sensor 15 and the output of the rotation detector 41 over time at predetermined sampling intervals (e.g., every 10 milliseconds) and stores them in the memory unit 46 as discrete value data 46b including each value that is continuous over time.

[0057] [Operation of the assist device] FIG. 6 is a flowchart showing the manner in which the processing unit 45 controls the motor 40. The processing unit 45 first performs the estimation process 45b (FIG. 4) (step S1 in FIG. 6). The estimation process 45b is a process for calculating an estimated weight of the baggage that the user U is carrying. After performing the estimation process 45b, the processing unit 45 performs the control process 45a (FIG. 4) (step S2 in FIG. 6). The control process 45a is a process for controlling the motor 40 based on the estimated weight to generate an assist torque. When the processing unit 45 obtains the estimated weight through the estimation process 45b, it generates an assist torque. In this way, before controlling the motor 40, the processing unit 45 first performs the estimation process 45b.

[0058] FIG. 7 is a flowchart illustrating an example of the estimation process. The processing unit 45 first executes the determination process 45b2 (FIG. 4) to determine whether or not the output of the rotation detector 41 satisfies a predetermined condition (steps S11 and S12 in FIG. 7). More specifically, the predetermined condition is that the arm angular velocity ω becomes greater than a preset threshold value Th when the assist arm 13 rotates in a direction in which the hip joint of the user U is extended from a flexed state. In other words, the predetermined condition is that the arm angular velocity ω becomes greater than a preset threshold value Th when the assist arm 13 rotates in a direction in which the arm angle β increases.

[0059] Therefore, the processing unit 45 determines whether the assist arm 13 is rotating in a direction that increases the arm angle β (step S11 in FIG. 7). When calculating the arm angular velocity ω, the processing unit 45 also obtains information indicating the rotation direction of the assist arm 13. The processing unit 45 determines whether the assist arm 13 is rotating in a direction in which the arm angle β increases, based on information indicating the rotation direction of the assist arm 13 acquired when calculating the arm angular velocity ω. The processing unit 45 repeats step S11 until it determines that the assist arm 13 is rotating in a direction in which the arm angle β increases.

[0060] When it is determined that the assist arm 13 is rotating in a direction in which the arm angle β increases, the processing unit 45 determines whether or not the most recently calculated arm angular velocity ω is greater than the threshold value Th (step S12 in FIG. 7). If it is determined that the arm angular velocity ω is not greater than the threshold value Th (is equal to or less than the threshold value Th), the processing unit 45 returns to step S11 again. In this way, by executing the judgment process 45b2, the processing unit 45 determines whether the arm angular velocity ω is greater than the threshold value Th when the assist arm 13 rotates in a direction in which the arm angle β increases (steps S11 and S12 in Figure 7).

[0061] If it is determined in step S12 that the arm angular velocity ω is greater than the threshold value Th, the processing unit 45 acquires the first time-series data and the second time-series data (step S13 in FIG. 7).

[0062] The first time series data and the second time series data are acquired from the discrete value data 46b stored in the storage unit 46. FIG. 8 is a diagram showing the contents of the first time series data and the second time series data acquired from the discrete value data 46b.

[0063] 8, the first time series data T1 is time series data based on the outputs of the plurality of rotation detectors 41 acquired during a past period P. The past period P is a period going back a predetermined time from the time when the arm angular velocity ω (output of the rotation detector 41) used for the determination in steps S11 and S12 was acquired. The first time series data T1 includes time series data T11 of the arm angle β and time series data T12 of the arm angular velocity ω.

[0064] The first time series data T1 is acquired from the discrete value data 46b stored in the storage unit 46. The discrete value data 46b includes discrete value data D11 of the arm angle β and discrete value data D12 of the arm angular velocity ω. The discrete value data D11 of the arm angle β includes a plurality of values ​​of the arm angle β that are continuous in time. Furthermore, the discrete value data D12 of the arm angular velocity ω includes a plurality of values ​​of the arm angular velocity ω that are continuous in time. The processing unit 45 acquires, from the discrete value data D11 of the arm angle β, a plurality of arm angles β acquired during the past period P as time series data T11 of the arm angle β. Therefore, the time series data T11 of the arm angle β includes the plurality of arm angles β acquired during the past period P as a plurality of elements. Furthermore, the processing unit 45 acquires, from the discrete value data D12 of the arm angular velocity ω, a plurality of arm angular velocities ω acquired during the past period P as time series data T12 of the arm angular velocity ω. Therefore, the time series data T12 of the arm angular velocity ω includes the plurality of arm angular velocities ω acquired during the past period P as a plurality of elements. As a result, the processing unit 45 acquires the first time series data T1. The processing unit 45 acquires, for each of the left and right assist arms 13, a plurality of arm angles β acquired during the past period P and a plurality of arm angular velocities ω acquired during the past period P. Therefore, the time series data T11 of the arm angle β and the time series data T12 of the arm angular velocity ω include data for each of the left and right assist arms 13.

[0065] The second time series data T2 is time series data based on the outputs of the multiple acceleration sensors 15 acquired during the past period P described above. The second time series data T2 includes time series data T21 of Y-direction acceleration and time series data T22 of Z-direction acceleration.

[0066] The second time series data T2 is also acquired from the discrete value data 46b stored in the storage unit 46. The discrete value data 46b includes discrete value data D21 of Y-direction acceleration and discrete value data D22 of Z-direction acceleration. The discrete value data D21 of Y-direction acceleration includes a plurality of temporally consecutive Y-direction acceleration values. The discrete value data D22 of Z-direction acceleration includes a plurality of temporally consecutive Z-direction acceleration values. The processing unit 45 acquires, from the discrete value data D21 of Y-direction acceleration, a plurality of Y-direction accelerations acquired during the past period P as time-series data T21 of Y-direction acceleration. Therefore, the time-series data T21 of Y-direction acceleration includes, as a plurality of elements, a plurality of Y-direction accelerations acquired during the past period P. Furthermore, the processing unit 45 acquires, from the discrete value data D22 of Z-direction acceleration, a plurality of Z-direction accelerations acquired during the past period P as time-series data T22 of Z-direction acceleration. Therefore, the time-series data T22 of Z-direction acceleration includes, as a plurality of elements, a plurality of Z-direction accelerations acquired during the past period P. As a result, the processing unit 45 acquires the second time series data T2.

[0067] Next, the processing unit 45 executes the weight estimation process 45b3 (Figure 4) to obtain an estimated weight of the luggage that the user U is trying to lift based on the first time series data T1 and the second time series data T2 (step S14 in Figure 7). The processing unit 45 uses a trained model 46a (FIG. 4) stored in the storage unit 46 to obtain the estimated weight. The trained model 46a is a model obtained by machine learning the relationship between the first time series data T1 and the second time series data T2 and the weight of the luggage. In other words, the trained model 46a is a model that uses the first time series data T1 and the second time series data T2 as explanatory variables and the weight of the luggage as a response variable.

[0068] There is a correlation between the output of the rotation detector 41 and the output of the acceleration sensor 15 immediately after the user U starts lifting the luggage, and the weight of the luggage being lifted by the user U. The trained model 46a is constructed based on this correlation.

[0069] Therefore, the processing unit 45 needs to acquire the first time series data T1 based on the output of the rotation detector 41 and the second time series data T2 based on the acceleration sensor 15 immediately after the user U starts lifting the baggage.

[0070] The output of the rotation detector 41 indicates the rotation state of the thighs BF relative to the waist BW. Therefore, if a predetermined condition (steps S11 and S12 in FIG. 7) is set so that the determination process 45b2 is performed immediately after the thighs BF start to rotate, the processing unit 45 can acquire the first time series data T1 and the second time series data T2 immediately after the user U starts to lift the baggage.

[0071] In this embodiment, the predetermined condition is that when the assist arm 13 rotates in a direction in which the arm angle β increases (step S11 in FIG. 7), the arm angular velocity ω becomes greater than a preset threshold value Th (step S12 in FIG. 7), so the processing unit 45 can determine the start of rotation of the thigh BF based on the threshold value Th. As a result, the processing unit 45 can appropriately acquire the first time series data T1 and the second time series data T2 immediately after the user U starts lifting the load.

[0072] The training data used in the machine learning of the trained model 46a is the first time series data T1 and the second time series data T2 obtained when the user U performs an action of lifting a load of known weight. The training data is acquired by the user U using the assist device 10. The trained model 46a is machine-trained in advance using the acquired training data. The machine learning algorithm may be either classification or regression, such as SVC (Support Vector Classification) or SVR (Support Vector Regression). Alternatively, a neural network, a decision tree, or a random forest may be used.

[0073] The processing unit 45 provides the first time series data T1 and the second time series data T2 acquired in step S13 to the trained model 46a to obtain estimated weights. After determining the estimated weight, the processing unit 45 returns to FIG. 6 and performs the control process 45a (FIG. 4) to control the motor 40 based on the estimated weight.

[0074] According to the assist device 10 having the above configuration, an estimation process 45b (FIG. 4) is executed to calculate an estimated weight of the luggage being lifted by the user U based on the output of the acceleration sensor 15 and the output of the rotation detector 41. Therefore, it is possible to control the torque (assist torque) according to the weight of the luggage without providing a dedicated sensor for detecting the weight of the luggage. As a result, the cost of the assist device 10 can be reduced, and the user U does not need to wear anything other than the first attachment 11 and the second attachment 12, making it easier for the user U to wear the assist device 10.

[0075] Here, the estimation process 45b of the assist device 10 when the user U successively lifts and puts down the baggage N in front of him as shown in FIG. 9 will be described.

[0076] FIG. 9 shows three states of the user U: an upright state, a gripping state, and a lifting state. In the upright state, the user U is in an upright posture. In the upright state, the upper body BU and thighs BF are substantially parallel to the vertical direction. In the holding state, the user U is in a crouching position and is holding a baggage N placed on a baggage stand D placed in front of the user U. In the holding state, the user U is not lifting the baggage N. In addition, the hip joints and knee joints of the user U are bent, the upper body BU is tilted forward, and the thighs BF are tilted backward. In the lifting state, the user U is in a crouching position and lifts the luggage N that he or she is holding from the luggage stand D. The hip and knee joints of the user U are bent, the upper body BU is tilted forward, and the thighs BF are tilted backward. The height position of the user U's waist BW in the lifting state is higher than the height position of the waist BW in the holding state.

[0077] Here, it is assumed that the user U lifts the luggage N by moving from an upright state, through a gripping state, to a lifted state. It is also assumed that the user U unloads the luggage N by moving from a lifted state to a gripped state and then to an upright state. It is also assumed that the user U performs the lifting and unloading operations by moving both legs, including the thighs BF, in the same way. Although Figure 9 shows a case where a load is grasped and lifted while in a crouching position, estimation process 45b can also be performed in the same way when a load is grasped and lifted while in a position where only the waist is bent without bending the knee joints.

[0078] FIG. 10 is a graph showing an example of the arm angular velocity ω when a user U continuously lifts and puts down an item N. In FIG. In FIG. 10, the horizontal axis represents time, and the vertical axis represents the arm angular velocity ω. In Figure 10, the arm angular velocity ω when the assist arm 13 rotates in a direction in which the arm angle β increases is displayed as a negative value, and the arm angular velocity ω when the assist arm 13 rotates in a direction in which the arm angle β decreases is displayed as a positive value. Note that the arm angular velocity ω displayed as a negative value in Fig. 10 represents the arm angular velocity ω when the assist arm 13 rotates in the direction in which the arm angle β increases, and does not indicate that the arm angular velocity ω is a negative numerical value. In Fig. 10, when the arm angular velocity ω indicates -50 degrees / second, this indicates that the arm angular velocity ω is 50 degrees / second when the assist arm 13 rotates in the direction in which the arm angle β increases.

[0079] In FIG. 10, from timing t1 to timing t2, the arm angular velocity ω is approximately 0, and the user U is in an upright position. Between timing t2 and timing t3, the assist arm 13 rotates in a direction in which the arm angle β decreases. In other words, between timing t2 and timing t3, the user U bends his hip joints and changes his posture from an upright state (upright posture) to a grasping state (crouching posture).

[0080] Between timing t3 and timing t4, the arm angular velocity ω is approximately 0, and the user U maintains the gripping state.

[0081] Between timing t4 and timing t6, the assist arm 13 rotates in a direction in which the arm angle β increases. In other words, between timing t4 and timing t6, the user U extends his hip joint, changes his body position from the grasping state (crouching position) to the lifting state (crouching position), and performs the lifting operation.

[0082] Between timing t6 and timing t7, the arm angular velocity ω is approximately 0, and the user U maintains the lifted state.

[0083] Between timing t7 and timing t8, the assist arm 13 rotates in a direction in which the arm angle β decreases. In other words, between timing t7 and timing t8, the user U bends his hip joint, changes his body position from a lifting state (crouching position) to a grasping state (crouching position), and performs a load-unloading operation.

[0084] Between timing t8 and timing t9, the arm angular velocity ω is approximately 0, and the user U maintains the gripping state.

[0085] Between timing t9 and timing t10, the assist arm 13 rotates in a direction in which the arm angle β increases. In other words, between timing t9 and timing t10, the user U extends his hip joints and changes his posture from a gripping state (squatting position) to an upright state (standing position).

[0086] Here, in FIG. 7, it is assumed that the threshold value Th in step S11 is 50 degrees / second. In this case, the assist arm 13 rotates in a direction in which the arm angle β increases between timing t4 and timing t6 in Fig. 10. In other words, the user U extends his hip joint between timing t4 and timing t6, changes his body position from the grasping state (crouching position) to the lifting state (crouching position), and performs the lifting operation.

[0087] 10, the arm angular velocity ω increases over time from time t4, and reaches 50 degrees / second, which is the threshold value Th, at time t5. Therefore, after time t5, the arm angular velocity ω becomes greater than the threshold value Th. In this case, the arm angular velocity ω when the assist arm 13 rotates in the direction in which the arm angle β increases becomes greater than the threshold value Th. Therefore, after passing the timing t5, the processing unit 45 determines that the arm angular velocity ω is greater than the threshold value Th (step S12 in FIG. 7), executes the weight estimation process 45b3 (FIG. 4), and acquires the first time series data T1 and the second time series data T2 (step S13 in FIG. 7).

[0088] Here, a description will be given of a mode in which the processing unit 45 acquires the time series data T12 of the arm angular velocity ω included in the first time series data T1. As described above, the processing unit 45 acquires, from the discrete value data D12 of the arm angular velocity ω, a plurality of values ​​of the arm angular velocity ω acquired during the past period P as time-series data T12 of the arm angular velocity ω. The past period P is a period extending back a predetermined time from the point in time when the arm angular velocity ω (the output of the rotation detector 41 that determined it) that is determined to be greater than the threshold value Th is acquired. In this embodiment, the time length of the past period P is set to 500 milliseconds. The processing unit 45 acquires, as time-series data T12 of the arm angular velocity ω, a plurality of values ​​of the arm angular velocity ω included in a past period P going back 500 milliseconds from the timing t5.

[0089] 10 shows the manner in which the processing unit 45 acquires the time series data T12 of the arm angular velocity ω, but the manner in which the processing unit 45 acquires the time series data T11 of the arm angle β is similar to the manner in which the processing unit 45 acquires the time series data T12 of the arm angular velocity ω. The processing unit 45 acquires a plurality of arm angles β included in the past period P as time series data of the arm angle β.

[0090] The manner in which the time series data T21 of the Y-direction acceleration and the time series data T22 of the Z-axis direction acceleration are acquired is similar to the manner in which the time series data T12 of the arm angular velocity ω is acquired. The processing unit 45 acquires multiple Y-direction accelerations included in the past period P as time series data T21 of the Y-direction acceleration, and acquires multiple Z-direction accelerations included in the past period P as time series data T22 of the Z-direction acceleration. As a result, the processing unit 45 acquires the first time series data T1 and the second time series data T2.

[0091] When the processing unit 45 acquires the first time series data T1 and the second time series data T2, it calculates an estimated weight (step S14 in FIG. 7). Furthermore, the processing unit 45 performs a control process 45a to control the motor 40 based on the estimated weight to generate an assist torque (step S2 in FIG. 6).

[0092] In addition, in FIG. 10, even during the period from timing t9 to timing t10, the arm angular velocity ω when the assist arm 13 rotates in the direction in which the arm angle β increases becomes larger than the threshold value Th. Therefore, the processing unit 45 executes the weight estimation process 45b3 (FIG. 4) between timing t9 and timing t10, and acquires the first time series data T1 and the second time series data T2. In this case, since the user U is not carrying baggage N, the estimated weight by the estimation process 45b is the value for when the user U is not carrying baggage N. The processing unit 45 generates an assist torque based on the estimated weight for when the user U is not carrying baggage N, and therefore can control the application of an assist torque greater than necessary to the user U.

[0093] [Regarding past periods] As described above, the processing unit 45 needs to acquire the first time series data T1 based on the output of the rotation detector 41 immediately after the user U starts lifting the luggage, and the second time series data T2 based on the acceleration sensor 15.

[0094] In this embodiment, the start of rotation of the thigh BF is detected using the threshold value Th, and the first time series data T1 and the second time series data T2 are obtained based on the output of the rotation detector 41 and the output of the acceleration sensor 15 acquired in the past period P, which is a period going back a predetermined time. In this way, the first time series data T1 based on the output of the rotation detector 41 immediately after the user U starts lifting the luggage, and the second time series data T2 based on the acceleration sensor 15 are obtained. For this reason, it is necessary to set the past period P (FIG. 10) appropriately.

[0095] The past period P is determined by the threshold value Th and the time length of the past period P. FIG. 11 is a graph showing an example of the f1 score when the estimated weight is calculated by changing the threshold value Th and the time length of the past period P. In FIG. 11, the horizontal axis represents the time length (seconds) of the past period P, and the vertical axis represents the f1 score. 11, graph G1 shows the f1 score when the threshold value Th is 20 degrees / second. Graph G2 shows the f1 score when the threshold value Th is 30 degrees / second. Graph G3 shows the f1 score when the threshold value Th is 40 degrees / second. Graph G4 shows the f1 score when the threshold value Th is 50 degrees / second.

[0096] As shown in FIG. 11, when the time length of the past period P is 0.3 seconds or less, a decrease in accuracy is observed, and when it is 0.4 seconds or more, relatively high accuracy is obtained. Furthermore, when the threshold value Th is 30 degrees / second or less, a decrease in accuracy is observed, and when it is in the range of 40 degrees / second to 50 degrees / second, relatively high accuracy is obtained. In this way, it can be seen that the estimation accuracy of the estimated weight can be improved by appropriately setting the past period P.

[0097] [Regarding the re-learning process] The assist device 10 of this embodiment has the function of executing the re-learning process 45c to receive externally provided teacher data and to re-learn the trained model 46a based on this teacher data.

[0098] When performing the re-learning process 45c, the processing unit 45 switches the mode from the normal mode to the learning mode. The normal mode is a mode for assisting the work of the user U, and is a mode for executing the estimation process 45b and the control process 45a described above. The learning mode is a mode for re-learning the trained model 46a.

[0099] In the learning mode, the processing unit 45 can receive teacher data via an input unit (not shown) for receiving input from the outside. In addition, in the learning mode, when a user U lifts a load of luggage whose weight is known, the processing unit 45 can use the first time series data T1 and second time series data T2 obtained at that time and the weight of the load as training data.

[0100] In this way, the trained model 46a is retrained using the relationship between the first time series data and the second time series data and the weight of the luggage when actually used by the user U as training data, so that the trained model 46a can be optimized according to the user U, and further, the accuracy of the estimated weight can be improved.

[0101] [Regarding the second embodiment] FIG. 12 is a block diagram showing the configuration of the processing unit 45 of the assist device 10 according to the second embodiment. This embodiment differs from the above embodiment in that the processing unit 45 has a function of executing muscle torque estimation processing 45b4. In the muscle torque estimation process 45b4, the processing unit 45 of this embodiment calculates an estimated muscle torque for rotating the thigh BF exerted by the muscle force of the user U based on the inclination angle α (Figure 5) of the upper body BU of the user U and the angular velocity of the upper body BU obtained from the output of the acceleration sensor 15, and the arm angle β and arm angular velocity ω obtained from the output of the rotation detector 41. The angular velocity of the upper body BU is obtained by acquiring the inclination angle α of the upper body BU over time and based on the increment of the inclination angle α per unit time. The estimated muscle torque is obtained by providing the above parameters to a pre-constructed muscle torque estimation model.

[0102] In this embodiment, when the processing unit 45 executes the judgment process 45b2 and determines that the assist arm 13 rotates in a direction that increases the arm angle β and that the arm angular velocity ω is greater than the threshold value Th (steps S11 and S12 in Figure 7), it acquires time series data of the estimated muscle torque (step S13 in Figure 7) and calculates an estimated weight based on the time series data of the estimated muscle torque (step S14 in Figure 7).

[0103] The time series data of the estimated muscle torque includes a plurality of estimated muscle torque values ​​obtained based on the outputs of a plurality of rotation detectors 41 and a plurality of acceleration sensors 15 acquired in the past period P.

[0104] The trained model 46a of this embodiment is a model obtained by machine learning the relationship between time-series data of estimated muscle torque and the weight of luggage. The processing unit 45 provides the time-series data of the estimated muscle torque to the trained model 46a to obtain the estimated weight.

[0105] In this embodiment as well, the estimated weight can be calculated based on the time-series data of the estimated muscle torque calculated in the past period P, and the estimated weight can be calculated with high accuracy.

[0106] [Regarding the third embodiment] FIG. 13 is a block diagram showing an example of the configuration of the control device 16 according to the third embodiment. This embodiment differs from the first embodiment in that the processing unit 45 further has a function of executing a correction process 45d, and the estimation process 45b further includes a motion determination process 45b6 and a generation process 45b5. The storage unit 46 also stores a first trained model 46a1, a second trained model 46a2, a third trained model 46a3, and a motion determination model 46c.

[0107] FIG. 14 is a flowchart showing the estimation process performed by the processing unit 45 of this embodiment. In FIG. 14, steps S11 and S12 are the same as steps S11 and S12 in FIG.

[0108] In FIG. 14, the processing unit 45 acquires the first time series data T1 and the second time series data T2 in step S13. The first time series data T1 of this embodiment includes average values ​​of values ​​relating to the left and right assist arms 13 at the same timing as multiple elements. That is, the time series data T11 of the arm angle β includes the average values ​​of the arm angles β of the left and right assist arms 13 at the same timing as multiple elements. Furthermore, the time series data T12 of the arm angular velocity ω includes, as a plurality of elements, the average values ​​of the arm angular velocities ω of the left and right assist arms 13 at the same timing. The processing unit 45 acquires multiple arm angles β and arm angular velocities ω acquired during the past period P from the discrete value data 46b, calculates the average value of each of the arm angles β and arm angular velocities ω, and acquires the first time series data T1. In the following description, the average value of the arm angle β will be simply referred to as the arm angle β, and the average value of the arm angular velocity ω will be simply referred to as the arm angular velocity ω.

[0109] In FIG. 14, when the first time series data T1 and the second time series data T2 are acquired in step S13, the processing unit 45 executes a generation process 45b5. In the generation process 45b5, the processing unit 45 generates the first down-sampled data and the second down-sampled data (step S21 in FIG. 14).

[0110] FIG. 15 is a diagram showing a manner in which the first down-sampled data and the second down-sampled data are generated from the first time-series data and the second time-series data. As shown in FIG. 15, the processing unit 45 down-samples the first time series data T1 and the second time series data T2 at a preset first sampling rate to generate first down-sampled data DD1. Furthermore, the processing unit 45 down-samples the first time series data T1 and the second time series data T2 at a preset second sampling rate to generate second down-sampled data DD2.

[0111] The first down-sampled data DD1 and the second down-sampled data DD2 respectively include down-sampled data of the time series data of the arm angle β, down-sampled data of the time series data of the arm angular velocity ω, down-sampled data of the time series data of the Y-direction acceleration, and down-sampled data of the time series data of the Z-direction acceleration.

[0112] In this embodiment, downsampling refers to a process of sampling multiple elements contained in each of the time series data T11 of the arm angle β, the time series data T12 of the arm angular velocity ω, the time series data T21 of the Y-direction acceleration, and the time series data T22 of the Z-direction acceleration, at a fixed rate, and thinning out elements other than the sampled elements to reduce the number of elements contained in the data. The sampling rate is a value indicating the ratio at which elements contained in the data before downsampling are sampled. In this embodiment, the first sampling rate is set to 1 / 10, and the second sampling rate is set to 1 / 2.

[0113] FIG. 16 is a diagram for explaining the downsampling performed by the processing unit 45 of this embodiment. The first time series data T1 and the second time series data T2 are shown in the upper part of Fig. 16. In Fig. 16, "○○" displayed in the upper column indicates the elements included in each time series data. For example, if the sampling interval of the discrete value data 46b is 10 milliseconds, the length of the past period P is 500 milliseconds, and the end time of the past period P is n (n is an integer and is a time expressed in 10 millisecond units), each time series data includes 51 elements corresponding to each time from time n-50 to time n, every 10 milliseconds.

[0114] The bottom row in Fig. 16 shows second downsampled data DD2 obtained by downsampling the time series data T1 and T2 shown in the top row at the second sampling rate. In Fig. 16, a "-" displayed in the bottom row indicates that the element at that time has been thinned out and does not exist.

[0115] The second down-sampled data DD2 includes down-sampled data of the arm angle β, down-sampled data of the arm angular velocity ω, down-sampled data of the Y-direction acceleration, and down-sampled data of the Z-direction acceleration.

[0116] When downsampling at the second sampling rate, the 51 elements contained in each time series data are sampled at a rate of one in two. Therefore, the 51 elements contained in each time series data are sampled at 20-millisecond intervals. In the illustrated example, each downsampled data contains 26 elements corresponding to each 20-millisecond time from time n-50 to time n. In this way, the number of elements contained in the second downsampled data DD2 is approximately half the number of elements contained in the time series data before downsampling.

[0117] Furthermore, when downsampling is performed at the first sampling rate, one in ten of the 51 elements contained in each time series data is sampled. Therefore, the 51 elements contained in each time series data are sampled at a cycle of 100 milliseconds. In this case, each downsampled data includes six elements corresponding to time n-50, time n-40, time n-30, time n-20, time n-10, and time n, respectively. In this way, the number of elements contained in the first downsampled data DD1 is approximately 1 / 10 of the number of elements contained in the time series data before downsampling.

[0118] In this way, the processing unit 45 generates the first downsampled data DD1 and the second downsampled data DD2 by downsampling the first time series data and the second time series data, thereby reducing the amount of data to be processed later.

[0119] After generating both down-sampled data DD1 and DD2 in step S21 in FIG. 14, the processing unit 45 executes the motion determination process 45b6 (FIG. 13) (step S22 in FIG. 14). In the movement determination process 45b6, the processing unit 45 determines the movement of the user U based on the first down-sampled data DD1. The processing unit 45 determines the movement of the user U when he lifts up a baggage.

[0120] The processing unit 45 performs a motion determination of the user U using a motion determination model 46c (FIG. 13) stored in the storage unit 46. The movement determination model 46c is a model obtained by machine learning the relationship between the first down-sampling data and a plurality of preset movement patterns.

[0121] In this embodiment, three movement patterns (movement pattern A, movement pattern B, and movement pattern D) are set as movement patterns when the user U lifts the baggage N. Each of the three motion patterns includes the state of the user U, which may be an upright state, a holding state, or a lifting state, as shown in Fig. 9. Each of the three motion patterns represents a pattern of the motion of the user U when lifting and unloading the luggage N as the state of the user U changes sequentially.

[0122] Of the three motion patterns, motion pattern A is the same as the motion of lifting luggage N in the first embodiment. That is, motion pattern A is a pattern in which a user U squats down and lifts luggage N placed on a luggage tray D with both hands, as shown in Fig. 9 . In the grasping state of the movement pattern A, the user U bends his knees and assumes a crouching position, and grasps the baggage N on the baggage stand D with both hands.

[0123] FIG. 17 is a diagram for explaining the operation pattern B when the user U lifts the baggage N. In FIG. As shown in FIG. 17, the movement pattern B is a pattern in which a user U lifts a baggage N placed on a baggage stand D with both hands by bending the waist without bending the knees. In the grasping state in motion pattern B, the knee joints of the user U are hardly bent, and the upper body BU is tilted forward due to the flexion of the hip joints and waist BW. In the grasping state, the user U grasps the baggage N on the baggage stand D with both hands. The lifting state in operation pattern B is almost the same as the lifting state in operation pattern A.

[0124] FIG. 18 is a diagram for explaining an operation pattern D when a user U lifts a baggage N. In FIG. As shown in FIG. 18, the movement pattern D is a pattern in which a user U crouches down and lifts a baggage N placed on a floor Y with both hands. In the gripping state in movement pattern D, similar to movement pattern A, the hip and knee joints of the user U are bent, the upper body BU is tilted forward, and the thighs BF are tilted backward. In the movement pattern D, the luggage N is placed on the floor surface Y, so the upper body BU is tilted further forward than in the movement pattern A. The lifting state in operation pattern D is almost the same as the lifting state in operation pattern A.

[0125] The movement determination model 46c (FIG. 13) is a model obtained by machine learning the relationship between the first down-sampling data (first time-series data T1 and second time-series data T2) and the above-mentioned three movement patterns. In other words, the movement determination model 46c is a model that uses the first down-sampling data as an explanatory variable and the movement pattern as a target variable. There is a correlation between the output of the rotation detector 41 and the output of the acceleration sensor 15 immediately after the user U starts lifting the luggage, and the movement of the user U when lifting the luggage. The movement determination model 46c is constructed based on this correlation.

[0126] The training data used for machine learning of the motion determination model 46c is the first time series data T1 and the second time series data T2 obtained when the user U performs the motion of lifting a load in each of the three motion patterns. The training data for the motion determination model 46c is acquired by the user U performing the motion of lifting a load in each of the three motion patterns. The motion determination model 46c is machine-learned in advance using the training data. The machine learning algorithm may be either classification or regression, such as SVC (Support Vector Classification) or SVR (Support Vector Regression). Alternatively, a neural network, a decision tree, or a random forest may be used.

[0127] The processing unit 45 provides the first down-sampled data to the movement determination model 46c, and determines which of the three movement patterns the movement of the user U belongs to (step S22 in FIG. 14).

[0128] After generating the first down-sampled data, the processing unit 45 determines the behavior of the user U based on the first down-sampled data DD1 (step S22 in FIG. 14).

[0129] Next, the processing unit 45 proceeds to step S23 and executes a weight estimation process 45b3 (FIG. 13). In the weight estimation process 45b3, the processing unit 45 obtains an estimated weight of the baggage that the user U is trying to lift based on the determination result of the motion determination and the second down-sampled data DD2. In step S23, the processing unit 45 switches the process for calculating the estimated weight depending on the result of the motion determination.

[0130] If the determination result of the motion determination is motion pattern A, the processing unit 45 proceeds to step S25 in Fig. 14. In step S25, the processing unit 45 obtains an estimated weight using the first trained model 46a1 (Fig. 13) stored in the storage unit 46. The processing unit 45 provides the second down-sampling data DD2 to the first trained model 46a1 to obtain an estimated weight. If the determination result of the motion determination is motion pattern B, the processing unit 45 proceeds to step S26 in Fig. 14. In step S26, the processing unit 45 obtains an estimated weight using the second trained model 46a2 (Fig. 13) stored in the storage unit 46. The processing unit 45 provides the second down-sampling data DD2 to the second trained model 46a2 to obtain an estimated weight. If the determination result of the motion determination is motion pattern D, the processing unit 45 proceeds to step S27 in Fig. 14. In step S27, the processing unit 45 obtains an estimated weight using the third trained model 46a3 (Fig. 13) stored in the storage unit 46. The processing unit 45 provides the second down-sampling data DD2 to the third trained model 46a3 to obtain an estimated weight.

[0131] The first trained model 46a1 is a model for movement pattern A. In other words, the first trained model 46a1 is a model obtained by machine learning the relationship between the first time series data T1 and the second time series data T2 and the weight of the luggage N when the user U lifts the luggage N with an action that conforms to the movement pattern A.

[0132] The second trained model 46a2 is a model for movement pattern B. In other words, the second trained model 46a2 is a model obtained by machine learning the relationship between the first time series data T1 and the second time series data T2 and the weight of the luggage N when the user U lifts the luggage N with an action in accordance with movement pattern B.

[0133] The third trained model 46a3 is a model for movement pattern D. In other words, the third trained model 46a3 is a model obtained by machine learning the relationship between the first time series data T1 and the second time series data T2 and the weight of the luggage N when the user U lifts the luggage N with an action that conforms to the movement pattern D.

[0134] The trained models 46a1, 46a2, and 46a3 are machine-trained in advance by the same method as the trained model 46a in the first embodiment. The processing unit 45 selects a trained model to be used according to each movement pattern, provides the second down-sampling data DD2 to the selected trained model, calculates the estimated weight, and ends the estimation process.

[0135] In other words, the motion determination process 45b6 determines which of a plurality of pre-set motion patterns the motion of the user U when lifting the luggage N corresponds to, and the weight estimation process 45b3 selectively uses a plurality of trained models that have been trained for each of a plurality of motion patterns to determine the estimated weight of the luggage, based on the relationship between the first time series data T1 and the second time series data T2 and the weight of the luggage.

[0136] As a result, a trained model according to the movement pattern of the user U can be used, and the estimation accuracy of the estimated weight can be further improved.

[0137] FIG. 19 is a flowchart showing the manner of control of the motor 40 performed by the processing unit 45 of this embodiment. After the processing unit 45 completes the estimation process 45b (FIG. 13) (step S1 in FIG. 19), it performs the correction process 45d (FIG. 13) (step S20 in FIG. 19). The correction process 45d is a process for correcting the estimated weight obtained by the estimation process 45b (FIG. 13).

[0138] The processing unit 45 corrects the estimated weight using a sigmoid function. The processing unit 45 obtains the corrected weight from the estimated weight using, for example, a sigmoid function shown in the following equation (1). Corrected weight=20 / (1+exp(ax-b))...(1)

[0139] In formula (1), x is the estimated weight. Note that the above formula shows the case where the estimated weight is calculated within the range of 0 to 20 kg. In equation (1), parameters a and b are set by fitting the above equation to each of the movement patterns A, B, and D of user U determined when calculating the estimated weight. Therefore, three parameter sets including the parameters a and b are prepared corresponding to the three operation patterns, respectively. In other words, three equations (1) are prepared corresponding to the three operation patterns, respectively. Equation (1) and the three parameter sets are stored in the storage unit 46.

[0140] The processing unit 45 selects one of the three parameter sets according to the movement pattern of the user U obtained by the movement determination process 45b6, and applies the selected parameter set to the equation (1), thereby correcting the estimated weight.

[0141] Next, the processing unit 45 performs a control process 45a (step S2 in FIG. 19). The processing unit 45 controls the motor 40 based on the corrected weight obtained in the correction process 45d to generate an assist torque. That is, the processing unit 45 uses the corrected weight as an estimated weight.

[0142] As described above, in this embodiment, the processing unit 45 further executes a correction process 45d that corrects the estimated weight obtained by the estimation process 45b using a plurality of sigmoid functions (Equation (1)) corresponding to each of a plurality of operation patterns, and the control process 45a is configured to use the corrected weight obtained by the correction process 45d as the estimated weight. In this case, the estimated weight can be nonlinearly corrected between a minimum value (0 kg) and a maximum value (20 kg). More specifically, if the estimated weight is closer to the minimum value, it is corrected to approach the minimum value, and if the estimated weight is closer to the maximum value, it is corrected to approach the maximum value. This allows the assist torque to be adjusted to an appropriate value for the user U.

[0143] In this embodiment, the estimation process 45b includes an acquisition process 45b1, a determination process 45b2, a weight estimation process 45b3, and a movement determination process 45b6. The movement determination process 45b6 is a process for determining the movement of the user U when lifting a load based on the first time series data T1 and the second time series data T2 acquired in the past period P when it is determined that the output of the rotation detector 41 satisfies a predetermined condition. In this case, the estimated weight is calculated based on the result of the motion determination of the user U, so that the accuracy of the estimated weight can be further improved.

[0144] Furthermore, in this embodiment, the estimation process 45b includes a generation process 45b5. The generation process 45b5 is a process for generating first downsampled data DD1 by downsampling the first time series data T1 and the second time series data T2 at a first sampling rate, and second downsampled data DD2 by downsampling the first time series data T1 and the second time series data T2 at a second sampling rate, when it is determined that the output of the rotation detector 41 satisfies a predetermined condition. In the motion determination process 45b6, motion determination is performed based on the first down-sampled data DD1, and in the weight estimation process 45b3, an estimated weight is obtained based on the determination result of the motion determination and the second down-sampled data DD2. This makes it possible to reduce the amount of data to be processed in the motion determination process 45b6 and the weight estimation process 45b3, and to reduce the processing load on the processing unit 45.

[0145] [Regarding the Fourth Embodiment] FIG. 20 is a diagram showing the storage unit 46 of the control device 16 according to the fourth embodiment. In this embodiment, the first time series data T1 includes time series data of the average values ​​of the left and right assist arms 13, time series data of the values ​​of the left assist arm 13, and time series data of the values ​​of the right assist arm 13, and differs from the third embodiment in that the processing unit 45 that executes the movement determination process 45b6 uses these time series data selectively to perform a more detailed movement determination of the user U.

[0146] In this embodiment, it is determined which of seven types of movement patterns the movement of the user U corresponds to, and an estimated weight is calculated according to the determined movement pattern. As shown in FIG. 20, the storage unit 46 stores a weight estimation model group 48 and a motion determination model group 50. The weight estimation model group 48 includes seven models (first weight estimation model 48a, second weight estimation model 48b, third weight estimation model 48c, fourth weight estimation model 48d, fifth weight estimation model 48e, sixth weight estimation model 48f, and seventh weight estimation model 48g) corresponding to seven types of movement patterns. The group of movement determination models 50 includes five models (a first movement determination model 50a, a second movement determination model 50b, a third movement determination model 50c, a fourth movement determination model 50d, and a fifth movement determination model 50e).

[0147] [About the movement pattern] In this embodiment, as described above, seven types of motion patterns (motion pattern A-1, motion pattern A-2, motion pattern B-1, motion pattern B-2, motion pattern C-1, motion pattern C-2, and motion pattern D) are set.

[0148] The motion pattern A-1 is the same as the motion pattern A in the second embodiment. That is, the motion pattern A is a pattern in which a user U crouches down and lifts a baggage N placed on a baggage tray D with both hands, as shown in Fig. 9 . Movement pattern A-2 is a pattern in which the user U lifts the baggage N with one hand in the movement of movement pattern A-1.

[0149] Movement pattern B-1 is the same as movement pattern B in the second embodiment. That is, movement pattern B-1 is a pattern in which a user U bends the waist without bending the knees, tilts the upper body BU forward, and lifts luggage N placed on a luggage tray D with both hands, as shown in Fig. 17 . Movement pattern B-2 is a pattern in which the user U lifts the baggage N with one hand in the movement of movement pattern B-1.

[0150] FIG. 21 is a diagram for explaining the operation pattern C-1 and the operation pattern C-2. As shown in FIG. 21, the movement pattern C-1 is a pattern in which a user U lifts a baggage N placed on a baggage stand D with both hands while bending one knee and leaning the upper body BU. In the grasping state in movement pattern C-1, one knee of the user U is bent and extended forward, and the other knee is pulled backward. In the grasping state, the user U grasps the luggage N on the luggage stand D with both hands. The lifted state in the operation pattern C-1 is a state in which the other leg of the user U is pulled further back than the one leg. The movement pattern C-2 is a pattern in which the user U lifts the baggage N with one hand in the movement of the movement pattern C-1.

[0151] Movement pattern D is the same as movement pattern D in the second embodiment. That is, movement pattern D is a pattern in which a user U crouches down and lifts a baggage N placed on a floor surface Y with both hands, as shown in Fig. 18 .

[0152] Furthermore, in this embodiment, the seven types of motion patterns are classified into three groups as shown below. Movement group 1: Movement pattern A-1, Movement pattern A-2 Movement group 2: Movement pattern B-1, Movement pattern B-2, Movement pattern C-1, Movement pattern C-2 Movement group 3: Movement pattern D

[0153] Action group 1 is a pattern in which user U lifts baggage N on baggage stand D while crouching. Action group 2 is a pattern in which the user U leans the upper body BU forward and lifts the baggage N on the baggage stand D. Action group 3 is a pattern in which the user U crouches down and picks up a bag N placed on the floor Y with both hands.

[0154] In this embodiment, first, it is determined which of the three groups the user U's movement belongs to, and then the movement is determined within each group.

[0155] [About estimation processing] FIG. 22 is a flowchart showing the estimation process performed by the processing unit 45 of this embodiment. In FIG. 22, steps S11 and S12 are the same as steps S11 and S12 in FIG.

[0156] In FIG. 22, the processing unit 45 acquires the first time series data T1 and the second time series data T2 in step S13. When the first time series data T1 and the second time series data T2 are acquired, the processing unit 45 proceeds to step S21, executes a generation process 45b5, and generates first down-sampled data and second down-sampled data.

[0157] FIG. 23 is a diagram showing a manner in which the first down-sampled data and the second down-sampled data are generated from the first time-series data and the second time-series data in this embodiment. 23 , the first time series data T1 of the present embodiment includes average angle data T11A, left angle data T11L, right angle data T11R, average angular velocity data T12A, left angular velocity data T12L, and right angular velocity data T12R. When acquiring the first time series data T1, the processing unit 45 acquires, from the discrete value data 46b, data on a plurality of arm angles β and data on a plurality of arm angular velocities ω acquired during the past period P. The processing unit 45 uses these data to generate the average angle data T11A, left angle data T11L, right angle data T11R, average angular velocity data T12A, left angular velocity data T12L, and right angular velocity data T12R.

[0158] The average angle data T11A is time-series data that includes, as multiple elements, average values ​​of the arm angles β of the left and right assist arms 13 at the same timing. The left angle data T11L is time-series data that includes the arm angle β (first value) of the left assist arm 13 as a plurality of elements. The right angle data T11R is time-series data that includes the arm angle β (second value) of the right assist arm 13 as a plurality of elements. The average angular velocity data T12A is time-series data that includes, as a plurality of elements, average values ​​of the arm angular velocities ω of the left and right assist arms 13 at the same timing. The left angular velocity data T12L is time-series data that includes the value of the arm angular velocity ω (first value) of the left assist arm 13 as a plurality of elements. The right angular velocity data T12R is time-series data that includes the values ​​of the arm angular velocity ω (second value) of the right assist arm 13 as multiple elements.

[0159] As described above, the processing unit 45 acquires time series data of the average values ​​of the left and right assist arms 13 and time series data when the values ​​of the left and right assist arms 13 are treated individually.

[0160] Furthermore, the processing unit 45 downsamples the first time series data T1 and the second time series data T2 at a first downsampling rate to generate first downsampled data DD1. Furthermore, the processing unit 45 downsamples the first time series data T1 and the second time series data T2 at a second downsampling rate to generate second downsampled data DD2. The first down-sampled data DD1 and the second down-sampled data DD2 each include the following data: Downsampled average angle data T11A Downsampled left angle data T11L Downsampled right angle data T11R Downsampled average angular velocity data T12A Downsampled left angular velocity data T12L Downsampled right angular velocity data T12R Downsampled data of Y-direction acceleration time series data Downsampled data of Z-direction acceleration time series data

[0161] The downsampling is performed in the same manner as in the second embodiment. In this embodiment, the first sampling rate is set to 1 / 5, and the second sampling rate is set to 1 / 2.

[0162] After generating both down-sampled data DD1 and DD2 in step S21 in FIG. 22, the processing unit 45 executes processing related to motion determination and calculation of estimated weight (step S30 in FIG. 22).

[0163] FIG. 24 is a flowchart showing the process of determining movement and calculating estimated weight in this embodiment. The processing unit 45 proceeds to step S41 and first determines which of the action groups 1 to 3 the action of the user U belongs to. In step S41, the processing unit 45 uses the first action determination model 50a (FIG. 20) stored in the storage unit 46.

[0164] The first motion determination model 50a (FIG. 20) is a model that uses the first down-sampled data as an explanatory variable and the motion group as a response variable.

[0165] The training data used for machine learning of the first movement determination model 50a is the following data obtained when the user U performs a movement to lift a load using each of the movement patterns of the three movement groups. Average angle data T11A included in the first time series data T1 Average angular velocity data T12A included in the first time series data T1 Second time series data T2

[0166] The training data for the first movement determination model 50a is acquired by the user U performing the movement pattern of each of the three movement groups to lift a load. The first movement determination model 50a is machine-learned in advance using the training data.

[0167] In step S41 in FIG. 24, the processing unit 45 determines which of the action groups 1 to 3 the action of the user U belongs to based on the following data included in the first down-sampled data DD1. Downsampled average angle data T11A Downsampled average angular velocity data T12A Downsampled data of Y-direction acceleration time series data Downsampled data of Z-direction acceleration time series data

[0168] If it is determined in step S41 that the action belongs to action group 1, the processing unit 45 proceeds to step S43 and determines whether the user U is lifting the baggage with one hand or both hands (step S43 in FIG. 24). That is, the processing unit 45 determines whether the movement of the user U is movement pattern A-1 or movement pattern A-2. In step S43, the processing unit 45 uses the second movement determination model 50b (FIG. 20) stored in the storage unit 46.

[0169] The second movement determination model 50b (FIG. 20) is a model that uses the first down-sampled data as an explanatory variable and the movement pattern (movement pattern A-1 or movement pattern A-2) as a response variable.

[0170] The training data used for machine learning of the second movement determination model 50b is the following data obtained when the user U performs the movement pattern A-1 and the movement pattern A-2 to lift a load. Left angle data T11L included in the first time series data T1 Right angle data T11R included in the first time series data T1 Left angular velocity data T12L included in the first time series data T1 Right angular velocity data T12R included in the first time series data T1 Second time series data T2

[0171] The training data for the second movement determination model 50b is acquired by the user U performing the movement of lifting a load using the above two movement patterns. The second movement determination model 50b is machine-learned in advance using the above training data.

[0172] In step S43 in FIG. 24, the processing unit 45 determines whether the movement of the user U is movement pattern A-1 or movement pattern A-2 based on the following data included in the first down-sampled data DD1. Downsampled left angle data T11L Downsampled right angle data T11R Downsampled left angular velocity data T12L Downsampled right angular velocity data T12R Downsampled data of Y-direction acceleration time series data Downsampled data of Z-direction acceleration time series data

[0173] If it is determined in step S43 that the user U is not lifting the baggage with one hand (if the action pattern is determined to be A-1), the processing unit 45 proceeds to step S44 and calculates the estimated weight using the first weight estimation model 48a (FIG. 20) stored in the storage unit 46. The processing unit 45 provides the second down-sampling data DD2 to the first weight estimation model 48a to calculate the estimated weight.

[0174] The first weight estimation model 48a is a model obtained by machine learning the relationship between the first time series data T1 and the second time series data T2 and the weight of the luggage N when the user U lifts the luggage N in accordance with the movement pattern A-1. The training data used for machine learning of the first weight estimation model 48a is the following data obtained when the user U performs the action of lifting a load using the action pattern A-1. Average angle data T11A included in the first time series data T1 Average angular velocity data T12A included in the first time series data T1 Second time series data T2

[0175] Therefore, in step S44, the processing unit 45 obtains the estimated weight using the following data included in the second down-sampled data DD2. Downsampled average angle data T11A Downsampled average angular velocity data T12A Downsampled data of Y-direction acceleration time series data Downsampled data of Z-direction acceleration time series data

[0176] When the estimated weight is calculated in step S44, the processing unit 45 ends the process and returns to the process of FIG.

[0177] If it is determined in step S43 that the user U is lifting the baggage with one hand (if it is determined to be motion pattern A-2), the processing unit 45 proceeds to step S45, where it calculates the estimated weight using the second weight estimation model 48b (FIG. 20) stored in the storage unit 46. The processing unit 45 provides the second down-sampling data DD2 to the second weight estimation model 48b to calculate the estimated weight.

[0178] The second weight estimation model 48b is a model obtained by machine learning the relationship between the first time series data T1 and the second time series data T2 and the weight of the luggage N when the user U lifts the luggage N in accordance with the movement pattern A-2. The training data used for machine learning of the second weight estimation model 48b is the following data obtained when the user U performs the action of lifting a load using the action pattern A-2. Left angle data T11L included in the first time series data T1 Right angle data T11R included in the first time series data T1 Left angular velocity data T12L included in the first time series data T1 Right angular velocity data T12R included in the first time series data T1 Second time series data T2

[0179] Therefore, in step S45, the processing unit 45 obtains the estimated weight using the following data included in the second down-sampled data DD2. Downsampled left angle data T11L Downsampled right angle data T11R Downsampled left angular velocity data T12L Downsampled right angular velocity data T12R Downsampled data of Y-direction acceleration time series data Downsampled data of Z-direction acceleration time series data

[0180] When the estimated weight is calculated in step S45, the processing unit 45 ends the process and returns to the process of FIG.

[0181] If it is determined in step S41 that the action belongs to action group 2, the processing unit 45 proceeds to step S46 and determines whether the user U is lifting a load with one knee bent (step S46 in FIG. 24). That is, the processing unit 45 determines whether the movement of the user U is movement pattern B-1, B-2 or movement pattern C-1, C-2. In step S46, the processing unit 45 uses the third movement determination model 50c (FIG. 20) stored in the storage unit 46.

[0182] The third movement determination model 50c (FIG. 20) is a model that uses the first down-sampled data as an explanatory variable and the state of the knee of one leg (whether it is bent or not) as a response variable.

[0183] The training data used for machine learning of the third movement determination model 50c is the following data obtained when user U performs the movement pattern B-1, B-2, and movement pattern C-1, C-2 to lift a load. Left angle data T11L included in the first time series data T1 Right angle data T11R included in the first time series data T1 Left angular velocity data T12L included in the first time series data T1 Right angular velocity data T12R included in the first time series data T1 Second time series data T2

[0184] The training data for the third movement determination model 50c is acquired by the user U performing the movement of lifting a load using the above four movement patterns. The third movement determination model 50c is machine-trained in advance using the above training data.

[0185] In step S46 of FIG. 24, the processing unit 45 determines whether or not the user U is bending one knee based on the following data included in the first down-sampled data DD1. Downsampled left angle data T11L Downsampled right angle data T11R Downsampled left angular velocity data T12L Downsampled right angular velocity data T12R Downsampled data of Y-direction acceleration time series data Downsampled data of Z-direction acceleration time series data

[0186] If it is determined in step S46 that one knee is not bent (if the user U's movement is determined to be movement pattern B-1 or B-2), the processing unit 45 proceeds to step S47 and determines whether the user U is lifting the luggage with one hand or both hands (step S47 in Figure 24). That is, the processing unit 45 determines whether the movement of the user U is movement pattern B-1 or movement pattern B-2. In step S47, the processing unit 45 uses the fourth movement determination model 50d (FIG. 20) stored in the storage unit 46.

[0187] The fourth movement determination model 50d (FIG. 20) is a model that uses the first down-sampled data as an explanatory variable and the movement pattern (movement pattern B-1 or movement pattern B-2) as a response variable.

[0188] The training data used for machine learning of the fourth movement determination model 50d is the following data obtained when the user U performs the movement pattern B-1 and the movement pattern B-2 to lift a load. Left angle data T11L included in the first time series data T1 Right angle data T11R included in the first time series data T1 Left angular velocity data T12L included in the first time series data T1 Right angular velocity data T12R included in the first time series data T1 Second time series data T2

[0189] The training data for the fourth movement determination model 50d is acquired by the user U performing the movement of lifting a load using the above two movement patterns. The fourth movement determination model 50d is machine-trained in advance using the above training data.

[0190] In step S47 of FIG. 24, the processing unit 45 determines whether the movement of the user U is movement pattern B-1 or movement pattern B-2 based on the following data included in the first down-sampled data DD1. Downsampled left angle data T11L Downsampled right angle data T11R Downsampled left angular velocity data T12L Downsampled right angular velocity data T12R Downsampled data of Y-direction acceleration time series data Downsampled data of Z-direction acceleration time series data

[0191] If it is determined in step S47 that the user U is not lifting the baggage with one hand (if the action pattern is determined to be B-1), the processing unit 45 proceeds to step S48, where it calculates the estimated weight using the third weight estimation model 48c (FIG. 20) stored in the storage unit 46. The processing unit 45 provides the second down-sampling data DD2 to the third weight estimation model 48c to calculate the estimated weight.

[0192] The third weight estimation model 48c is a model obtained by machine learning the relationship between the first time series data T1 and the second time series data T2 and the weight of the luggage N when the user U lifts the luggage N in accordance with the movement pattern B-1. The training data used for machine learning of the first weight estimation model 48a is the following data obtained when the user U performs the action of lifting a load using the action pattern A-1. Average angle data T11A included in the first time series data T1 Average angular velocity data T12A included in the first time series data T1 Second time series data T2

[0193] Therefore, in step S48, the processing unit 45 obtains the estimated weight using the following data included in the second down-sampled data DD2. Downsampled average angle data T11A Downsampled average angular velocity data T12A Downsampled data of Y-direction acceleration time series data Downsampled data of Z-direction acceleration time series data

[0194] When the estimated weight is calculated in step S48, the processing unit 45 ends the process and returns to the process of FIG.

[0195] If it is determined in step S47 that the user U is lifting the baggage with one hand (if the motion pattern is determined to be B-2), the processing unit 45 proceeds to step S49, where it calculates the estimated weight using the fourth weight estimation model 48d (FIG. 20) stored in the storage unit 46. The processing unit 45 provides the second down-sampling data DD2 to the fourth weight estimation model 48d to calculate the estimated weight.

[0196] The fourth weight estimation model 48d is a model obtained by machine learning the relationship between the first time series data T1 and the second time series data T2 and the weight of the luggage N when the user U lifts the luggage N in accordance with the movement pattern B-2. The training data used for machine learning of the fourth weight estimation model 48d is the following data obtained when the user U performs the action of lifting a load using the action pattern B-2. Left angle data T11L included in the first time series data T1 Right angle data T11R included in the first time series data T1 Left angular velocity data T12L included in the first time series data T1 Right angular velocity data T12R included in the first time series data T1 Second time series data T2

[0197] Therefore, in step S49, the processing unit 45 obtains the estimated weight using the following data included in the second down-sampled data DD2. Downsampled left angle data T11L Downsampled right angle data T11R Downsampled left angular velocity data T12L Downsampled right angular velocity data T12R Downsampled data of Y-direction acceleration time series data Downsampled data of Z-direction acceleration time series data

[0198] When the estimated weight is calculated in step S49, the processing unit 45 ends the process and returns to the process of FIG.

[0199] If it is determined in step S46 that one knee is bent (if the user U's movement is determined to be movement pattern C-1 or C-2), the processing unit 45 proceeds to step S50 and determines whether the user U is lifting the luggage with one hand or both hands (step S50 in Figure 24). That is, the processing unit 45 determines whether the movement of the user U is movement pattern C-1 or movement pattern C-2. In step S50, the processing unit 45 uses the fifth movement determination model 50e (FIG. 20) stored in the storage unit 46.

[0200] The fifth movement determination model 50e (FIG. 20) is a model that uses the first down-sampled data as an explanatory variable and the movement pattern (movement pattern C-1 or movement pattern C-2) as a response variable.

[0201] The training data used for machine learning of the fifth movement determination model 50e is the following data obtained when the user U performs the movement pattern C-1 and the movement pattern C-2 to lift a load. Left angle data T11L included in the first time series data T1 Right angle data T11R included in the first time series data T1 Left angular velocity data T12L included in the first time series data T1 Right angular velocity data T12R included in the first time series data T1 Second time series data T2

[0202] The training data for the fifth movement determination model 50e is acquired by the user U performing the movement of lifting a load using the above two movement patterns. The fifth movement determination model 50e is machine-trained in advance using the above training data.

[0203] In step S50 in FIG. 24, the processing unit 45 determines whether the movement of the user U is movement pattern C-1 or movement pattern C-2 based on the following data included in the first down-sampled data DD1. Downsampled left angle data T11L Downsampled right angle data T11R Downsampled left angular velocity data T12L Downsampled right angular velocity data T12R Downsampled data of Y-direction acceleration time series data Downsampled data of Z-direction acceleration time series data

[0204] If it is determined in step S50 that the user U is not lifting the luggage with one hand (if the motion pattern is determined to be C-1), the processing unit 45 proceeds to step S51, where it calculates the estimated weight using the fifth weight estimation model 48e (FIG. 20) stored in the storage unit 46. The processing unit 45 provides the second down-sampling data DD2 to the fifth weight estimation model 48e to calculate the estimated weight.

[0205] The fifth weight estimation model 48e is a model obtained by machine learning the relationship between the first time series data T1 and the second time series data T2 and the weight of the luggage N when the user U lifts the luggage N in accordance with the movement pattern C-1. The training data used for machine learning of the fifth weight estimation model 48e is the following data obtained when the user U performs the action of lifting a load using the action pattern C-1. Average angle data T11A included in the first time series data T1 Average angular velocity data T12A included in the first time series data T1 Second time series data T2

[0206] Therefore, in step S51, the processing unit 45 obtains an estimated weight using the following data included in the second down-sampled data DD2: Downsampled average angle data T11A Downsampled average angular velocity data T12A Downsampled data of Y-direction acceleration time series data Downsampled data of Z-direction acceleration time series data

[0207] When the estimated weight is calculated in step S51, the processing unit 45 ends the processing and returns to the processing of FIG.

[0208] If it is determined in step S50 that the user U is lifting the baggage with one hand (if the motion pattern is determined to be C-2), the processing unit 45 proceeds to step S52, where it calculates the estimated weight using the sixth weight estimation model 48f (FIG. 20) stored in the storage unit 46. The processing unit 45 provides the sixth weight estimation model 48f with the second down-sampling data DD2 to calculate the estimated weight.

[0209] The sixth weight estimation model 48f is a model obtained by machine learning the relationship between the first time series data T1 and the second time series data T2 and the weight of the luggage N when the user U lifts the luggage N in accordance with the movement pattern C-2. The training data used for machine learning of the sixth weight estimation model 48f is the following data obtained when the user U performs the action of lifting a load using the action pattern C-2. Left angle data T11L included in the first time series data T1 Right angle data T11R included in the first time series data T1 Left angular velocity data T12L included in the first time series data T1 The right angular velocity data T12R is included in the first time series data T1. Second time series data T2

[0210] Therefore, in step S52, the processing unit 45 obtains the estimated weight using the following data included in the second down-sampled data DD2: Downsampled left angle data T11L Downsampled right angle data T11R Downsampled left angular velocity data T12L Downsampled right angular velocity data T12R Downsampled data of Y-direction acceleration time series data Downsampled data of Z-direction acceleration time series data

[0211] When the estimated weight is calculated in step S52, the processing unit 45 ends the process and returns to the process of FIG.

[0212] If the action is determined to be action group 3 in step S41, the processing unit 45 proceeds to step S53, where it calculates an estimated weight using the seventh weight estimation model 48g (FIG. 20) stored in the storage unit 46. In this case, the processing unit 45 can determine that the action of the user U is action pattern D. In step S53, the processing unit 45 provides the second down-sampled data DD2 to the seventh weight estimation model 48g to obtain an estimated weight.

[0213] The seventh weight estimation model 48g is a model obtained by machine learning the relationship between the first time series data T1 and the second time series data T2 and the weight of the luggage N when the user U lifts the luggage N in accordance with the movement pattern D. The training data used for machine learning of the seventh weight estimation model 48g is the following data obtained when the user U performs the action of lifting a load using the action pattern D. Average angle data T11A included in the first time series data T1 Average angular velocity data T12A included in the first time series data T1 Second time series data T2

[0214] Therefore, in step S53, the processing unit 45 obtains the estimated weight using the following data included in the second down-sampled data DD2. Downsampled average angle data T11A Downsampled average angular velocity data T12A Downsampled data of Y-direction acceleration time series data Downsampled data of Z-direction acceleration time series data

[0215] When the estimated weight is calculated in step S53, the processing unit 45 ends the process and returns to the process of FIG. According to this embodiment, the first time series data T1 contains time series data of the average values ​​of the left and right assist arms 13 (average angle data T11A, average angular velocity data T12A) and time series data when the values ​​of the left and right assist arms 13 are treated individually (left and right angle data T11L, T11R, left and right angular velocity data T12L, T12R). By using these data appropriately, it becomes possible to determine, as a movement pattern, movements of the user U that have differences between the left and right, and the number of movement patterns available when determining the movements of the user U can be further increased.

[0216] In this embodiment, movement patterns A-2, B-2, and C-2 are set for lifting luggage with one hand, but in the case of one hand, it is also possible to use a learning model to determine whether the luggage is lifted with the left or right hand.

[0217] 〔others〕 The embodiments disclosed herein are illustrative in all respects and are not restrictive. In the above embodiment, the second time series data T2 includes time series data T21 of Y-direction acceleration and time series data T22 of Z-direction acceleration, but the second time series data T2 may further include time series data of the inclination angle α of the upper body BU and time series data of the angular velocity of the upper body BU.

[0218] In addition, in the above first embodiment, an example was given of acquiring the first time series data T1 for each of the left and right assist arms 13, but it may also be configured to acquire the first time series data T1 for either one of the left and right assist arms 13, for example. In this case, the amount of data processed by the processing unit 45 can be reduced, and the load on the processing unit 45 can be reduced. When the first time-series data T1 is acquired for each of the left and right assist arms 13, it is possible to make a determination for each of the left and right assist arms 13, thereby further improving the estimation accuracy of the estimated weight.

[0219] Furthermore, in each of the above embodiments, the processing unit 45 has exemplified the case where it determines the Y-direction acceleration of the acceleration sensor 15 and the Z-direction acceleration of the acceleration sensor 15 based on the output of the acceleration sensor 15, but it may also determine the acceleration in a direction set based on the first attachment 11 based on the output of the acceleration sensor 15 and use this as the discrete value data 46b and the second time series data T2. That is, as shown in (a) and (b) in Figure 25, the direction parallel to the longitudinal direction of the upper body BU of the user U is set as the ZZ direction based on the first attachment 11, and the direction perpendicular to the ZZ direction and the left-right direction of the user U is set as the YY direction. In this case, the processing unit 45 may obtain the acceleration in the YY direction and the acceleration in the ZZ direction based on the output of the acceleration sensor 15, and may obtain the discrete value data 46b and the second time series data T2 based on these accelerations.

[0220] The scope of the present invention is not limited to the above-described embodiments, but includes all modifications within the scope of equivalents to the configurations described in the claims. [Explanation of symbols]

[0221] 10 Assist Device 11 First attachment 12 Second attachment 13 Assist arm 14 actuator 15 acceleration sensor 16 control device 21 waist support part 21a front belt 21b rear belt 21c Waist side pad 22 Jacket part 22a Shoulder belt 22b Chest belt 23 Frame cover 24 Backpack section 25 Rotation mechanism 25a Case 25b Shaft 39 Frame pipe 40 Motor 40a Output shaft 40b spiral spring 41 rotation detector 42 reducer 42a Input shaft 43 Drive pulley 44 Wire 45 Processing section 45a Control processing 45b Estimation processing 45b1 Acquisition process 45b2 Judgment process 45b3 Weight estimation processing 45b4 Muscle torque estimation processing 45b5 Generation process 45b6 Operation determination process 45c Re-learning process 45d Correction process 46 Memory section 46a Trained model 46a1 First trained model 46a2 Second trained model 46a3 Third trained model 46b Discrete data 46c Model for motion detection 48 Weight estimation model group 48a First weight estimation model 48b Second weight estimation model 48c Third weight estimation model 48d Fourth weight estimation model 48e Fifth weight estimation model 48f 6th weight estimation model 48g 7th weight estimation model 50 Motion determination model group 50a First motion determination model 50b Second motion judgment model 50c Third motion judgment model 50d 4th movement judgment model 50e 5th movement judgment model BB Chest BF Thigh BS Shoulder BU Upper body BW Lower back D Luggage stand 11 Discrete Value Data D12 Discrete Value Data D21 Discrete Value Data D22 Discrete Value Data DD1 1st downsampling data DD2 Second downsampling data N luggage P Past period T1 First time series data T11 Time series data T12 Time series data T2 Second time series data T21 Time series data T22 Time series data Th Threshold U User g Vertical line α Inclination angle β Arm angle γ Angle ω Arm angular velocity

Claims

1. a first wearing tool to be worn at least on the waist of a user; an arm disposed along the thigh of the user and rotatable relative to the first wearing device; a motor that generates torque to rotate the arm; a second attachment provided on the arm and attached to the thigh; an acceleration sensor provided in the first wearing tool; a rotation detector that detects the rotation state of the arm; a control device for controlling the motor, The control device an estimation process for calculating an estimated weight of the baggage being lifted by the user based on the output of the acceleration sensor and the output of the rotation detector; a control process for controlling the motor based on the estimated weight; Assist device.

2. The estimation process includes: an acquisition process of acquiring the output of the rotation detector and the output of the acceleration sensor over time; a determination process for determining whether or not the output of the rotation detector satisfies a predetermined condition; and a weight estimation process for, when it is determined that the output of the rotation detector satisfies the predetermined condition, calculating the estimated weight based on first time series data based on a plurality of outputs of the rotation detector acquired in a past period going back a predetermined time from the time when the output of the rotation detector that satisfied the predetermined condition was acquired, and second time series data based on a plurality of outputs of the acceleration sensor acquired in the past period. The assist device according to claim 1 .

3. The predetermined condition is that the angular velocity of the arm obtained from the output of the rotation detector when the arm rotates in a direction in which the hip joint of the user is extended is greater than a preset threshold value. The assist device according to claim 2 .

4. the first time series data includes time series data of an angle of the arm with respect to the first attachment and time series data of an angular velocity of the arm, The second time series data includes time series data of acceleration in the up-down direction of the acceleration sensor and time series data of acceleration in the front-back direction of the user of the acceleration sensor. The assist device according to claim 2 or 3.

5. In the weight estimation process, the estimated weight is calculated using a trained model that has learned the relationship between the first time-series data, the second time-series data, and the weight of the luggage. The assist device according to any one of claims 2 to 4.

6. The processing unit a process of receiving training data indicating a relationship between the first time series data, the second time series data, and the weight of the luggage; and further executing a process of relearning the trained model based on the training data. The assist device according to claim 5 .

7. The estimation process includes: an acquisition process of acquiring the output of the rotation detector and the output of the acceleration sensor over time; a muscle torque estimation process for determining an estimated muscle torque for rotating the thigh exerted by the muscle force of the user, based on the tilt angle and angular velocity of the upper body of the user obtained from the output of the acceleration sensor and the angle and angular velocity of the arm obtained from the output of the rotation detector; a determination process for determining whether or not the output of the rotation detector satisfies a predetermined condition; a weight estimation process for, when it is determined that the output of the rotation detector satisfies the predetermined condition, calculating the estimated weight based on a plurality of rotation detector outputs and a plurality of estimated muscle torques calculated based on a plurality of acceleration sensor outputs acquired during a past period up to a predetermined time prior to the time when the output of the rotation detector that satisfied the predetermined condition was acquired. The assist device according to claim 1 .

8. The estimation process includes: an acquisition process of acquiring the output of the rotation detector and the output of the acceleration sensor over time; a determination process for determining whether or not the output of the rotation detector satisfies a predetermined condition; a motion determination process that, when it is determined that the output of the rotation detector satisfies the predetermined condition, determines the motion of the user when lifting the luggage, based on first time series data based on a plurality of outputs of the rotation detector acquired in a past period going back a predetermined time from the time when the output of the rotation detector that satisfied the predetermined condition was acquired, and second time series data based on a plurality of outputs of the acceleration sensor acquired in the past period; a weight estimation process for calculating the estimated weight based on a determination result of the operation determination, the first time-series data, and the second time-series data. The assist device according to claim 1 .

9. In the movement determination process, it is determined whether the movement of the user when lifting the luggage corresponds to any one of a plurality of preset movement patterns; In the weight estimation process, the estimated weight is obtained by selectively using a plurality of trained models that have been trained for each of the plurality of movement patterns to determine the relationship between the first time-series data and the second time-series data and the weight of the luggage. The assist device according to claim 8.

10. The estimation process includes: a generation process for generating first downsampled data by downsampling the first time series data and the second time series data at a first sampling rate and second downsampled data by downsampling the first time series data and the second time series data at a second sampling rate when it is determined that the output of the rotation detector satisfies the predetermined condition, In the motion determination process, the motion determination is performed based on the first down-sampling data, In the weight estimation process, the estimated weight is calculated based on the determination result of the motion determination and the second down-sampling data. The assist device according to claim 8 or 9.

11. The processing unit further performing a correction process to correct the estimated weight obtained by the estimation process using a plurality of sigmoid functions corresponding to the plurality of movement patterns, respectively; In the control process, the corrected weight obtained in the correction process is used as the estimated weight. The assist device according to claim 9.

12. Another arm is arranged along another thigh of the user and is rotatable relative to the first wearing device; another motor that generates torque to rotate the other arm; Another second wearing tool is provided on the other arm and attached to the other thigh; and a rotation detector for detecting a rotation state of the other arm. In the acquisition process, in addition to the output of the rotation detector and the output of the acceleration sensor, the output of the other rotation detector is acquired over time; The first time series data includes time series data of a first value based on the output of the rotation detector, time series data of a second value based on the output of the other rotation detector, and time series data of an average value of the first value and the second value. The assist device according to any one of claims 8 to 11.

Images