JP7181032B2 - power assist device - Google Patents

Description

The present invention relates to a power assist device.

In recent years, the development of AAN (Assist-as-Needed) control has been advanced in the control method of the power assist device that assists the movement of the wearer. AAN control is a control technique that provides assistance only when the wearer needs assistance. That is, in AAN control, when the wearer can follow the target trajectory by himself, the degree of support is reduced, and when the error from the target trajectory (difference between the target trajectory and the actual trajectory of the wearer's body) is large, Try to increase the degree of support. Various studies have been made on control methods for such a power assist device.

X.Fang, 3 others, "Adaptive Velocity Field Control of Upper-limb Rehabilitation Robot", 2016 28th Chinese Control and Decision Conference, 2016

In the power assist device, the present inventors have developed a control method for the AAN control system by adaptive control using a neural network in consideration of the dynamic mutual interference between the power assist device and the wearer. The characteristics of this control method are that (1) the target trajectory tracking accuracy and the magnitude of the force/torque support can be specified by two parameters, (2) the force/torque of the support determined by the control signal becomes excessively large, and the wearer The control signal can be limited to any range so as not to harm the

In the above control method and the control method of a general power assist device, basically, the power assist device is controlled so as to follow the target trajectory. However, in these control methods, it is difficult to imagine that the actually worn power assist device imagines a target time trajectory and performs control so as to accurately follow the time trajectory. Rather, it is considered that the robot creates an image of the target route and performs vague control along the image of the target route.

Thus, in order to achieve motion along the target path in the power assist device, it is not always necessary to follow the target time trajectory. In particular, when there is a large error between the target time trajectory and the current position, if the time trajectory is followed, an excessive input is generated and the load on the wearer and the power assist device increases. For this reason, if there is a large error between the target's time trajectory and its current position, the movement on the trajectory is realized by executing velocity control so that the target trajectory is reached at a certain speed instead of the time trajectory. It is important to. Therefore, it is considered that the power assist control is suitable for defining a velocity field toward the target path and performing control along the velocity field.

Non-Patent Document 1 discusses power assist control using velocity field control. However, in practice, impedance control is performed on the target time trajectory, velocity trajectory, and acceleration trajectory generated from the target velocity field. That is, in the sense of generating a trajectory from a target velocity field, only indirect velocity field control is performed.

The present invention has been made in view of the above background, and provides a method of designing a velocity field along a target path expressed by a differentiable parametric curve. In designing the velocity field, we developed an interpolation processing algorithm that enables real-time processing, and developed an adaptive AAN velocity field control system that has AAN characteristics and follows the designed target velocity, as in Patent Document 1.

The present invention includes a drive source that applies a control torque to each joint of the wearer with each joint of the wearer as a rotation axis, and an AAN control system so that the wearer is assisted when necessary. and a control unit for controlling torque, wherein the AAN control system differentiates a pre-obtained target path with time as a parameter using control points and basis functions, and is parametric. A target path with control points as parameters is designed by interpolation using an interpolation function capable of forming a closed curve, and a tangent vector at the control point closest to the current position of the target part of the wearer's body and the current position A velocity field design mechanism that calculates a target velocity at the current position based on a vector directed from the position to the nearest control point, and an unknown dynamic characteristic that is dynamic mutual interference between the wearer and the power assist device. Based on the considered mathematical model, a closed loop system that performs feedback control on the speed following error, which is the difference between the actual speed at the current position and the target speed at the current position, the current position, and the actual speed at the current position an unknown dynamics estimating mechanism for estimating the unknown dynamic characteristics by means of a neural network using a velocity, a target velocity at the current position, and a differential value of the target velocity at the current position as inputs; A dead zone is provided in the error, and a nonlinear function that bounds the velocity following error is used to ensure the boundedness of the weighting matrix W in the updating rule of the weighting matrix W of the neural network of the unknown dynamics estimation mechanism. The AAN control system calculates the control torque based on the control input from the closed loop system and the control input from the unknown dynamics estimation mechanism.

In the power assist device, an AAN control system composed of a closed loop system and an unknown dynamics estimation mechanism based on a mathematical model considering unknown dynamics is applied. In the AAN control system, the inventors have found that by controlling the AAN control system to follow the target speed instead of following the target route, it is possible to realize a smoother walking motion during free walking. I found what I could do. In addition, since feedback control is performed on the speed following error in the closed loop system and the unknown dynamic characteristics are estimated by the neural network in the unknown dynamics estimation mechanism, it is possible to reduce dynamic mutual interference between the wearer and the power assist device. . As a result, dynamic mutual interference between the wearer and the power assist device can be suppressed, and the stability of the control system can be ensured.

Furthermore, a uniform B-spline function or a NURBS function is used as the interpolation function. These functions, using control points and basis functions, can produce closed curves that are differentiable and parametric.

Furthermore, when the distance between the current position and the nearest point on the target path is greater than a predetermined distance, the target speed at the current position is corrected according to the distance. By doing so, when the toe coordinates of the leg on which the wearer's power assist device is worn deviate greatly from the designed target path, It is possible to prevent the magnitude of the target velocity at the toe coordinates of the leg from becoming excessively large.
can.

Further, the velocity field design mechanism previously calculates target velocities assuming the current position at each intersection of grid lines installed in the working coordinate space, and the target velocity at the current position is It is calculated by interpolating using the target speed of at least one or more intersections located in the vicinity of the current position among the intersections of the grid lines. By doing so, it is possible to reduce the computational load required to calculate the target velocity in the toe coordinates, and it is possible to shorten the calculation time, so that smoother movement is realized in the power assist device. can do.

According to the present invention, it is possible to provide a power assist device that can suppress dynamic mutual interference between the wearer and the power assist device even when walking freely, and that ensures the stability of the control system. .

1 is a schematic diagram showing a schematic configuration of a power assist device according to an embodiment; FIG. It is a schematic diagram which shows the state which the wearer has mounted|worn the power assist apparatus. It is a block diagram showing the flow of sensor signals and control signals in the power assist device. 1 is a schematic diagram showing the configuration of a control system of a power assist device according to an embodiment; FIG. FIG. 4 is a schematic diagram showing a curve of a cubic basis function N _i ³ (t); FIG. 4 is a schematic diagram showing control points obtained from toe coordinates of a swinging leg of a healthy person during walking measured by motion capture. FIG. 4 is a schematic diagram illustrating a procedure for obtaining a velocity vector of the tip of the leg on which the power assist device is worn by the wearer. FIG. 4 is a schematic diagram illustrating a procedure for obtaining a velocity vector of the tip of the leg on which the power assist device is worn by the wearer. FIG. 4 is a schematic diagram illustrating a procedure for obtaining a velocity vector of the tip of the leg on which the power assist device is worn by the wearer. FIG. 4 is a schematic diagram showing an example of a target velocity field set in a work coordinate space; FIG. 10 is a schematic diagram illustrating a method of interpolating a target velocity at a current position using target velocities of intersection points of grid lines; FIG. 5 is a schematic diagram showing the details of the configuration of the portion surrounded by the solid line A in FIG. 4; 4 is a flow chart showing an example of the control flow of the power assist device; 4 is a flow chart showing another example of the flow of control of the power assist device; 4 is a graph showing a target route and a trajectory of the foot coordinates of the leg on the side of the wearer wearing the power assist device, which is actually measured. It is a graph which shows the change of the assistance torque of a power-assisting device at the time of weakness and muscular strength exertion.

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described with reference to the drawings.
First, a schematic configuration of a power assist device 1 according to the present embodiment will be described with reference to FIG. FIG. 1 is a schematic diagram showing a schematic configuration of a power assist device 1 according to this embodiment. FIG. 2 is a schematic diagram showing a state in which a wearer wears the power assist device 1. As shown in FIG.

The power assist device 1 includes a driving source that applies control torque to each joint of the wearer with the joints of the wearer as rotation axes, and an AAN control system that provides control torque so that the wearer can be assisted when necessary. and a control unit for controlling the That is, as shown in FIGS. 1 and 2, the power assist device 1 includes a foot support portion 21, a lower leg support member 22, an upper leg support member 23, an ankle joint rotation member 24, and a knee joint rotation member 25. , a hip joint rotation member 26, and a control box (I/O board) 30 as a control unit.

The foot support part 21 supports the sole. The lower leg support member 22 is a rod-shaped member that is arranged along the lower leg when worn by the wearer. The upper leg support member 23 is a rod-shaped member arranged along the upper leg.

The ankle joint rotating member 24 rotatably supports the foot support portion 21 and the lower leg support member 22, and includes an actuator 24a such as a servomotor for actively rotating and driving, and an angle sensor 24b such as a rotary encoder. , has The knee joint rotating member 25 rotatably supports the lower leg support member 22 and the upper leg support member 23, and includes an actuator 25a such as a servomotor for actively rotating and driving, and an angle sensor 25b such as a rotary encoder. and have The hip joint rotary member 26 rotatably supports the upper thigh support member 23 and the waist support member 15, and includes an actuator 26a such as a servomotor for actively rotating and driving, and an angle sensor 26b such as a rotary encoder. , have Each actuator (actuator 24a, actuator 25a, actuator 26a) in the ankle joint rotating member 24, the knee joint rotating member 25, and the hip joint rotating member 26 controls each joint of the wearer with respect to each joint of the wearer as a rotation axis. It is a drive source that imparts torque.

A control box 30 as a control unit controls the control torque so that the wearer can be assisted when necessary by the AAN control system. The control box 30 detects sensor signals from each sensor (angle sensors 24b, 25b, 26b) and controls driving of each actuator (actuators 24a, 25a, 26a). The power assist device 1 may include at least one pressure sensor 35 such as a uniaxial force sensor for detecting the walking condition of the wearer on the sole of the foot support portion 21 . In addition, the power assist device 1 may include a gyro sensor 34 that measures the inclination of the wearer's torso in the vicinity of the wearer's back.

FIG. 3 is a block diagram showing the flow of sensor signals and control signals in the power assist device 1. As shown in FIG. As shown in FIG. 3, the control box 30 is preliminarily input with the trajectory of the toe coordinates of the free leg of a healthy person during walking measured by motion capture. Rotation angle sensor signals of the ankle joint rotating member 24, the knee joint rotating member 25, and the hip joint rotating member 26 are fed back to the control box 30 from the angle sensors 24b, 25b, and 26b, respectively. The control box 30 calculates a control torque for performing necessary power support using a velocity follow-up error, which will be described later, and outputs drive signals to the actuators 24a, 25a, 26a via the servo driver. , 26a. A sensor signal from the sole pressure sensor 35 is used to detect whether the foot is in contact with the floor surface.

Next, a method for controlling the power assist device 1 according to this embodiment will be described.
In designing the control system of the power assist device, it is necessary to consider dynamic mutual interference between the wearer and the power assist device as well as the stability of the control system. A mathematical model that considers the dynamic interaction between the wearer and the power assist device 1 is expressed as in Equation (1).

に含まれる質量や慣性モーメントなどの物理パラメータはすべて未知である。また、Ｔ_ｄは物理的にモデル化できないその他の未知動特性である。これを未知外乱と呼ぶ。

Here, the meaning of each symbol in the mathematical model is as follows. However, due to the dynamic interaction between the wearer and the power assist device 1,

All physical parameters such as mass and moment of inertia involved in are unknown. Also, _Td is another unknown dynamic characteristic that cannot be physically modeled. This is called an unknown disturbance.

Furthermore, the above equation of motion is transformed into the following equation (2) using a new variable, the velocity following error r. The velocity following error is the difference between the actual velocity at the current position of the target part of the wearer's body and the target velocity at the current position. That is, the velocity follow-up error r is the difference between the derivative of the joint angle q (joint angular velocity) and the target velocity (target angular velocity) θ(q) in the joint coordinate space, as shown in Equation (3). The target velocity θ(q) in the joint coordinate space is obtained by transforming the target velocity v(b) in the work coordinate space using the generalized inverse matrix of the Jacobian matrix J(q), as shown in Equation (4). can be obtained by f _d is an unknown dynamic characteristic and is represented by Equation (6).

FIG. 4 is a schematic diagram showing the configuration of the control system of the power assist device 1 according to this embodiment. As shown in FIG. 4, the control system of the power assist device 1 is an adaptive control system that calculates the control torque τ to be output to each actuator (servo motor) based on the mathematical model described above. The control system of the power assist device 1 is composed of a velocity field design mechanism, a closed loop system (AAN controller), and an unknown dynamics estimation mechanism (AAN RBFNN).

The velocity field design mechanism interpolates a previously acquired target path with time as a parameter using an interpolation function that can create a differentiable and parametric closed curve using control points and basis functions. to design. Then, the target velocity at the current position is calculated based on the tangent vector at the control point closest to the current position of the target part of the wearer's body and the vector directed from the current position to the closest control point. Here, the target part of the wearer's body is, for example, the toes or joints of the leg on which the power assist device 1 is worn by the wearer.

In order to calculate the target velocity at the current position of the target part of the wearer's body, the toe coordinates of the leg on the side where the power assist device 1 is worn by the wearer q _f =[x z] ^T 's target path must be designed. Furthermore, the target path should be a differentiable and parametric closed curve with control points as parameters. For this reason, first, motion capture is used to measure the toe coordinates of the free leg of a healthy person during walking at predetermined sampling time intervals. That is, the toe coordinates obtained by measurement are the target route with time as a parameter. Then, a closed curve is obtained by interpolating the foot trajectory of the swinging leg obtained by the measurement with an interpolation function that can create a differentiable and parametric closed curve using the control points and the basis function, and the second-order continuously differentiable ( Design a C ^2- continuous parametric curve P(t). That is, by interpolating the target route with time as a parameter using the interpolation function, the target route with control points as parameters can be designed.

Interpolation functions that can create a differentiable and parametric closed curve using control points and basis functions are, for example, B-Spline functions and NURBS (Non-Uniform Rational B-Spline) functions. A B-spline function (here, a uniform cubic B-spline function) is expressed as in Equation (7). where k is the order (here, cubic), K is the number of segments, P _i is the control points, and N _i ^k is the basis function.

The basis function N _i ³ (t) for the cubic case is N _i ³ (t)=N ₃ ³ (t−(i−3)). Here, N ₃ ³ is expressed as in Equation (8) according to the range of t.

FIG. 5 is a schematic diagram showing a curve of a cubic basis function N _i ³ (t). FIG. 6 is a schematic diagram showing control points obtained from the toe coordinates of the free leg of a healthy person during walking measured by motion capture. Here, the toe coordinates of the free leg of a healthy person during walking measured by motion capture are indicated by P _i , and the control point is indicated by P(t _i ). As shown in FIGS. 5 and 6, a control point P(t _i ) is obtained using three consecutive toe coordinates P _i . For example, the control point P(t ₃ ) is obtained using the toe coordinates P ₀ , P ₁ , P ₂ .

Note that the toe coordinates P _i for obtaining the control points may be obtained by appropriately thinning out the point group obtained by motion capture instead of all of the point group obtained by motion capture. For example, in the parametric curve P(t), the number of toe coordinates P _i for obtaining control points is relatively increased at locations where the curvature is large, and the toe coordinates P i for obtaining control points are relatively increased at locations where the curvature is small. The point cloud obtained by motion capture may be thinned out so that the number of _i points is relatively reduced.

Using the target route designed as described above, the velocity vector of the tip of the leg on which the power assist device 1 is worn by the wearer is obtained. 7 to 9 are schematic diagrams for explaining the procedure for obtaining the velocity vector of the toe of the leg on which the power assist device 1 is worn by the wearer. In the graph, the horizontal axis is the x-coordinate [m] and the vertical axis is the z-coordinate [m]. As shown in FIG. 7, first, the latest coordinates q _f (=(x, z)) of the toe of the wearer's leg on the parametric curve P(t) where the power assist device 1 is worn. A nearby point (nearest point) Q _q (u) (=(P _x (u), P _z (u)) is searched for, and the position of the leg on the side of the wearer wearing the power assist device 1 is searched. A vector directed from the toe coordinate q _f to the searched nearest point Q _q (u), that is, q _f −Q _q (u) is obtained.

Subsequently, as shown in FIG. 8, the tangent vector of the parametric curve P(t) at the searched nearest neighbor point Q _q (u) is obtained. A tangent vector is represented by Formula (9).

Subsequently, as shown in FIG. 9, the target velocity at the toe coordinate _qf of the leg on which the power assist device 1 is worn by the wearer is obtained. The target velocity at the toe coordinate _qf of the leg on which the power assist device 1 is worn by the wearer is expressed as in Equation (10). where ζ is an arbitrary constant greater than zero (ζ>0). Also, the target velocity expressed by equation (10) is a continuous function.

When the toe coordinate _qf of the leg on which the power assist device 1 is worn of the wearer deviates greatly from the designed target path, the leg on the side of the wearer wearing the power assist device 1 In order to prevent the magnitude of the target velocity at the toe coordinate q _f of from becoming excessive, the condition of equation (11) is added. That is, when the distance between the current position of the target part of the wearer's body and the nearest point on the target path is greater than a predetermined distance, the target velocity at the current position is corrected according to the distance.

When the target velocity at the toe coordinate q _f is obtained by the method described above, the closest point Q _q (u) is searched from the toe coordinate q _f , and the tangent vector at the nearest point Q _q (u) is calculated to obtain the velocity vector It is necessary to repeat the process of asking for Therefore, the method described above requires a considerable amount of computation time. Therefore, in order to achieve smoother walking motion, the target velocity field at the current position of the target part of the wearer's body is set in the work coordinate space by the velocity field design mechanism (see FIG. 4). The speed may be calculated by interpolation.

FIG. 10 is a schematic diagram showing an example of the target velocity field set in the work coordinate space. As shown in FIG. 10, at each intersection of grid lines set in advance in the work coordinate space, target velocities are calculated assuming the current position of the target part of the wearer's body. Set the velocity field. In FIG. 10, the grid width in the z direction is 0.01 [m] and the grid width in the x direction is 0.02 [m].

FIG. 11 is a schematic diagram illustrating a method of interpolating the target velocity at the current position of the target part of the wearer's body using the target velocity of the intersection point group of the grid lines. As shown in FIG. 11, the target velocity at the current position (x, z) of the target part of the wearer's body is at least one or more of the intersection points of the grid lines located near the current position (x, z). It is calculated by interpolating using the target velocity at the intersection. Here, the target velocity at the current position (x, z) is the grid line intersection points ((x ₁ , z ₁ ), (x ₁ , z ₂ ), (x ₂ , z ₁ ), (x ₂ , z ₂ )). That is, the target velocity at the current position (x, z) is calculated by equation (12).

By doing so, it is possible to reduce the computational load required to calculate the target speed at the toe coordinate _qf , and it is possible to shorten the calculation time, so that the power assist device 1 can be operated more smoothly. movement can be realized.

Referring to FIG. 4 again, in the closed loop system, feedback control is performed on the speed follow-up error r. In the unknown dynamics estimation mechanism, a neural network (AAN RBFNN ) to estimate the unknown dynamics. FIG. 12 is a schematic diagram showing the details of the configuration of the portion surrounded by the solid line A in FIG. As shown in FIG. 12, in the unknown dynamics estimation mechanism, z including the wearer's actual joint angles and angular velocities is input to the neural network. Moreover, by using the projection function for the update rule, the weight matrix does not diverge and always stays at a bounded value.

In the control system of the power assist device 1, the control torque τ is represented by Equation (13). Here, the first term is the term relating to the unknown dynamics estimation mechanism, that is, the estimated value of the unknown dynamics. K _r and γ are parameters that are set arbitrarily.

A range in which the absolute value of _ri is smaller than γ is defined as a dead zone (dead zone) as in equation (15). By providing a dead zone in π(r _i , γ) in this way, it is possible to make it insensitive, that is, not to perform any control when the velocity follow-up error r is very small. As a result, in the power assist device 1, unnecessary assistance is not provided when the speed following error r is minute. Also, by using the dead zone function Tanh(π(r, γ)), the control input can be bounded. The parameter γ is a parameter specified by the designer to specify the AAN control mode and the track following control mode, as will be described later.

The first term in the equation for the control torque τ is a term relating to the unknown dynamics estimation mechanism, that is, an estimated value of the unknown dynamic characteristics, and is expressed as in Equation (16).

Here, σ(z) is a radial basis function as an activation function, which is a nonlinear function defined as in Equation (18). Here, L is the number of units in the neural network, and n is the number of joints of the power assist device 1 .

Furthermore, an update rule using a projection function is defined as in Equation (20) below.

In the above update rule, ρ is a forgetting factor. The compensation level in the estimate of unknown dynamics is adjusted by the forgetting factor ρ. The forgetting factor ρ is a parameter specified by the designer to specify the AAN control mode and the velocity field following control mode, as will be described later.

When the forgetting factor ρ is increased, the influence of the velocity following error r on updating W is reduced. That is, when the forgetting factor ρ is increased, W in the equation (20) converges to 0 (minor value) regardless of the speed following error r, so the compensation for unknown dynamic characteristics becomes weaker, and the load on the wearer is reduced. increases. In other words, the ratio of the exerted torque of the wearer to the total torque of the operation becomes large.

On the other hand, when the forgetting factor ρ is decreased, the influence of the speed following error r on updating W increases. That is, when the forgetting factor ρ is decreased, W in the equation (20) becomes a large value when the speed following error r is large, so that the unknown dynamic characteristic is strongly compensated and the load on the wearer is reduced.
That is, the ratio of the control torque of the power assist device 1 to the total operating torque increases.

Since the above update rule is a differential equation starting from inside the convex set, the solution always stays inside the convex set. That is, it is possible to ensure the boundedness of the solution in the update rule. By limiting the control input signal to an arbitrary range in this way, the stability of the control system can be ensured. As a result, even when the speed follow-up error r is large, the control torque of the power assist device 1 does not become excessive. Thus, the forgetting factor ρ can be used to adjust the compensation level for unknown dynamics. The activation function that can be used in the neural network in the unknown dynamics estimation mechanism is not limited to the radial basis function. Any function other than the radial basis function may be used as long as it is a nonlinear function that ensures the boundedness of the solution.

As described above, the dead zone range can be changed by changing the parameter γ. Also, by changing the forgetting factor ρ, it is possible to change the influence of the velocity following error r on updating W. FIG. By changing the parameter γ and the forgetting factor ρ, the power assist device 1 can be operated in two control modes, ie, the AAN control mode and the velocity field following control mode. The AAN control mode is a control mode in which assistance is provided only when the wearer needs assistance, and the parameter γ is set to a relatively small value and the forgetting factor ρ is set to a relatively large value. On the other hand, the velocity field follow-up control mode is a control mode in which the target velocity is followed by a relatively large parameter γ and a relatively small forgetting factor ρ.

Next, the control flow of the power assist device 1 according to this embodiment will be described.
FIG. 13 is a flow chart showing an example of the control flow of the power assist device 1. As shown in FIG. As shown in FIG. 13, first, each parameter is set (step S101).

Subsequently, the control end time T is specified, and the control operation is started at time t=0 (step S102). Subsequently, each joint angle is measured to calculate the current position of the target part of the wearer's body, and the target velocity at the current position is calculated (step S103). Subsequently, each joint angular velocity is calculated from each measured joint angle, and the target velocity at the current position is converted into the velocity of each joint to calculate the velocity following error r, and the corresponding input (here, control torque τ) is calculated (step S104). Subsequently, a drive signal is output to the actuator of each joint via the servo driver (step S105).

Subsequently, it is determined whether or not the elapsed time t has reached the control end time T (t=T) (step S106). In step S106, if it is determined that the elapsed time t has not reached the control end time T (if NO), the elapsed time t=t+Δt (where Δt is the sampling time, for example Δt=0.001 [ seconds]), and the process returns to step S103. If it is determined in step S106 that the elapsed time t has reached the control end time T (YES), the process ends.

FIG. 14 is a flow chart showing another example of the control flow of the power assist device 1. As shown in FIG. The difference from the processing float shown in FIG. 13 is that the target velocity field (see FIG. 10) is set in the work coordinate space, and the target velocity at the current position of the target part of the wearer's body is interpolated using the set target velocity field. The point is that it is calculated. As shown in FIG. 14, first, each parameter is set (step S201). Subsequently, a target velocity field in the work coordinate space is set using a predesigned parametric curve P(t) (step S202).

Subsequently, the control end time T is specified, and the control operation is started at time t=0 (step S203). Subsequently, each joint angle is measured to calculate the current position of the target part of the wearer's body, and the target velocity at the current position is calculated from the set target velocity field (step S204). Subsequently, each joint angular velocity is calculated from each measured joint angle, and the target velocity at the current position is converted into the velocity of each joint to calculate the velocity following error r, and the corresponding input (here, control torque τ) is calculated (step S205). Subsequently, a drive signal is output to the actuator of each joint via the servo driver (step S206).

Subsequently, it is determined whether or not the elapsed time t has reached the control end time T (t=T) (step S207). In step S207, if it is determined that the elapsed time t has not reached the control end time T (in the case of NO), the elapsed time t=t+Δt (where Δt is the sampling time, for example Δt=0.001 [ seconds]), and the process returns to step S204. If it is determined in step S207 that the elapsed time t has reached the control end time T (YES), the process ends.

In the processing flows shown in FIGS. 13 and 14, the processing is terminated when the time t reaches the control end time T, but the processing is not limited to this. For example, the end state of the power assist device 1 may be specified, and the process may be terminated when the power assist device 1 reaches the end state. Here, the end state is, for example, a state in which the positions of the joints of the power assist device 1 have reached the target positions.

Next, an experiment for confirming the effect of the power assist device 1 according to this embodiment will be described. In this experiment, the power assist device 1 was operated in the AAN control mode according to the flow of control described with reference to FIG. 13, and changes in assist torque and trajectory following were evaluated between when the wearer was weak and when the wearer was exerting muscle strength. Specifically, since the joints of the power assist robot to be controlled are the knee and hip joints, n=2 and the number of units L of the neural network is 30. The parameter γ was set to 10 and the forgetting factor ρ to 0.04. Other parameters were set to K _r =5, Γ=diag{30, . . . , 30}. Note that Γ is a 30×30 matrix with 30 diagonal elements and 0s in the other elements. In this experiment, measurements were taken for four walking cycles, the first two cycles of the walking cycle being when the wearer was weak, and the following two cycles being when the wearer exerted muscle strength.

FIG. 15 is a graph showing the target route and the trajectory of the foot coordinates of the leg on the side of the wearer wearing the power assist device 1 that has actually been measured. Here, the horizontal axis is the x-coordinate [m] and the vertical axis is the z-coordinate [m]. The target path is indicated by a dashed line, the foot coordinate trajectory at the time of weakness is indicated by a solid line, and the foot coordinate trajectory at the time of muscle exertion is indicated by a two-dot chain line. As shown in FIG. 15, a trajectory substantially close to the target path is followed in both cases of muscle weakness and muscle exertion. The square root of the mean square error between the target path and the actual foot trajectory of the wearer's leg on which the power assist device 1 is worn is 0.72 [m] at the time of weakness and 0.29 [m] at the time of exertion of muscle strength. [m]. Thus, the present invention does not perform trajectory tracking control that directly follows the target trajectory, but AAN velocity field control that follows the velocity field along the target trajectory. I understand.

FIG. 16 is a graph showing changes in the assist torque of the power assist device 1 when the muscles are weak and when the muscles are exerted. Here, the horizontal axis is the elapsed time [s], and the vertical axis is the assist torque (control torque of the power assist device 1) [N·m]. As shown in FIG. 16, the support torque is changed with a relatively large amplitude (amplitude 13.63 [N m]) during weakness (time 0 to 40 [s]), and when muscle strength is exerted (time 40 to 40 [s]). 80 [s]), the assist torque is changed with a relatively small amplitude (amplitude 6.29 [N·m]). That is, the degree of support increases when the wearer is weak, and the degree of support decreases when the wearer exerts muscle strength. From the above, it was confirmed that the power assist device 1 according to the present embodiment appropriately provided support only when the wearer needed support.

As described above, in the power assist device 1, the following accuracy with respect to the target speed and the degree of assistance in AAN control can be adjusted by the parameter γ specifying the dead zone. Also, the unknown dynamics compensation level can be adjusted by the forgetting factor ρ included in the unknown dynamics estimation mechanism by the adaptive control system using the feedforward neural network.

As described above, in the power assist device according to the present embodiment, an AAN control system composed of a closed loop system and an unknown dynamics estimation mechanism based on a mathematical model considering unknown dynamics is applied. In the AAN control system, the inventors have found that by controlling the AAN control system to follow the target speed instead of following the target trajectory, it is possible to realize a smoother walking motion when performing free walking. I found what I could do. In addition, since feedback control is performed on the speed following error in the closed loop system and the unknown dynamic characteristics are estimated by the neural network in the unknown dynamics estimation mechanism, it is possible to reduce dynamic mutual interference between the wearer and the power assist device. . As a result, dynamic mutual interference between the wearer and the power assist device can be suppressed, and the stability of the control system can be ensured.

It should be noted that the present invention is not limited to the above embodiments, and can be modified as appropriate without departing from the scope of the invention.

1 Power assist device 15 Waist support member 21 Leg support member 22 Lower leg support member 23 Upper leg support member 24 Ankle joint rotation members 24a, 25a, 26a Actuators 24b, 25b, 26b Angle sensor 25 Knee joint rotation member 26 Hip joint rotation member 30 Control Box 34 Gyro sensor 35 Pressure sensor

Claims (4)

A drive source that applies a control torque to each joint of the wearer with each joint of the wearer as a rotation axis, and an AAN control system that controls the control torque so that the wearer can be assisted when necessary. A power assist device comprising a control unit,
The AAN control system is
A plurality of control points corresponding to the plurality of positions shown in the first target path are calculated using a first target path obtained in advance and representing the relationship between time and the position of the target part of the body. determining and interpolating the determined plurality of control points using an interpolation function capable of constructing a differentiable and parametric closed curve using the plurality of points and basis functions , the control points being parameters ; 2 target paths are generated , a control point closest to the current position of the target part of the wearer's body is specified from among the control points included in the second target path , and the closest control point is specified. a velocity field design mechanism for calculating a target velocity at the current position based on a tangent vector of the second target path at the control point and a vector directed from the current position to the nearest control point;
Velocity tracking error, which is the difference between the actual velocity at the current position and the target velocity at the current position, based on a mathematical model that takes into account unknown dynamic characteristics that are dynamic mutual interference between the wearer and the power assist device A closed loop system that performs feedback control for
an unknown dynamics estimating mechanism that inputs the current position, the actual velocity at the current position, the target velocity at the current position, and the differential value of the target velocity at the current position, and estimates the unknown dynamic characteristics using a neural network. is,
The first target path and the second target path represent target paths of the target part,
In the closed loop system, using a non-linear function that provides a dead zone in the velocity following error and bounds the velocity following error,
In the update rule of the weight matrix W of the neural network of the unknown dynamics estimation mechanism, using a nonlinear function that ensures boundedness of the weight matrix W,
The power assist device, wherein the AAN control system calculates a control torque based on a control input from the closed loop system and a control input from the unknown dynamics estimation mechanism. 2. The power assist device according to claim 1, wherein a uniform B-spline function or a NURBS function is used as said interpolation function. when the distance between the current position and the control point closest to the current position on the second target path is a predetermined distance or more , correcting the target speed at the current position according to the distance; The power assist device according to claim 1 or 2. The velocity field design mechanism calculates in advance a target velocity assuming the current position at each intersection of grid lines set in the work coordinate space, and the target velocity at the current position is calculated from the grid 4. The power assist device according to any one of claims 1 to 3, wherein the target speed of at least one or more intersections located near the current position among the intersections of the lines is interpolated to calculate the target speed.

Images