JP2004202652A - Biped robot walking with trunk twisting and method therefor - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve a walking speed of a biped walking robot by designating a trunk position vector P and a trunk twist vector Q. <P>SOLUTION: This robot walks while twisting a trunk in a range of an angle formed by a line for connecting left-right feet and the walking direction by preparing the trunk twist vector Q for walking command data. For example, the robot walks while maintaining the proportional relationship between the angle formed by the line for connecting the left-right feet and the walking direction and an angle formed by a line for connecting the left and right of the trunk and the walking direction. The robot walks at a high speed by securing a step larger than a case of a robot for walking without twisting the trunk. <P>COPYRIGHT: (C)2004,JPO&NCIPI

Description

[0001]
The present invention relates to a technique for improving the walking speed of a biped walking robot.
[0002]
2. Description of the Related Art In order to cause a bipedal walking robot to walk as intended by an operator, usually the following vectors indicating the position and posture of the robot are prepared.
Trunk position vector P: indicates the trunk position of the robot viewed from the coordinate origin.
Trunk posture vector R: indicates the direction of the trunk of the robot. 3 shows a direction in which a member corresponding to a human spine extends.
Left toe position vector C: indicates the left toe position of the robot viewed from the coordinate origin. It can also be specified by a vector from the trunk position vector P.
Left toe posture vector D: indicates the direction of the left toe.
Right toe position vector E: Indicates the right toe position of the robot viewed from the coordinate origin. It can also be specified by a vector from the trunk position vector P.
Right toe posture vector F: indicates the direction of the right toe.
The operator specifies these six types of vectors one after another according to a time series. By teaching the walking command data created in this way to the robot, the robot walks in the posture specified by the course specified by the operator.
The robot is controlled by a technique commonly called “inverted pendulum control” during walking, and is controlled so as not to fall during walking. The “trunk posture of the robot” is corrected by “inverted pendulum control”. In addition, control is performed by a technique commonly referred to as “ZMP compensation control”, and control is performed so as not to fall when unexpected irregularities or the like are encountered. The trunk position of the robot is corrected by the “ZMP compensation control”.
An example of the prior art is described in Patent Document 1.
[Patent Document 1]
JP-A-5-305579
[0003]
In the case of a bipedal walking robot, it is possible to walk by designating the trunk position vector P and the trunk posture vector R as described above.
There is a limit to the power of the actuator that can be mounted on the robot, and it must be said that the walking speed of the robot that walks using the above-described walking control technology is low.
The present invention has been developed for the purpose of realizing a technology that allows a robot to walk at high speed without changing the mechanical configuration of the robot by improving the technology for controlling the walking motion of the robot.
[0004]
The robot of the present invention walks on the left and right legs. The robot of the present invention walks while twisting the trunk so that the line connecting the left and right sides of the trunk falls within the range of the angle formed by the line connecting the left and right feet and the line perpendicular to the walking direction. It is characterized. The trunk here refers to the portion that becomes the base of the left and right feet. When the trunk is separated into the waist (lower torso) and the chest (upper torso), the waist corresponds to the trunk. If the lower torso and the upper torso are integrated, the entire torso corresponds to the trunk.
In the case of this robot, even if the rotation speed of the foot relative to the trunk is the same, a larger stride can be ensured than in the case of a robot that walks without twisting the trunk, and in the end, it walks faster. It becomes possible.
When the robot is caused to walk while twisting the trunk, as described in claim 3, a trunk torsion vector may be prepared in the walking command data, but from the toe position vector or the like taught to the robot. Alternatively, an operation of twisting the trunk by calculating the torsion angle of the trunk may be realized.
[0005]
Walking while maintaining a proportional relationship between the angle between the line connecting the left and right feet and the line orthogonal to the walking direction and the angle between the line connecting the left and right trunk and the line orthogonal to the walking direction Is preferred. For example, when the left foot lands and the right foot floats in the air and moves forward, when the angle between the line connecting the left foot and the right foot and the line orthogonal to the traveling direction is θ, the line connecting the left and right of the trunk and the walking direction Walk while adjusting the angle formed by the line perpendicular to to αθ. Here, α is a constant less than 1. In this case, the trunk can be twisted in synchronization with the forward movement of the foot, and a longer stride can be secured than in the case of a robot that walks without twisting the trunk.
[0006]
A body that twists a line connecting the left and right of the trunk within the angle range between the line connecting the left and right legs and the line orthogonal to the walking direction in the walking command data that commands the posture of the robot walking while walking on two legs. By preparing the trunk torsion vector, the robot can walk while twisting the trunk.
Data specifying the time-series change of the trunk torsion vector can be created in various modes, and the operator specifies the trunk torsion vector at the same time as specifying the trunk position vector and the trunk posture vector. Alternatively, the torsion vector may be automatically calculated by a computer from the toe position vector.
[0007]
FIG. 5 shows an apparatus 1 for creating walking instruction data for teaching a robot walking with two feet. This walking command data creation device 1 is used prior to actually operating the biped walking robot, and is used to create walking command data necessary for actually operating the biped walking robot. The walking command data created by the device 1 is taught to the biped walking robot. The biped walking robot walks according to the taught walking command data.
[0008]
In the walking command data creating device 1, an operator inputs data of a course or the like that the biped walking robot wants to walk. Specifically, six vectors shown in the lower left part (A) of FIG. 5 are input. The vector P is a trunk position vector, and indicates the trunk position of the robot viewed from the coordinate origin. The vector R is a trunk posture vector, and indicates the direction of the trunk of the robot. 3 shows a direction in which a member corresponding to a human spine extends. When the torso is separated into the waist (lower torso) and the upper torso, the waist corresponds to the trunk, the trunk position vector P defines the center position of the waist, and the trunk posture vector R is Define the direction perpendicular to the waist. The vector C is a left toe position vector, and indicates the left toe position of the robot viewed from the coordinate origin. It can also be specified by a vector from the trunk position vector P. Vector D is a left toe posture vector, and indicates the direction of the left toe. The vector E is a right toe position vector, and indicates the right toe position of the robot viewed from the coordinate origin. It can also be specified by a vector from the trunk position vector P. The vector F is a right toe posture vector, and indicates the direction of the right toe. The operator inputs these six types of vectors one after another according to a time series. Note that data input support software has been developed for this purpose. By specifying data at the main timing, intermediate data in a time series is complementarily calculated. In addition, the input amount can be reduced by giving an instruction to repeatedly use the walking command data for one step.
[0009]
The operator further inputs the target ZMP position in time series. In this case, since it is known from the vectors C to F that the left foot contact state, the two foot contact state, and the right foot contact state are known, the position in the left foot contact plane is designated in the left foot contact state, and the position in the right foot contact plane in the right foot contact state. Specify the position of. (B) shown at the lower right of FIG. 5 shows an example of a time-series change of the target ZMP in the Y direction, and there is no need to input the target ZMP within the foot contact period. The same applies to the target ZMP in the walking direction (X direction). In the left foot contact state, the position in the left foot contact plane is input, and in the right foot contact state, the position in the right foot contact plane is input. It is not necessary to input the target ZMP in the X direction during the both-foot contact period.
[0010]
FIG. 1 shows a method of setting a target ZMP. In this case, the robot walks in the X direction. A direction orthogonal to the walking direction is defined as a Y direction. The robot walks while alternately repeating the two-foot contact period and the one-foot contact period. The target ZMP within the one foot contact period must be within the contact surface of the contact foot. Otherwise, even if the ZMP actually matches the target ZMP, the robot falls. The target ZMP in the two-foot contact period may be in a region connecting the two contact surfaces. Considering the phenomenon in the Y direction, it is clear that the target ZMP must move from the inside of the left foot contact surface to the inside of the right foot contact surface during the period in which both feet are in contact with the ground (see the slope line C2). Alternatively, it is necessary to move from the right foot contact surface to the left foot contact surface (see the inclined line C4). The horizontal line C1 indicates the target ZMP when only the left foot is in contact with the ground, and the target ZMP is maintained in the left foot contact surface. The horizontal line C3 indicates the target ZMP when only the right foot is in contact with the ground, and the target ZMP is maintained in the right foot contact surface.
[0011]
The target ZMP in the X direction needs to satisfy almost the same condition. The horizontal line C6 indicates the target ZMP in the X direction when only the left foot is in contact with the ground, and the target ZMP in the X direction is maintained in the left foot contact surface. The horizontal line C8 indicates the target ZMP in the X direction when only the right foot is in contact with the ground, and the target ZMP in the X direction is maintained in the right foot contact surface. During the period in which both feet touch the ground, the target ZMP in the X direction must move from the inside of the left foot contact plane to the inside of the right foot contact plane (see slope C7), or must move from the right foot contact plane to the left foot contact plane. (See slope C9). The horizontal line C10 indicates the target ZMP in the X direction when only the left foot is touching again, and the target ZMP in the X direction is maintained in the left foot contact surface. In this case, since the robot walks and moves in the X direction, the X coordinate of the horizontal line C6 and the X coordinate of the horizontal line C10 have changed by the stride.
[0012]
By setting the target ZMP shown in FIG. 1 and controlling the actual ZMP observed during walking to match the target ZMP, the robot can walk without falling.
However, the target ZMP shown in FIG. 1 repeats stationary / constant speed change / stationary / constant speed change... With respect to time. At the timing of switching from stationary to constant velocity change or the timing of switching from constant velocity change to stationary, infinite acceleration or deceleration is required, and a large load acts on the robot. Also, the operation of the robot becomes awkward. These factors keep the maximum walking speed of the robot low.
[0013]
Some attempts have been made to modify the target ZMP change pattern where infinite acceleration or deceleration is required. FIG. 2 (B) is obtained by smoothing the pattern of FIG. 2 (A) by delay processing. By smoothing, the required maximum values of acceleration and deceleration can be reduced.
Since the target ZMP smoothed by the delay processing is delayed compared to the walking pattern, there is a problem that the walking posture of the robot is unstable. P12 in FIG. 2B corresponds to the timing of transition from the two-foot contact state to the right-foot contact state. At this time, the target ZMP must be within the right foot contact surface. However, as shown in FIG. 2 (B), the change pattern of the target ZMP smoothed by the delay processing is too slow, and may not be in the right foot contact surface yet when it has to be in the right foot contact surface, for example. The delay makes the walking posture of the robot unstable, and in a severe case, the robot falls.
In this specification, the term “right foot contact state” means a state where only the right foot is contacted. Even when both feet are in contact, the right foot is in contact with the ground, but in that case, it is called a two-foot contact state, not a right-foot contact state. The same applies to the left foot contact state.
[0014]
FIG. 2C shows a target ZMP obtained by performing a smoothing process on a sharp change point of the target ZMP. By performing the smoothing process, the required maximum values of the acceleration and the deceleration can be kept small.
The target ZMP subjected to the smoothing processing is too early or too late as compared with the walking pattern, and thus a problem occurs in that the walking posture of the robot is not stable. P13 in FIG. 2C corresponds to a timing immediately before shifting to the both-foot contact state and still in the left-foot contact state. Nevertheless, the target ZMP has begun to deviate from within the left foot contact surface. This is too early. P14 in FIG. 2C corresponds to the timing of transition from the two-foot contact state to the right-foot contact state. At this time, the target ZMP must be within the right foot contact surface. Nevertheless, the target ZMP is not yet in the right foot contact surface. This is too late. The target ZMP subjected to the smoothing process is too early or too late compared to the walking pattern, and the deviation makes the walking posture of the robot unstable, and in a severe case, the robot falls down.
[0015]
With the current technology, the problem of the target ZMP trajectory shown in FIG. 2A cannot be solved without causing a new problem, and is one of the factors that hinders the high-speed walking of the robot. I have. In this embodiment, this problem is dealt with as follows.
[0016]
As illustrated in FIG. 5B, the input target ZMP is discontinuous when viewed along the time axis. The walking command data creating device 1 is provided with a processing unit 6 that calculates a trajectory connecting a target ZMP that is discontinuous with respect to time. In the processing unit 6, as shown in FIG. 5C, the temporally discontinuous period (corresponding to the two-foot contact period) illustrated in FIG. The trajectory connected to B at the constant deceleration trajectory C is calculated.
In the uniform acceleration trajectory A, the maximum acceleration that the bipedal robot can realize without difficulty is maintained. The moving speed of the target ZMP is increased by the constant acceleration trajectory A. When the moving speed of the target ZMP reaches the maximum speed of the bipedal robot, the constant velocity trajectory B is adopted thereafter. When approaching the target ZMP on the opposite foot side, it switches to the constant deceleration trajectory C. In the constant deceleration trajectory C, the maximum deceleration that the bipedal robot can easily realize is maintained. The moving speed of the target ZMP becomes zero by the constant deceleration trajectory C. When it reaches zero, a trajectory is calculated where the target ZMP matches the target ZMP on the opposite foot. This is the same as a train accelerating, maintaining the maximum speed, decelerating and stopping, and is a motion pattern that does not waste excessively. Various techniques for calculating a trajectory for decelerating and stopping at a predetermined position have been developed.
FIG. 2D shows an example of a target ZMP trajectory connected by a trajectory including a constant acceleration trajectory and a constant deceleration trajectory. Since the ZMP moves to the other foot side, there is no problem that the target ZMP changes too quickly. On the other hand, as shown at point P16, the target ZMP has moved to a position within the one-foot contact surface before the change from the two-foot contact state to the one-foot contact state, and there is no problem that the change of the target ZMP is too slow. FIG. 2 (E) shows the time change of the moving speed of the target ZMP. The target ZMP accelerates in a constant acceleration trajectory, maintains the maximum speed when the maximum speed is reached, and then decelerates at a constant deceleration.
[0017]
In FIG. 2, it is assumed that the target ZMP within one foot contact period is stationary. However, the target ZMP within the one foot contact period does not need to be stationary, and for example, may move from the heel direction to the toe direction. The target ZMP within the one-foot contact period only needs to be within the contact surface of the contacted foot. In the case of a human, the target ZMP moves from the heel direction to the toe direction.
FIG. 3 shows a trajectory C14 in which trajectories (C11, C12, C13) of the target ZMP moving from the heel direction to the toe direction during the one foot contact period are connected by a constant acceleration trajectory, a constant velocity trajectory having the highest speed, and a constant deceleration trajectory. Is exemplified. The constant acceleration trajectory starts at the moving speed of the target ZMP moving from the heel direction to the toe direction during the one foot contact period, and the constant deceleration trajectory ends at the moving speed of the target ZMP during the one foot contact period. The moving speed of the target ZMP changes continuously, and no discontinuous change occurs.
C15 in FIG. 3 shows the trajectory of the target ZMP connected by the conventional technique, and the robot is required to operate at a large acceleration or deceleration because the speed fluctuates discontinuously.
[0018]
The trajectories illustrated in FIGS. 2 and 3 show changes in the X coordinate or the Y coordinate with respect to the time axis. Here, the time axis is set assuming that the robot walks at the maximum speed. In a constant acceleration trajectory, a constant velocity trajectory, and a constant deceleration trajectory, calculation is performed based on a maximum acceleration, a maximum speed, and a maximum deceleration that are possible within the range of the robot ability.
The data created by the walking command data creating device 1 may be taught to the robot by extending the time axis. For example, as shown in FIG. 4B, a target ZMP trajectory set in a pattern illustrated in FIG. 4A with respect to a time axis is gradually converted into a slowly slower pattern to teach the robot. There is. In this case, the operation of the robot gradually becomes slow. When the one-second data created by the walking command data creation device 1 is taught to the robot as two-second data, the operation speed of the robot is reduced by half. Alternatively, at the stage of executing the walking instruction data taught by the robot, the operation may be slowed down by extending and executing the time axis. At the stage created by the walking command data creation device 1, the robot was a constant acceleration trajectory, a constant velocity trajectory, and a constant deceleration trajectory. May not be. This technology is used to calculate the target ZMP trajectory assumed at the maximum speed. At that stage, it is important to connect with a constant acceleration trajectory, a constant velocity trajectory, and a constant deceleration trajectory. In the operation stage, it does not have to be a constant acceleration trajectory, a constant velocity trajectory, or a constant deceleration trajectory.
The change pattern of the trunk posture vector R, the toe position vectors C, E, the toe posture vectors D, F, and the target ZMP for one step can be standardized, and the change for one step is standardized. For example, walking command data can be generated by designating the trunk position vector P. The trunk position vector P can be designated using a joystick or the like. It is possible to input the trunk position vector P almost in real time by operating the joystick while observing the walking robot, and the walking command data creating device 1 creates walking command data at high speed and instructs the robot. Thereby, the walking course of the bipedal walking robot can be specified almost in real time.
[0019]
The walking command data generating device 1 includes a device 2 that calculates the dynamics of a bipedal walking robot based on the six types of vectors input by the operator and calculates the ZMP that may actually occur. Here, the inertial force generated by realizing the operation described in the time series of the six types of vectors input by the operator is calculated, and the moment due to the reaction force acting on the robot from the road surface is calculated from this and the gravity. Is calculated.
The calculated ZMP is compared with the target ZMP by the comparison device 3. The target ZMP used here is connected by the device 6, and changes smoothly with time as illustrated in FIG. 5C.
[0020]
If the calculated ZMP does not match the target ZMP, the time series of the six types of vectors input by the operator is unreasonable. Therefore, in this embodiment, the trunk position vector P is corrected by the device 4. Here, the trunk position vector P is corrected so that the calculated ZMP approaches the target ZMP. The ZMP calculation unit 2 calculates the ZMP again using the corrected trunk position vector P. The above cycle is repeated until the calculated ZMP substantially matches the target ZMP.
[0021]
When the trunk position vector P is corrected so that the calculated ZMP substantially matches the target ZMP, the corrected trunk position vector P, the inputted five vectors, and the inputted and connected target ZMP This allows the robot to walk. That is, the walking command data that is established dynamically is created. Therefore, the walking command data is completed by the corrected trunk position vector P, the input five vectors, and the input and connected target ZMP, and these are stored (device 5). The completed walking command data is taught to the bipedal walking robot.
As described above, if the walking posture for one step is standardized, and the trunk position vector P is input almost in real time by the joystick, it is possible to teach the biped walking robot in almost real time, and it is possible to perform biped walking. The next walking operation can be commanded while observing the behavior of the robot.
[0022]
Next, the mechanical structure of the bipedal walking robot will be described with reference to FIGS. In this specification, the front-back direction of the foot (the traveling direction of the robot) is the X axis, the left-right direction is the Y axis, and the vertical axis is the Z axis. Each axis is orthogonal to each other. FIG. 6 is a front view of both lower limbs of the biped walking robot of the present embodiment, FIG. 7 is a side view of the left lower limb, FIG. 8 is a diagram for explaining the structure of the ankle joint, and FIG. It is a figure explaining the detail of an actuator, and FIGS. 10-13 is a figure explaining movement of a foot part. The left and right lower limb shapes are mirror symmetric
[0023]
As shown in FIG. 6, the left and right lower limbs 12 of the robot 10 of the present embodiment include a thigh 14, a lower thigh (shin) 16, and a foot 18, and the thigh 14 and the torso 20 are hip joints. The thigh 14 and the lower leg 16 are connected by a knee joint 24, and the lower leg 16 and the foot 18 are connected by an ankle joint 26.
[0024]
First, the hip joint 22 will be described. As shown in FIG. 7, a disk 36 that rotates around the Z-axis is attached to the plate-like pelvis 28 extending substantially horizontally by bearings 34. 6, a pair of disks 36 is provided on the left and right in FIG. A shaft 30 extending from the pelvis 28 to the thigh 14 (extending in the Z-axis direction) is fixed to the center of each disk 36. The shaft 30 rotates around the Z axis with respect to the pelvis 28. The trunk is fixed to the plate-like pelvis 28.
The upper end of the thigh 14 is connected to the lower end of the shaft 30 by a universal joint 32. The universal joint 32 includes a cross-shaped universal joint, and allows the thigh 14 to rotate around the X axis and the Y axis with respect to the shaft 30. The hip joint 22 has a shaft 30 that can rotate around the Z axis with respect to the pelvis portion 28, and a universal joint 32 that allows the thigh portion 14 to rotate around the X axis and the Y axis with respect to the shaft 30, This is a three-axis joint that allows rotation about each of the X, Y, and Z axes.
[0025]
Next, the knee joint 24 will be described. At the lower end of each thigh portion 14, two flanges 40 arranged in parallel in the Y-axis direction extend downward, and at the upper end of the shaft 42 constituting each lower leg portion 16, two flanges 40 arranged in parallel in the Y-axis direction. The flange 44 is provided upward. The knee joint 24 includes a shaft 46 extending through the flanges 40 and 44 in the Y-axis direction. The knee joint 24 allows the lower leg 16 to rotate around the Y axis with respect to the thigh 14.
[0026]
Next, the ankle joint 26 will be described. FIG. 8 is a simplified and deformed view for explaining the structure of the ankle joint 26, and does not always match the actual shape and dimensions. At the lower part of the shaft 42 of the lower leg 16, two flanges 58 arranged in parallel in the X-axis direction extend downward. Two flanges 60 arranged in parallel in the Y-axis direction extend upward on the upper surface of the foot portion 18. The flange 58 of the lower leg 16 and the flange 60 of the foot 18 are connected by a cross-shaped universal joint 62 to form a universal joint. The ankle joint 26 allows the foot 18 to rotate around the X axis and the Y axis with respect to the lower leg 24. That is, the ankle joint 26 is a two-axis joint having degrees of freedom in each of the X and Y axes.
[0027]
Each joint is driven by a wire (excluding the rotation of the hip joint about the Z axis. This rotation is directly performed by a motor without using a wire). One end of each wire is attached to a distal member, and the other end is connected to an actuator composed of a motor and a lead screw. When the feed screw (extending in the Z direction) is rotated by the motor, the nut screwed to the feed screw is fed in the feed screw direction, and the tip of the wire connected to the nut advances and retreats in the Z-axis direction. By moving the wire tip in the Z-axis direction, the distal member can be pulled or loosened by the wire.
[0028]
First, the wire group for rotating the ankle joint will be described with reference to FIGS. Wire end guides 70a, 70b, 70c are fixed to the foot 18 by a mounting plate (not shown). Each of the wire termination guides 70a, 70b, 70c has an arc shape, the center axis of each arc extends in the Y-axis direction, and the arc surface has a predetermined width (distance extending along the Y axis). . The wire end guide 70a is located forward of the Y axis of the ankle joint 26 and is arranged on the X axis. The arc surface faces forward in the X-axis. The wire end guides 70b and 70c are located behind the Y axis of the ankle joint 26. The wire end guide 70b is located outside the X axis of the ankle joint 26, and the wire end guide 70c is located inside the X axis of the ankle joint 26. The arc surfaces of the wire termination guides 70b and 70c face the rear of the X-axis.
The lower ends of the three wires 66a, 66b, 66c are fixed to the respective wire connection points 72a, 72b, 72c at the lower ends of the wire termination guides 70a, 70b, 70c, and the other ends of the wires 66a, 66b, 66c. Extend to the knee joint 24 side. The arcuate surface of each wire termination guide 70a, 70b, 70c prohibits the wires 66a, 66b, 66c from sharply bending with a small radius of curvature.
[0029]
The wire connection point 72a is located ahead of the Y axis of the ankle joint 26, and when the wire 66a is pulled toward the knee joint 24, the foot 18 rotates around the Y axis of the ankle joint 26 and lifts the toe. The wire connection point 72a is arranged on the X axis of the ankle joint 26, and does not affect the rotation angle of the foot 18 around the X axis even if the wire 66a is pulled toward the knee joint 24. The wire connection point 72b is located outside the X axis of the ankle joint 26. When the wire 66b is pulled toward the knee joint 24, the foot 18 rotates around the X axis of the ankle joint 26, and Lift outside. The wire connection point 72c is located inside the X axis of the ankle joint 26, and when the wire 66c is pulled toward the knee joint 24, the foot 18 rotates around the X axis of the ankle joint 26 and Lift the inside. When the inside of the foot 18 is lifted, the wire 66c is pulled and simultaneously the wire 66b is loosened to allow the outside of the foot 18 to go down. Similarly, when lifting the outside of the foot 18, the wire 66b is pulled and the wire 66c is loosened at the same time to allow the inside of the foot 18 to go down. When the foot 18 is rotated around the X axis of the ankle joint 26, there is no need to operate the wire 66a. When the wires 66b and 66c are simultaneously pulled, the foot 18 rotates around the Y-axis of the ankle joint 26 to lift the heel. In this case, the wire 66a is loosened to allow the toe to lower. When the toe is lifted by pulling the wire 66a, the wires 66b and 66c are loosened to allow the heel to fall.
With the three wires 66a, 66b, 66c, the rotation angle of the ankle joint 26 around the X axis and the rotation angle around the Y axis can be adjusted independently.
[0030]
Note that the wire connection points may be arranged on both sides of the X-axis in front of the Y-axis of the ankle joint 26 and on the X-axis behind the Y-axis. Even when the wire connection points are arranged in this manner, the rotation angle of the ankle joint 26 around the X axis and the rotation angle around the Y axis can be independently adjusted by the wires.
[0031]
10 to 13 are schematic diagrams for explaining the movement of the foot 18, and FIGS. 10 and 11 are diagrams for explaining rotation about the X axis. FIG. 10 is a plan view of the foot 18, and FIG. 11 is a rear view of the foot 18. At the ends of the wires 66a, 66b, 66c, the wire end guides 70a, 70b, 70c are omitted, and only the wire connection points 72a, 72b, 72c are shown.
[0032]
FIG. 10 shows that the wire (not shown) connected to the wire connection point 72a contracts the effective length of the wire 66b connected to the wire connection point 72b while maintaining the neutral state, and is connected to the wire connection point 72c. This shows that the effective length of the wire 66c is extended. At this time, the foot 18 rotates around the X axis in the direction of the arrow as shown by the broken line in FIG. When the extension / contraction of the effective length of the wire is reversed, the foot 18 rotates in the direction opposite to the arrow. That is, by adjusting the extension / contraction of the effective length of the wire in this way, the foot 18 can be freely rotated around the X axis.
[0033]
FIGS. 12 and 13 are diagrams for explaining rotation about the Y axis. FIG. 12 is a plan view of the foot 18, and FIG. 13 is a side view of the foot 18. At the ends of the wires 66a, 66b, 66c, the wire end guides 70a, 70b, 70c are omitted, and only the wire connection points 72a, 72b, 72c are shown. FIG. 12 shows a case where the effective length of the wire 66a connected to the wire connection point 72a is reduced, and the effective lengths of the wires 66b and 66c connected to the wire connection points 72b and 72c are both increased. At this time, the foot 18 rotates around the Y axis in the direction of the arrow as shown by the broken line in FIG. When the extension / contraction of the effective length of the wire is reversed, the foot 18 rotates in the direction opposite to the arrow. By adjusting the extension / contraction of the effective length of the wire in this way, the foot 18 can be freely rotated around the Y axis.
Note that the wire tension required to lift the rear side of the foot 18 is greater than the wire tension required to lift the front side of the foot 18. For this reason, it is preferable that one of the three wire connection points 72a, 72b, 72c is set to the front side and the other is set to the rear side, and the heel is raised by two wires and two actuators. In this case, the performance of each actuator can be made equal.
[0034]
Although not shown, the foot 18 can be simultaneously rotated about the X axis and the Y axis. For example, when the effective length of the wire 66b is contracted at the speed ab, the effective length of the wire 66c is extended at the speed a + b (that is, contracted at -ab), and the effective length of the wire 66a is contracted at b, The foot 18 rotates around the X axis at a speed a and rises outward, and rotates around the Y axis at a speed b and rises forward. When the effective lengths of the three wires are adjusted at the same time, the foot 18 can be freely and simultaneously rotated around the X axis and the Y axis. Also, the rotation speed around the X axis and the rotation speed around the Y axis can be freely adjusted. From these facts, it is possible to adjust the X and Y axes independently of each other by using three wires for the two axes of X and Y, that is, the number of axes plus one wire.
[0035]
As shown in FIG. 8, three pulleys 64 a, 64 b, and 64 c which can freely rotate around a Y-axis axis 46 penetrating the flange 44 are provided on the upper part of the shaft 42 of the lower leg 16. And are arranged alternately. One wire 66a, 66b, 66c is wound on each pulley 64a, 64b, 64c. The wires 66a, 66b, 66c are separated from the pulleys in front of the pulleys 64a, 64b, 64c. The wires 66a, 66b, 66c apply a pulling force to the foot 18 from a position in front of the knee joint. For this reason, when the three wires 66a, 66b, 66c are simultaneously contracted at the same speed, the rotation of the foot 18 with respect to the lower leg 16 is not changed (the ankle joint 26 is not rotated), and the lower leg is not rotated. 16 can be rotated forward about the knee joint 24.
[0036]
The upper ends of the three wires 66a, 66b, 66c clearly shown in FIG. 8 are connected to actuators 68a, 68b, 68c (see FIGS. 6, 7). 6 and 7, the actuators are simplified for clarity. FIG. 9 schematically shows the details of the actuator 68 (all the actuators have the same structure, and therefore, will be described in common with suffixes omitted). They are connected by guide rods 108, 110, 112. A feed screw 120 is arranged between the pair of flanges so as to be rotatable and immovable in the axial direction. The feed screw 120 is rotated by a motor 114 and gears 116 and 118. The movable plate 104 has a nut that is screwed into the feed screw 120. The movable plate 104 is guided by the guide rods 108, 110, 112 and is movable in the axial direction and cannot rotate. The tip of the wire 66 is fixed to the movable plate 104.
When the motor 114 rotates, the feed screw 120 rotates and the movable plate 104 slides along the guide rod, and the wire 66 is pulled in or loosened.
A motor 114 of the actuator 68 and a pair of flanges 102 and 106 are fixed to the thigh 14. The guide rods 108, 110, 112 extend in the longitudinal direction of the thigh 14, and the motor is pulled or loosened in the longitudinal direction of the thigh 14 by the actuator 114 rotating the actuator 68. Or
If the distance between the pulleys 64a, 64b, 64c of the wires 66a, 66b, 66c and the connection points 72a, 72b, 72c is the effective length of the wire, the effective length of the wires 66a, 66b, 66c is extended by the motor 114. The actuator groups 68a, 68b, 68c for extending the effective lengths of the wires 66a, 66b, 66c are arranged on the thigh 14 near the hip joint 22.
[0037]
As shown in FIG. 9, a controller 200 is connected to the actuator 68. A signal for instructing the rotation angle of the ankle joint 26 and the tension of each wire (66a, 66b, 66c) is input to the controller 200 from another controller (not shown) that controls the entire movement of the robot 10. . The controller 200 controls the rotation angle and / or torque of the motor 114.
[0038]
As shown in FIG. 7, one end of a wire 66 d for rotating the lower leg 42 backward around the knee joint 24 is connected to the lower leg 42. The wire 66d is connected to the movable plate 104 of the actuator 68d through the rear of a pulley 64d (see FIG. 6) rotatably arranged at the knee joint. The movable plate 104 of the actuator 68d moves forward and backward by a motor. When the movable plate 104 moves forward and backward, the wire 66d is pulled or loosened.
[0039]
As described above, the following posture change is realized.
(1) The toe is raised by contracting the actuator 68a and loosening the actuators 68b and 68c. By loosening the actuator 68a and contracting the actuators 68b and 68c, the toe is lowered.
(2) The outside of the foot 18 is raised by contracting the actuator 68b and loosening the actuator 68c. The inside is raised by loosening the actuator 68b and contracting the actuator 68c.
(3) The lower leg 16 rotates forward by contracting the actuators 68a, 68b, 68c and loosening the actuator 68d. By loosening the actuators 68a, 68b, 68c and contracting the actuator 68d, the lower leg 16 rotates rearward.
With four actuators and four wires, the rotation angle of the ankle joint 26 around the X axis (the second rotation), the rotation angle of the ankle joint 26 around the Y axis (the one rotation), and the knee joint 24 The rotation angle around (the above three rotations) can be adjusted independently.
The actuators appear redundant because the four actuators adjust the angle of rotation about three axes. However, the rigidity with respect to the rotation angle can be adjusted using this redundancy. This point will be described later.
[0040]
Since not only the knee joint 24 but also the actuator for adjusting the rotation angle of the ankle joint 26 is arranged on the thigh 14, the tip of the artificial leg is light and the moment of inertia around the hip joint is small. Therefore, a lower limb that can be rotated around the hip joint 22 at a high speed with a small torque is obtained.
[0041]
Next, a wire and an actuator for adjusting the rotation angle around the hip joint 22 will be described. As shown in FIGS. 6 and 7, three arc-shaped wire end guides 48a, 48b, and 48c are attached at predetermined positions on the upper portion of the thigh portion 14, and the wires 50a, 50b, and 50c are respectively provided. Are hung one by one. The lower ends of the wires 50a, 50b, 50c are fixed to the lower ends 49a, 49b, 49c of the wire end guides 48a, 48b, 48c, respectively. A pulley 54 is arranged in the middle of the wire 50c attached to the rear side, and the pulley 54 is located behind the Y-axis of the hip joint 22. The upper ends of the wires 50a, 50b, 50c are connected to the movable plates of the actuators 52a, 52b, 52c, respectively. Since the feed screws of the actuators 52a and 52b are rotated by motors (not shown), the movable plates screwed to the feed screws advance and retreat by rotation of the motors. As a result, the effective lengths of the wires 50a, 50b, 50c extend and contract. The actuators 52a, 52b, 52c and the motors 56 and the like for the actuators are arranged on the body, and do not increase the moment of inertia of the lower limb rotating around the hip joint 22 at all.
[0042]
The actuators 52a and 52b are located forward of the Y axis of the hip joint 22, and when contracted, rotate the thigh 14 forward around the Y axis of the hip joint 22. The pulley 54 that guides the wire 50c is located behind the Y axis of the hip joint 22, and when the actuator 52c contracts, rotates the thigh 14 backward about the Y axis of the hip joint 22. The disk 36 rotatable on the pelvis 28 is rotated about the Z axis by a motor 38. The motor 38 is fixed to the pelvis 28.
[0043]
The wire connection point 49c is located behind the Y axis of the hip joint 22, and when the wire 50c is pulled, the thigh 14 rotates backward around the Y axis of the hip joint 22. The wire connection point 49c is arranged on the X-axis of the hip joint 22, and does not affect the rotation of the thigh 14 around the X-axis even if the wire 50c is pulled. The wire connection point 49a is located outside the X axis of the hip joint 22, and when the wire 50a is pulled, the thigh 14 rotates around the X axis of the hip joint 22 to open the thigh 14. The wire connection point 49b is located inside the X axis of the hip joint 22, and when the wire 50b is pulled, the thigh 14 rotates around the X axis of the hip joint 22 to close the thigh 14. When closing the thigh 14, the wire 50b is pulled and simultaneously the wire 50a is loosened to allow the thigh 14 to close. Similarly, when opening the thigh 14, the wire 50a is pulled and the wire 50b is simultaneously loosened to allow the thigh 14 to open. When rotating the thigh 14 around the X-axis of the hip joint 22, there is no need to operate the wire 50c. When the wires 50a and 50b are simultaneously pulled, the thigh 14 rotates forward around the Y axis of the hip joint 22 to lift the thigh 14. In this case, the wire 50c is loosened to allow the thigh 14 to rotate forward. When the thigh 14 is rotated backward by pulling the wire 50c, the wires 50a and 50b are simultaneously loosened to allow the thigh to descend.
[0044]
Thus, the hip joint 22 is adjusted as follows.
(1) The thigh 14 rotates backward by contracting the actuator 52c and loosening the actuators 52a and 52b. By loosening the actuator 52c and contracting the actuators 52a and 52b, the thigh 14 rotates forward. While a large torque is required to raise the thigh 14 forward, a large torque is not required to lower it rearward. Two actuators and two wires are used on the side where a large torque is required, and one actuator and one wire are used on the side where only a small force is required.
(2) The thigh 14 is lifted outward by contracting the actuator 52a and loosening the actuator 52b. The thigh 14 is closed by loosening the actuator 52a and contracting the actuator 52b.
The rotation angle of the hip joint 22 around the X axis (the second rotation) and the rotation angle of the hip joint 22 around the Y axis (the first rotation) can be independently adjusted by using three actuators and three wires.
[0045]
Since the actuator for moving the thigh 14 around the hip joint 22 is arranged on the trunk side, it is not necessary to move the thigh 14 for each actuator. The moment of inertia around the hip joint 22 is small. Therefore, the lower limb can be rotated at high speed around the hip joint 22 with a small torque.
[0046]
A non-linear spring 140 shown in FIGS. 14 and 15 is inserted in the middle of the three wires 66a, 66b, 66c and the like clearly shown in FIG. The spring 140 is made of spring steel and includes a flat plate portion 122, a pair of flanges 126, 126, and another pair of flanges 130, 130. A shaft 128 extends between the pair of flanges 126 and 126, and a shaft 132 extends between the pair of flanges 130 and 130. A ridge portion 124 extending parallel to the shafts 128 and 132 is formed in the flat plate portion 122. The wire 66 passes below the shaft 128, above the ridge 124, and below the shaft 132 while bending.
As shown in FIG. 16, when the wire 66 is pulled strongly, the flat plate portion 122 made of spring steel is bent and the wire 66 is stretched.
[0047]
Since the spring 140 is inserted into the wire, the wire tension can be adjusted by the actuator.
In FIG. 6, it is assumed that the amount of retraction of the actuator 68b is equal to the amount of retraction of the actuator 68c, and the foot 18 is adjusted around the X axis at right angles to the shaft 42 of the lower leg 16. From this state, it is assumed that the actuator 68b and the actuator 68c are further retracted at the same speed. In this case, since the wire 66b and the wire 66c are drawn at the same speed, the foot 18 does not rotate around the X axis. However, as the wires 66b and 66c are pulled in, the spring 140 is deformed as shown in FIG. 16, and the tension of the wires 66b and 66c increases. That is, since the robot pulls one of the two wires and rotates it clockwise, and pulls the other wire and rotates it counterclockwise, the rotation angle is obtained by pulling both wires simultaneously. Can be changed, and only the wire tension can be increased. Similarly, by loosening both wires simultaneously, only the wire tension can be reduced without changing the rotation angle.
[0048]
Wire tension determines the stiffness of the angle of rotation about the joint. For example, when the foot 18 of FIG. 6 is grounded on the ground, if the tension of both wires is low and the rigidity is low, the right side of the foot 18 is grounded if the grounded ground is high on the left side of the foot and low on the right side. The lifting wire extends, and the foot 18 is inclined according to the inclination of the ground, so that the entire foot 18 is grounded. If the rigidity is low, it flexibly follows external events.
On the other hand, when the foot 18 on one side is in the air, if the tension of both wires for adjusting the aerial posture is weak and the stiffness is low, a slight external force acts on the robot to extend the wire and the aerial posture is adjusted. It becomes unstable. In order to stabilize the posture, it is preferable that the rigidity is high. When the rigidity is high, the movement of the actuator and the rotation angle around the joint are in good agreement, and the rotation or operation around the joint can be performed at high speed.
[0049]
The robot of this embodiment is of a type (pull-pull type) in which one of two wires is pulled and rotated clockwise, and the other wire is pulled and rotated counterclockwise (pull-pull type). , The rigidity can be adjusted independently of the robot posture. When it is necessary to flexibly follow, the rigidity can be reduced, and when it is necessary to stabilize the posture, the rigidity can be adjusted to be high.
[0050]
The reason why the rigidity around the joint can be adjusted by the pull-pull method and the non-linear spring will be described with reference to FIGS. In this description, a simple example in which a foot is rotated around the Y axis by two wires will be described. FIG. 17 schematically illustrates such a configuration. As shown in FIG. 17, the foot 302 is integrated with a cylindrical pulley 303. The pulley 303 is rotatably supported around the Y axis 303c. The front wire 304 and the rear wire 306 are wound around the pulley 303, and one end of each is connected to the pulley 303 at wire connection points 303a and 303b. The other ends of the front wire 304 and the rear wire 306 are connected to a front actuator 312 and a rear actuator 314, respectively. The front actuator 312 and the rear actuator 314 are fixed to fixing members 322, 324. Actuators 312, 314 pull in and loosen wires 304, 306. A front nonlinear spring 305 and a rear nonlinear spring 307 are mounted in the middle of the wires 304 and 306.
[0051]
FIG. 18 is a graph showing the spring characteristics of the front nonlinear spring 305 and the rear nonlinear spring 307. The vertical axis (y-axis) indicates the spring force, and the horizontal axis (x-axis) indicates the amount of extension of the spring. The curve on the right side of the y-axis represents the spring characteristic of the rear nonlinear spring 307, and the left side of the y-axis represents the spring characteristic of the front nonlinear spring 305. As is clear from FIG. 18, the spring characteristics of the front nonlinear spring 305 and the rear nonlinear spring 307 have nonlinearity in which the spring force suddenly increases (the curve becomes steeper) as the elongation increases. . That is, it does not follow Hooke's law.
The manner in which the rigidity of the foot 302 around the rotation axis 303c is adjusted will be specifically described with reference to FIG. For example, it is assumed that the actuators 312 and 314 are further retracted by A (mm) from the operation amounts of the actuators 312 and 314 when the expansion amounts of the springs 305 and 307 are zero and the angle of the foot portion 302 is adjusted to a predetermined position. Then, the springs 305 and 307 are extended, and a spring force of B (kg) is generated (see points D and F). The wire tension is adjusted to B (kg). Since the tensions of the front wire 304 and the rear wire 306 are equal, the foot 302 does not rotate and maintains the adjusted position. In this state, it is assumed that a clockwise moment is applied to the foot 302 and the foot 302 is rotated to extend the rear spring 307 by C (mm) (point D → point E). On the other hand, when the rear spring 307 extends by C (mm), the front spring 305 contracts by the same amount (C (mm)) (point F → point G). Accordingly, in order to rotate the foot 302 to expand and contract the springs (305, 307) by C (mm), it is necessary to apply a force of H (kg), which is the difference between the spring force of the point E and the point G, to the pulley 303. There is.
[0052]
It is assumed that the actuators 312 and 314 are greatly retracted by J (mm) (points L and M). At this time, the spring force generated by the springs 305 and 307 is K (kg). Even in this case, since the tension of the wires 304 and 306 is equal, the foot portion 302 keeps its position without rotating. It is assumed that a clockwise moment is applied to the foot 302 in this state, and the rear spring 307 is extended by C (mm) (point L → point N). The front spring 305 contracts by C (mm) (point M → point P). Accordingly, in order to rotate the foot 302 to expand and contract the springs 307 and 305 by C (mm), it is necessary to apply a force of Q (kg), which is the difference between the spring force between the point N and the point P, to the pulley 303. .
[0053]
The force required to expand and contract the springs 307 and 305 by C (mm) from the state of tension B realized by the actuators 312 and 314 being retracted by A (mm) is H (kg). The force required to expand and contract the springs 307 and 305 by C (mm) from the state of tension K realized by the actuators 312 and 314 being retracted by J (mm) is Q (kg). Obviously, Q (kg) is larger than H (kg). That is, when the actuators 312 and 314 are largely retracted to generate a large tension on the wires 304 and 306, the rigidity of the foot 302 (the rotational moment about the Y axis required to rotate the foot 302 by a predetermined angle) is better. ) Will be higher. By changing the amount by which the actuators 312, 314 draw the wires 304, 306, the rigidity of the foot 302 can be adjusted.
When used in combination with the pull-pull method and the nonlinear spring, the rigidity can be adjusted independently of the rotation angle of the joint.
[0054]
The inclination angle of the spring characteristic graph of FIG. 18 is proportional to the rigidity. Therefore, the amount of elongation of the spring whose inclination angle corresponds to the intended rigidity is obtained, and the intended rigidity can be adjusted by giving the amount of elongation.
Regardless of the posture of the robot, if the length of the perpendicular from the rotation center to the wire, that is, the length of the moment arm is substantially constant, the tension or the amount of elongation can be determined from the rigidity.
However, when the length of the perpendicular from the rotation center to the wire, that is, the length of the moment arm, changes, the length of the moment arm must be considered before the tension is determined from the rigidity. . For example, it is assumed that the wire tension is 1 kg. At this time, the length from the rotation center to the point of application of the 1 kg tension is 10 cm (case 1) and 20 cm (case 2). At this time, the arm of the moment is longer in the case 2 and the moment is also larger. Case 2 has a greater degree of maintaining the joint rotation angle at a predetermined value against external force, even though the tension is the same. Stiffness is determined by the length of the arm of tension and moment.
When the stiffness is specified, the controller 200 calculates the tension required to adjust the specified stiffness and the moment of the moment to the specified stiffness from the arm length, and then adjusts the tension. Calculate the length of wire elongation required for The flexibility around the joint can be adjusted to the specified rigidity independently of the posture of the robot.
The rotation angle of the ankle joint 26 about the X axis is determined by the amount of pulling of the wire 66b by the actuator 68b, and the wire 66c is redundant. In this embodiment, the rigidity of the ankle joint 26 against rotation around the X axis is controlled by using the redundant amount of pulling in the wire 66c by the actuator 68c.
The rotation angle of the ankle joint 26 about the Y axis is determined by the amount of the wire 66a pulled by the actuator 68a, and the wire 66b and the wire 66c are redundant. In the present embodiment, the rigidity of the ankle joint 26 against rotation around the Y axis is controlled using the redundant amount of the wire 66b pulled by the actuator 68b and the amount of wire 66c pulled by the actuator 68c.
The rotation angle around the knee joint 24 is determined by the actuators 68a, 68b, 68c, and the wire 66d is redundant. In the present embodiment, the rigidity with respect to the rotation around the knee joint 24 is controlled using the redundant amount of pulling in the wire 66d by the actuator 68d.
The rotation angle of the hip joint 22 about the X axis is determined by the amount of the wire 50a pulled by the actuator 52a, and the wire 50b is redundant. In the present embodiment, the rigidity of the hip joint 22 against rotation about the X-axis is controlled by using the redundant amount of pull-in of the wire 50b by the actuator 52b.
The rotation angle of the hip joint 22 around the Y axis is determined by the amount of pulling of the wire 50c by the actuator 52c, and the wire 50a and the wire 50b are redundant. In the present embodiment, the rigidity of the hip joint 22 against rotation around the Y axis is controlled using the redundant amount of pulling in the wire 50a by the actuator 52a and the amount of pulling in the wire 50b by the actuator 52b.
[0055]
The robot of the present invention is driven by a wire, and there is no need to mount an actuator on each joint. Since the actuator can be mounted at a position away from the joint, the joint can be reduced in size and weight, and the degree of freedom of the mounting position of the actuator increases. Also, in the case of the present invention, the number of wires may be three for a two-axis joint, four for a three-axis joint, that is, the number of axes +1 for one joint. Normally, two wires are required for one degree of freedom of the joint. Since the required number of wires and the number of actuators are reduced, the limbs and the like can be made slimmer and lighter. As a result, the movement of the distal member can be accelerated, and a robot similar in appearance and operation to a human or animal can be realized.
[0056]
The actuator group is not limited to being arranged on the thigh. For example, an actuator group may be arranged on the upper arm of the upper limb, and the individual actuators may be operated to rotate the forearm with respect to the upper arm or to rotate the hand with respect to the forearm.
The controller of the above-described actuator receives the rotation angle and the tension of the joint, but may be configured to input the rigidity of the joint instead of the tension.
In such a configuration, the controller calculates the length of the moment arm between the rotation center of the distal member (eg, foot) and the connection point of the wire from the rotation angle of the joint, and calculates the length of the moment arm. Then, the wire tension at which the joint has the desired rigidity is calculated from the following equation. Then, the actuation amount of the actuator, which makes the wire tension a calculated value, is output to the actuator. In a configuration in which the moment arm changes greatly with the rotation of the distal member (a configuration in which there is no wire end guide), the rigidity is input to the controller instead of the tension as described above, and the calculation is performed. Thereby, the rigidity of the joint can be controlled more accurately.
[0057]
Next, a control device of a bipedal walking robot having the above-described mechanical configuration will be described.
FIG. 19 shows a configuration of a control device for controlling bipedal walking of the robot, and a trunk position vector P created by the walking command data creating device 1 (this is input by an operator through the walking command data creating device 1). ), The trunk posture vector R, the left toe position vector C, the left toe posture vector D, the right toe position vector E, and the right toe posture vector F, and calculate the inverse kinematics. A computing device 304 is provided. However, the torso position vector P is not only corrected by the walking command data generating device 1 but also further corrected by the calculating device 304 in order to execute “inverted pendulum control” and “tracing control” described later. Is entered. Also, regarding the trunk posture vector R, a process of correcting the target trunk posture vector is performed so that the actual trunk posture vector matches the target trunk posture vector, and is input to the calculation device 304.
[0058]
The calculation device 304 calculates the input trunk position vector P (which has been corrected), the trunk posture vector R (which has also been corrected), the left toe position vector C, and the left toe posture vector D Then, based on the right toe position vector E and the right toe posture vector F, the rotation angle θ of each joint necessary to realize the position and posture described by the input vectors is calculated. The rotation angle is calculated based on the rotation amount of a motor that rotates each joint. Although only two actuators are displayed in FIG. 19, the motor rotation amounts of all the actuators are calculated. In this calculation, an inverse kinematics operation is performed.
The target rotation amount θ calculated for each actuator is input to drivers 306 and 316 for each actuator. Each driver adjusts the torque applied to the motors 308 and 318 of each actuator based on the deviation between the target rotation amount and the actual rotation amount. The actual rotation amount is detected by encoders 310 and 320 provided for each motor of the actuator. By this feedback control, the motors 308 and 318 of the actuator are feedback-controlled so that the actual rotation amounts coincide with the target rotation amounts.
[0059]
The target ZMP is input to the inverted pendulum model 334. The actual trunk position vector P is also input to the inverted pendulum model 334. The actual trunk position vector P is obtained by computing the signal of the gyro 328 provided on the trunk of the robot by the computing device 330.
The inverted pendulum model 334 uses the target ZMP and the actual trunk position vector P to calculate the inclination φ from the vertical of the inclination line from the target ZMP to the trunk position vector P. The calculated φ is input to the ZMP correction amount calculation device 336. The ZMP correction amount calculating device 336 calculates the ZMP correction amount according to the following equation: ΔZMP = k1 × φ + k2 × dφ / dt. That is, a value proportional to the inclination angle φ and a value proportional to the time derivative of the inclination angle φ are added to calculate the ZMP correction amount. The calculated ZMP correction amount ΔZMP is added to the input target ZMP, and the target ZMP is corrected by ΔZMP. The process of correcting the target ZMP by ΔZMP as described above is usually called inverted pendulum control.
When walking, particularly when running at high speed, humans lean forward on the trunk to facilitate running. By tilting the trunk forward, the center of gravity is moved forward, and the feet are moved forward so as to follow the center of gravity moved forward. Inverted pendulum control corresponds to this type of control by a human, and performs control to incline the trunk position forward.
[0060]
The target ZMP (340) corrected by the inverted pendulum model is compared with the actual ZMP (326), and the deviation (342) is calculated. The actual ZMP (326) is calculated by calculating the outputs of the plurality of force sensors 322 provided on the soles of the robot by the calculation device 324. The deviation 342 is multiplied by a gain k3 and converted into a dimension of the trunk position vector P. The deviation ΔP converted into the dimension of the trunk position vector P is added to the trunk position vector P. The process of converting the deviation (342) between the target ZMP (340) and the actual ZMP (326) into a dimension of the trunk position vector P and adding the same to the trunk position vector P is usually called "tracing control process". .
[0061]
On the plane on which the robot walks, a trunk position vector P, a trunk posture vector R, a left toe position vector C, a left toe posture vector D, a right toe position vector E, a right toe posture vector F, At the stage of determining the target ZMP, unexpected irregularities may be present, and the robot's foot may step on a convex, for example. This is equivalent to a human being knocked down. The human then bends his knees and forwards his waist position to keep him from falling. The profile control process corresponds to this type of human control, and performs control to translate the trunk forward.
When the knee is bent and the trunk is moved forward in equilibrium to prevent the fall, the walking cannot be continued as it is. After the fall is prevented, it is necessary to extend the knee and return to the normal walking posture. The inverted pendulum model executes control corresponding to it.
Metaphorically speaking, the following model corresponds to the flexibility of the body, and the inverted pendulum model corresponds to the ability to maintain the walking posture.
[0062]
In the present embodiment, the target ZMP is corrected by the inverted pendulum control system 334, and the correction amount limiting device 346 is incorporated in correcting the target trunk position vector P by the deviation (342) between the target ZMP and the actual ZMP by the copying control system. This limits the amount of correction.
The correction amount of the target trunk position vector calculated by the following control model is defined as ΔP. When the correction amount ΔP is added to the input trunk position vector P by vector, and the inverse kinematics is solved by the inverse kinematics calculation device 304 to calculate the joint angle, the calculated joint angle indicates the allowable rotation range of the joint. May exceed. In this case, due to mechanical restrictions of the robot, the control device in FIG. 19 cannot execute the control as calculated. In this embodiment, the correction amount ΔP of the target torso position vector calculated in the following control loop is α ⁿ To correct the trunk position vector P. Here, α is a number between 0 and 1, n is the number of times described below, and changes as n = 0, 1, 2, 3,.
That is, instead of correcting the trunk position vector P uniformly by the equation of P + ΔP as in the related art, P + α ⁿ Correct the trunk position vector P according to × ΔP. First, if n = 0, P + α ⁰ Correct by × ΔP (= P + ΔP). If the second n = 1 can be used, P + α ¹ Correct with × ΔP. If n = 2 for the third time, P + α ² Correct with × ΔP. If n = 3 for the fourth time, P + α ³ Correct with × ΔP.
The specific processing procedure is as follows.
(1) The correction amount of the target trunk position vector calculated by the following control model is defined as ΔP.
(2) First, assuming n = 0, P + α ⁰ Correct by × ΔP (= P + ΔP).
(3) As a result of solving the inverse kinematics with P + ΔP to obtain the joint angle θ, if the rotation angles are within the allowable rotation angles for all the joints, the processing is ended there.
(4) In the above (3), if the calculated joint angle θ is out of the allowable rotation angle for at least one joint, the correction amount is reduced.
(5) For that purpose, n = 1, and correction is made by P + α × ΔP α is 1 or less, and the correction amount is reduced.
(6) As a result of solving the inverse kinematics at P + α × ΔP to obtain the joint angle θ, if all the joints are within the permissible rotation angles, the process is terminated.
(7) In the above (6), if the calculated joint angle θ of at least one joint is out of the allowable rotation angle, the correction amount is reduced.
(5) Therefore, assuming that n = 2, P + α ² Correct with × ΔP. α is 1 or less, and the correction amount is reduced.
Hereinafter, P + α while n = 3, 4, 5,... ⁿ The above procedure is repeated until all the joint rotation angles θ that realize the trunk position vector P corrected by × ΔP are within the allowable rotation angle.
[0063]
P + α ⁿ Since the correction is made as xΔP, the copying control is insufficient unless n = 0. However, if the robot does not have enough flexibility to effectively apply the copying control, it is inevitable that the copying control becomes insufficient. As a result of sufficiently applying the following control, the joint rotation angle rotates to the limit, and even if the rotation angle reaches the limit, if the rotation angle is still insufficient, there is a risk of falling. As in this embodiment, if the control amount of the control is reduced so that the joint rotation angle falls within the limit, the control is performed within the flexibility of the robot, and the robot may fall. Can be significantly reduced.
[0064]
Following control corresponds to body flexibility. The larger the gain k3 (conversion device 344) for the profile control, the more flexibly the posture of the limb of the robot changes, and the less shock it receives. Therefore, when a shock is received such as at the time of landing, it is preferable that the copying control works strongly.
On the other hand, if the copying control is always strong, the posture may be disturbed by the inertial force when one foot is standing, and the landing may be performed at a timing not scheduled to land, resulting in a large landing shock. The setting of the value of the gain k3 for the profile control involves the above-mentioned conflicting factors, and it is difficult to adjust the value to the optimum value.
[0065]
In the present embodiment, in order to solve this reciprocal relation, a method of increasing or decreasing the value of the gain k3 for the tracking control in accordance with the state of the robot is adopted.
FIG. 20A shows an example in which the gain k3 for the following control is changed in accordance with a change in the walking pattern and the like. Increase, otherwise keep low. In this case, the limb of the robot flexibly responds to the landing shock because the tracing control is effective before and after receiving the landing shock. At other times, the gain k3 for following control is small, and the posture maintaining force is kept high. The posture is not disturbed by the inertial force when one foot is standing, and the landing shock does not increase since the landing is performed at a timing not scheduled to land. The gain k3 for the following control can be increased or decreased in accordance with the time series change of the six vectors created by the walking command data creating device 1.
[0066]
(B) of FIG. 20 shows the force received by the force sensor installed on the left sole, (C) shows the force received by the force sensor installed on the right sole, and (D) shows both sensors. Shows the total force received by At the time of landing, the total force is large at the time of landing because of strong inertial force. If the total force and the gain k3 for the tracing control are made proportional, the gain k3 for the tracing control can be increased when a landing shock is received, and kept low at other times. A process 360 extending from the device 324 for calculating the signal of the force sensor 322 in FIG. 19 to the conversion device 344 indicates a process for making the gain k3 for the copying control proportional to the total force received by the force sensor 322. .
When the copying control is performed strongly, the gains k1 and k2 of the inverted pendulum model may be increased at the same time, but only the gain k3 for the copying control may be increased.
[0067]
In the inverted pendulum model, the inclination φ from the vertical of the inclination line from the target ZMP to the trunk position vector P is calculated (see 334 in FIG. 19). The calculated φ is input to the ZMP correction amount calculation device 336, and the ZMP correction amount ΔZMP is calculated by the equation of ΔZMP = k1 × φ + k2 × dφ / dt. In order for the robot to walk smoothly, it is necessary to adjust the values of the gains k1 and k2 to appropriate values. Further, it is necessary to adjust the gain k3 for the copying control to an appropriate value.
If the values of the gains k1, k2, and k3 are adjusted to appropriate values for walking control, even when the robot is stationary without walking, the robot tilts the trunk forward and corrects it. Is repeated. Here, it is called self-excited vibration. When self-excited vibration occurs, the power of the robot is wastefully consumed, which is inconvenient.
In order to solve this, while the robot is stationary without walking, the values of the gains k1 and k2 used in the equation of k1 × φ + k2 × dφ / dt for calculating the correction amount ΔZMP of the ZMP are set at the time of walking. It has been found that it is effective to lower the value below an appropriate value. In this case, the gain k3 for the copying control may be reduced at the same time. However, if the values of the gains k1 and k2 are reduced, the gain k3 for the copying control need not be reduced. It has been confirmed that even if the values of the gains k1 and k2 (and further the gain k3) are reduced while the robot is stationary without walking, the robot continues to stand upright. Power is not wasted on self-excited vibration.
FIG. 19 illustrates that the device 324 that calculates the signal of the force sensor 322 determines whether the robot is stationary or walking, and switches the gains k1 and k2 based on the result.
[0068]
The robot of this embodiment is of a type (pull-pull type) in which one of two wires is pulled and rotated clockwise, and the other wire is pulled and rotated counterclockwise. Since the spring is inserted, the rigidity can be adjusted independently of the posture of the robot. It is effective to take advantage of this characteristic while walking while changing the rigidity during walking.
During landing, unexpected irregularities may be stepped on, and the lower limbs are more susceptible to unexpected irregularities. If the rigidity is increased and faithfully controlled to the commanded posture, the robot may slip or fall because the influence of unexpected irregularities cannot be absorbed.
Therefore, in this embodiment, the rigidity is changed in accordance with the progress of the walking pattern. Here, the rigidity of the left and right lower limbs is changed separately.
FIG. 21 (A) shows a walking pattern and a change pattern of the rigidity of the left and right lower limbs. The rigidity of both the left and right lower limbs is maintained low during the contact with the ground, and is maintained high while moving in the air. The rigidity of the right lower limb is reduced immediately before the transition from the left foot contact state to the both foot contact state, and the rigidity of the left lower limb is reduced immediately before the transition from the right foot contact state to the both foot contact state. The stiffness of the left leg and the stiffness of the right leg are controlled separately. Because the rigidity of the lower limb during landing is low, even when stepping on an unexpected unevenness, the grounded lower limb flexibly responds to the influence of the unevenness. On the other hand, the rigidity of the foot in the air is kept high, and the posture maintaining force is kept high. Posture is not disturbed by inertia force when standing on one foot.
There are various variations in the rigidity change pattern illustrated in FIGS. 21A to 21C.
(A) The rigidity of the left lower leg is reduced between the left foot contact state and the both foot contact state, and increased during the right foot contact state. The rigidity of the right lower limb is reduced between the right foot contact state and the both foot contact state, and increased during the left foot contact state.
(B, C) The stiffness of the left lower limb is reduced immediately before the transition from the right foot contact state to the double foot contact state, and is increased until the next right foot contact state. The rigidity of the right lower limb is reduced immediately before the transition from the left foot contact state to the both foot contact state, and is increased until the next left foot contact state.
(B) The rigidity of the lower limb is continuously increased.
(C) Raise the rigidity of the lower limbs stepwise.
With any of the change patterns, the lower limb in contact with the ground flexibly absorbs the influence of unevenness, while the rigidity of the foot in the air is kept high, and the posture maintaining force is kept high.
[0069]
In the following, a description will be given of a technique of controlling the actuator by the control device of FIG. 19 so as to follow an instruction value related to a target rotation angle that changes over time and a rigidity that changes over time. Hereinafter, an example will be described in which the rotation angles and rigidity of the ankle joint 26 around the X axis are controlled by the actuators 68b and 68c so as to follow the specified rotation angle and rigidity.
[0070]
(First control technology)
FIG. 22 is a control block diagram of a controller 200b for an actuator 68b that rotates the ankle joint 26 outward around the X axis by a wire 66b, and a control of the controller 200c for an actuator 68c that rotates inward by a wire 66c. FIG. This corresponds to the details of the drivers 306 and 316 in FIG. In this case, the inverse kinematics calculation device 304 also indicates the rigidity of the joint against rotation. The indicated value of the stiffness changes over time.
First, the controller 200b will be described. The first converter 2b converts the rotation angle instruction value P1 of the ankle joint 26 into the rotation angle P2 of the motor 114b necessary to realize the rotation angle instruction value P1. Here, the calculation is performed assuming that the spring 140 does not expand. The calculated rotation angle P2 is compared with the actual rotation angle P7 of the motor 114b to obtain a deviation P3. The actual rotation angle P7 of the motor 114b is obtained from an encoder 115b built in the motor 114b. The second converter 4b converts the rotation angle deviation P3 into a rotation speed P4 of the motor 114b. Here, the larger the deviation P3 is, the larger the rotation speed P4 is. If the deviation P3 is zero, the rotation speed P4 is also zero. The third converter 6b converts the rotation speed P4 of the motor 114b into a torque P5 applied to the motor 114b. The motor 114b has a characteristic in which a current value to be supplied and a generated torque are proportional. The torque P5 applied to the motor 114b is commanded in units of current. The torque P5 commanded in the unit of current is equivalent to a torque increase / decrease value for feedback-controlling the actuator torque based on the deviation between the designated rotation angle and the actual rotation angle. Increase insufficient actuator torque and reduce excessive actuator torque for the indicated angle of rotation. As a result, the actual rotation angle is feedback-controlled to the specified rotation angle.
The stiffness instruction value Q1 is input to the fourth converter 10b. The fourth converter 10b calculates the arm length at which the tension of the wire 66b generates a moment around the ankle joint 26 from the rotation angle indicated value P1, and calculates the arm length from the rigidity indicated value Q1 and the moment arm length. Calculate the wire tension Q2 required to adjust to the indicated stiffness Q1. Here, the spring characteristic shown in FIG. 18 is referred to convert the spring constant and the arm length of the moment into the wire tension Q2 that realizes the spring characteristic that can be adjusted to the specified rigidity Q1. The fifth converter 12b converts the tension Q2 into a motor torque Q3. For this conversion, the characteristic formula of the actuator shown in FIG. 9 is used. Not all of the motor torque is reflected in wire tension. Part of the motor torque is spent on friction and inertia of the actuator and the like. The sixth converter 14b calculates the actually required motor torque Q4 by compensating for the motor torque Q3 that generates the required wire tension by the amount consumed for friction and inertia.
The motor torque Q4 that compensates for the amount spent for friction and inertia is added to the increase / decrease value P5 of the motor torque calculated by the feedback control of the rotation angle. Finally, the added motor torque P6 is commanded. Actually, since the torque of the motor 114b is proportional to the energizing current, a current proportional to the added torque P6 is applied to the motor 114b.
[0071]
The control block is the same for the controller 200c as well, and will not be described again. In this case, since the actuator 68b and the actuator 68c are symmetrical, P2, Q2, and Q3 have the same value with the same sign in the controllers 200b and 200c. However, since the inertia term and the friction term are not symmetrical, the value of Q4 is not equal.
[0072]
According to this control block, the motor torque Q4 necessary for adjusting to the specified rigidity is added. Therefore, the rigidity of the ankle joint 26 around the X axis is controlled to the designated rigidity Q1. The motor torque Q3 for adjusting to the instructed stiffness Q1 (here, the motor torque that becomes the wire tension) is balanced between the actuator 68b and the actuator 68c, and does not affect the rotation angle of the ankle joint 26 around the X axis. . The rotation angle of the ankle joint 26 about the X axis is feedback-controlled to the designated rotation angle P1 irrespective of the motor torque for adjusting the rigidity.
According to the control block of FIG. 22, even if the rotation angle P1 and the rigidity Q1 are independently specified, the control can be performed to the specified rotation angle P1 and the rigidity Q1.
The above-described event is always obtained during the operation in which the indicated values P1 and Q1 of the rotation angle and the rigidity change every moment, and can always follow the indicated values P1 and Q1 that change over time.
In order to change the stiffness instruction value Q1 with time according to the state of walking, it may be made to correspond to the change with time of the walking command data, or the output of the force sensor 322 grounded on the sole of the robot. May be changed based on the In the case of FIG. 19, the value of the rigidity instructed to the motor drivers 306 and 316 is changed by the force detected by the force sensor 322.
[0073]
(Second control technology)
FIG. 23 is a control block diagram of the controller 200b and the controller 200c according to the second control technique. This corresponds to the drivers 306 and 316 in FIG. In this case, the inverse kinematics calculation circuit 304 calculates the target rotation angle for the actuator that adjusts the joint rotation angle, and calculates the target stiffness value for the redundant actuator to adjust the joint rotation angle. In this case, the motor 114b is an actuator for adjusting the rotation angle of the joint, and the motor 114c is a redundant actuator for adjusting the rotation angle of the joint, and adjusts the rigidity.
In this case, the controller 200b performs feedback control of the rotation angle of the motor 114b based on the rotation angle instruction value P1, and the controller 200c performs feedback control of the torque of the motor 114c based on the rigidity instruction value. The actual torque Q6 of the motor 114c is measured by an ammeter 115c that measures a current flowing through the motor 114c. The motors 114b and 114c have a characteristic that the current value is proportional to the torque.
The control block diagram for the feedback control of the rotation angle of the motor 114b is substantially the same as that in FIG. 22, and the control block diagram for the feedback control of the torque of the motor 114c is also substantially the same as that in FIG. .
According to this control technique, a torque required to adjust the rigidity Q1 to the specified one is applied from the motor 114c. With this torque applied, the motor 114b is rotated as necessary to adjust to the designated rotation angle P1. At this time, as a result, the torque of the motor 114b is also adjusted to the torque necessary for adjusting to the specified rigidity Q1, and the rotation angle of the motor 114c is also adjusted to the required rotation angle P1. It is adjusted to an appropriate rotation angle.
The control technique of FIG. 23 uses the redundancy of using two motors for one degree of freedom of the rotation angle, and independently controls the rotation angle and the rigidity. ing. Humans also rotate around joints using a redundant number of muscles in view of the degree of freedom of the joints. The rigidity is adjusted using the redundancy.
According to the control block of FIG. 23, when the rotation angle and the rigidity are independently specified, the control can be performed to the specified rotation angle and the rigidity.
[0074]
22 and 23, the stiffness can be changed in accordance with the walking pattern. The stiffness can be changed when the posture needs to be changed flexibly, and the stiffness can be increased when the posture needs to be maintained. Can be maintained. In the present embodiment, as described with reference to FIG. 21, walking is performed while the rigidity of the grounded foot is reduced and the rigidity of the foot floating in the air is increased.
[0075]
In the present embodiment, the gain k3 for following control and the rigidity of the lower limb are changed according to the walking pattern. It is useful to use only one of the change in gain and the change in stiffness, but it is effective to use it in combination. Behavior can be flexible and stable.
[0076]
When the walking command data is created using the walking command data creating device 1 shown in FIG. 5, usually only the trunk posture vector R is specified for the trunk posture. Indeed, if the trunk posture vector R is specified, the robot can walk smoothly.
However, when walking at a higher speed, it is advantageous to walk while twisting the trunk, and it is advantageous to input not only the trunk vector but also the posture related to the torsion of the trunk.
[0077]
FIG. 24A shows a walking posture in the case of using the conventional teaching technique, and the trunk does not twist even though it may lean forward. In the first place, a vector specifying the torsion of the trunk is not taught.
FIG. 24B shows a walking posture obtained by teaching a trunk torsion vector Q indicating the direction of a line connecting the left and right of the trunk, in addition to the trunk vector R indicating the direction of the left-right symmetry axis of the trunk. Is shown. When the trunk is twisted in conjunction with the movement of the foot, the step length of one step can be increased even if the movement speed of the foot relative to the trunk is the same, and the robot can walk at high speed.
FIG. 24C shows a teaching example of the trunk torsion vector Q, and is viewed in plan. (1) shows a state in which the left and right feet are aligned, (2) shows a state in which the left foot is stepped, (3) shows a state in which the left foot is stepped further, and (4) shows a state in which the right foot is stepped. Is shown.
Since the left toe position vector C and the right toe position vector E are taught, the angle θ formed by the line L connecting the centers of the left and right feet with the line Y orthogonal to the traveling direction X can be obtained. Therefore, the torsion vector Q connecting the left and right sides of the trunk is taught so that the angle between the torsion vector Q and the line Y orthogonal to the traveling direction X is αθ. Here, α is set to 50%. The robot is to be taught with a trunk torsion vector Q in which the angle formed by the trunk torsion vector Q and the Y axis is 50% of the angle θ formed by the line L connecting the centers of the right and left feet with the Y axis. Note that the above ratio is not limited to 50%, but has a width before and after. If the ratio is too small, there is no point in twisting the trunk, and if the ratio is too large, a large amount of energy is consumed for twisting the trunk. Therefore, a range of about 30 to 60% is preferable. As shown in (1) to (4) of FIG. 24C, the ratio α may be kept constant, or the ratio α may be changed in one stroke.
By teaching the torso vector Q to be twisted in conjunction with the movement of the left and right legs, the stride of a robot having the same ability can be increased, and the robot can be walked at high speed.
Instead of specifying the trunk torsion vector Q, the trunk torsion vector Q may be calculated and calculated from the angle formed by the line L connecting the centers of the right and left feet with the Y axis.
As shown in FIG. 25, the upper body 250 of the robot is separated into a waist (lower torso) 254 and a chest (upper torso) 252, and both are configured to be relatively rotatable about a member corresponding to a human spine. be able to. In this case, it is possible to walk while twisting the waist 254, which is the base of the left and right feet, and keep the chest 252 in the traveling direction at all times. Also in this case, the stride length can be extended while keeping the movement speed of the foot relative to the trunk (lumbar region) the same. As shown in the plan view of FIG. 26, the stride can be increased by the distance D by twisting the trunk 254. In this case, the chest 252 is stable, and the walking posture of the robot is stable. In addition, since the stride length increases, it becomes possible for the robot to walk at high speed
When the footprints of the right and left feet are on a straight line 256, not only the walking posture is beautiful, but also it is possible to pass through a narrow passage. FIG. 27 is a plan view showing a case where the left and right feet are walking on a straight line 256 without twisting the trunk, and the left and right feet must swing at an angle α in a horizontal plane with respect to the trunk. Must. FIG. 28 shows a plan view in the case where the left and right feet are walking on a straight line 256 while twisting the trunk. When the left and right feet swing relative to the trunk by an angle β in a horizontal plane. Obviously, when walking while twisting the trunk, the swing angle required to walk the left and right feet on the straight line 256 is reduced, and the robot can walk on the straight line without difficulty. Even if the footprints of the left and right feet are not on a straight line in the strict sense, but on two parallel straight lines, if the distance between the two straight lines is small, the walking posture of the robot will be clear and stable It becomes. By twisting the trunk, the distance between the two straight lines can be easily reduced, and by twisting the trunk, the walking posture can be made clean and stable.
[0078]
In the control device of FIG. 19, a trunk posture vector R is input, inverse kinematics calculation is performed, a joint angle necessary for controlling the input trunk posture vector R is calculated, and a joint obtained in this manner is calculated. In order to perform feedback control using the angle as a target joint angle, the actual trunk vector R matches the instructed trunk vector R, that is, the trunk vector R obtained by the walking command data creating device 1 (FIG. 5). Should be. However, when a gyro was attached to the robot and the actual trunk vector was measured, it was found that they did not match. This is presumed to be caused by play of the joints, bending of the members, and the like.
If the target trunk posture vector does not match the actual trunk posture vector, the robot cannot take the intended trunk posture.
In the present embodiment, the trunk posture vector R (here, the posture that the operator wants the robot to take) created by the walking command data creation device 1 of FIG. If the teaching is performed as it is, the desired trunk posture vector and the actual trunk posture vector match, so that the actual trunk posture vector cannot be made as intended.) Therefore, the target trunk posture vector does not match the actual trunk posture vector. In anticipation of this, the target trunk posture vector is artificially corrected, and the corrected target trunk posture vector does not match the actual trunk posture vector. A matching method is adopted.
[0079]
(1) Specifically, first, the trunk posture vector R created by the walking command data creating device 1 is taught to the robot. The teached robot is operated to detect the actual trunk posture vector R.
Hereinafter, the target trunk posture vector is R1 and the actual trunk posture vector is R2. The actual trunk posture vector R2 is calculated by the device 330 for calculating the output of the gyro 328.
(2) The deviation vector ΔR between the target trunk posture vector R1 and the actual trunk posture vector R2 is calculated (the device 350 in FIG. 19). The displacement vector ΔR is R1 · R2 ^-1 Can be calculated by
(3) Therefore, the target trunk posture vector R1 is corrected in anticipation of the deviation vector ΔR. The corrected desired trunk posture vector is R1 ^* Then R1 ^* = R1 ・ ΔR.
(4) The robot calculates the corrected target trunk posture vector R1. ^* It works with the goal. The actual trunk posture vector realized by the robot is R1 ^* Does not. Deviate from that. As a result, it matches the desired trunk posture vector R1. Corrected target trunk posture vector R1 ^* Is corrected in anticipation of the deviation vector ΔR from the trunk posture vector R1 intended by the operator.
(5) The corrected desired trunk posture vector is R1 ^* When calculating
Y = α · ΔR + (1−α) · I is calculated, and then R1 ^* = R1 · Y. The correction device 352 of FIG. 19 performs this processing. Here, α is a positive value of 1 or less, and I is a unit vector.
In this case, this corresponds to correcting the target trunk posture vector R1 in anticipation of only α% of the deviation vector ΔR.
The correction process of the desired trunk posture vector can be performed at various stages. When the deviation vector ΔR is found, the trunk posture vector R created by the walking command data creation device 1 may be corrected, and the corrected target trunk posture vector may be taught to the robot. In this case, the robot control device does not require the devices 350 and 352 for correcting the target trunk posture vector. Instead, the robot teaches the target trunk posture vector before correction with the deviation vector ΔR, calculates the deviation vector ΔR one after another during the operation of the robot, and calculates the target trunk posture after the deviation vector ΔR is calculated. The vector may be corrected by the shift vector ΔR.
[0080]
As mentioned above, although the specific example of this invention was demonstrated in detail, these are only illustrations and do not limit a claim. The technology described in the claims includes various modifications and alterations of the specific examples illustrated above.
The technical elements described in the present specification or the drawings exhibit technical usefulness alone or in various combinations, and are not limited to the combinations described in the claims at the time of filing. The technology illustrated in the present specification or the drawings achieves a plurality of objects at the same time, and has technical utility by achieving one of the objects.
[Brief description of the drawings]
FIG. 1 is a view for explaining how a target ZMP moves.
FIG. 2 is a diagram illustrating various curves in which a target ZMP that changes sharply is smoothed;
FIG. 3 is a diagram illustrating another example in which a target ZMP moves.
FIG. 4 is a view for explaining that the speed of outputting data of a target ZMP is variable.
FIG. 5 is a diagram showing a configuration of a walking command data creating device.
FIG. 6 is a front view of both lower limbs of the robot according to the embodiment.
FIG. 7 is a side view of the left lower leg of the robot.
FIG. 8 is a view for explaining the structure of the ankle joint of the robot.
FIG. 9 is a diagram illustrating details of an actuator of the robot.
FIG. 10 is a view for explaining movement of a foot of the robot.
FIG. 11 is a view for explaining the movement of the foot of the robot.
FIG. 12 is a view for explaining the movement of the feet of the robot.
FIG. 13 is a view for explaining the movement of the foot of the robot.
FIG. 14 is a front view of a nonlinear spring in which a wire is incorporated.
FIG. 15 is a side view of a non-linear spring incorporating a wire.
FIG. 16 is a diagram illustrating a state where a nonlinear spring is deformed by wire tension.
FIG. 17 is a view for explaining rigidity around joints of the robot.
FIG. 18 is a graph illustrating non-linear spring characteristics and rigidity around a joint.
FIG. 19 is a control block diagram illustrating a configuration of a control device.
FIG. 20 is a diagram showing a pattern for increasing or decreasing a gain for following control following a walking pattern.
FIG. 21 is a diagram showing a pattern in which the rigidity of the lower limb is increased or decreased following a walking pattern.
FIG. 22 is a first control block diagram for independently controlling the joint rotation angle and the rigidity.
FIG. 23 is a second control block diagram for independently controlling the joint rotation angle and the rigidity.
FIG. 24 is a diagram schematically showing a method of teaching a trunk torsion vector and a teaching result.
FIG. 25 shows a posture in which a robot whose upper body is divided into a waist (lower torso) and a chest (upper torso) walks while twisting the trunk.
FIG. 26 is a plan view showing that the step length is widened by twisting the trunk.
FIG. 27 shows a swing angle in a horizontal plane of a foot necessary for walking on a straight line without twisting the trunk.
FIG. 28 shows a swing angle in a horizontal plane of a foot required for walking on a straight line while twisting the trunk.
[Explanation of symbols]
1: Walking command data creation device
2: ZMP operation unit
3: Comparison section
4: Trunk position vector correction unit
5: Completed walking command data
10: Robot
12: Lower limb
14: Thigh
16: Lower leg
18: foot
20: body
22: Hip joint
24: Knee joint
26: Ankle joint
28: Pelvic part
30: Shaft
32: Universal joint
34: Bearing
36: disk
38: Motor
40: Flange
42: Shaft
44: Flange
46: axis
48a, 48b, 48c: Wire termination guide
49a, 49b, 49c: wire connection points
50a, 50b, 50c: wire
52a, 52b, 52c: Actuator
54: Pulley
56: Actuator
58: Flange
60: Flange
62: Cross-shaped universal joint
64a, 64b, 64c, 64d: Pulley
66a, 66b, 66c, 66d: Wire
68a, 68b, 68c, 68d: Actuator (ball screw)
102: Flange
104: movable plate
106: Flange
108: Guide rod
110: Guide rod
112: Guide rod
114: Motor
116: Gear
118: Gear
120: Lead screw
122: flat part
124: Mine
126: flange
128: Shaft
130: Flange
132: Shaft
304: reverse kinematics calculation unit
334: Inverted pendulum model
k1, k2, k3: gain
322: Force sensor
328: Gyro
350: shift vector operation unit
352: Target trunk posture vector correction vector calculation device

Claims (3)

A bipedal robot that walks while twisting the trunk so that the line connecting the left and right sides of the trunk falls within the angle range defined by the line connecting the left and right feet and the line perpendicular to the walking direction. Walk while maintaining a proportional relationship between an angle formed by a line connecting the left and right feet and a line orthogonal to the walking direction and an angle formed by a line connecting the left and right sides of the trunk and a line orthogonal to the walking direction 2 Foot robot. The body that twists the line connecting the left and right of the trunk within the angle range between the line connecting the left and right legs and the line perpendicular to the walking direction in the walking command data that commands the posture of the robot walking while walking on two legs. A method in which the robot walks while twisting the trunk by preparing a trunk twist vector.

Images