JP3429048B2 - Walking control device for legged mobile robot - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is preferably used when a legged mobile robot having joints capable of driving these bodies and arms is walking and it is difficult to continue stable walking due to disturbance. The present invention relates to a technique for restoring stability by changing the gait of the upper body.

[0002]

2. Description of the Related Art As a walking control device for a legged mobile robot, particularly a bipedal legged mobile robot, a weight and a joint for driving the weight are provided on the upper part of the robot, and both weights are driven when walking. A method of canceling the inertial force generated to move the robot has been proposed ("Biped Walking Robot Data Book" (Revised 2nd Edition) Ministry of Education, Science, and Research (A) "Biped Walking Robot Mechanism" And Control Research "research group (February 1986), and" Development of a biped robot that compensates for three-axis moments by body movements "
(6th Intelligent Mobile Robot Symposium, May 1992)
21st and 22nd of the month)). Further, the present applicant also proposes, as one of posture stabilization controls, a technique for landing earlier than the scheduled time to recover the posture when the legged mobile robot encounters an unexpected uneven road surface and leans forward, for example. (Japanese Patent Laid-Open No. 5-254,465).

Disturbance refers to all elements that cannot be predicted at the stage of gait design, such as side winds, unevenness of the road surface, and slippage between the road surface and the sole, but the macroscopic friction coefficient between the road surface and the sole is Sliding due to changes (This macroscopic friction coefficient includes the friction coefficient as a simple physical property value when the entire sole is in close contact with the road surface, but in addition to this, due to unevenness of the road surface, the sole It only touches the ground partly, which causes the friction coefficient to drop and causes slippage).

The legged mobile robot can walk due to the frictional force between the road surface and the sole of the foot. This frictional force must withstand the inertial force generated by moving the entire body when the robot walks. I won't. The inertial force generated by swinging a heavier leg forward and backward is eventually transmitted to the road surface through the sole of the foot, and this becomes the propulsive force and the braking force.
In the legged robot, the left and right legs spatially separated from each other are swung in opposite directions, so that a spin force acts to turn the robot around a vertical axis. If the frictional force is high enough, this spin force is canceled by the frictional force, and the robot can proceed in the designed direction, but if sufficient frictional force cannot be secured, the robot will It rotates around the supporting leg, which causes the stability of walking to be impaired.

However, among the above-mentioned conventional techniques, the technique common to driving by providing a weight on the upper part has a constant gait, and since the upper body is swung in the same manner at all times, the energy consumption is large, However, there is the inconvenience of not being able to respond to disturbances. Moreover, even with the technique proposed by the applicant, it is still not sufficient to deal with an unknown disturbance.

Accordingly, an object of the present invention is to provide a legged mobile robot having a joint capable of driving the upper body, which is capable of driving the joint when an unknown disturbance such as a slip between the road surface and the sole of the foot is encountered. It is an object of the present invention to provide a walking control device for a legged mobile robot capable of surely continuing walking without losing stability.

[0007]

In order to solve the above problems SUMMARY OF THE INVENTION This invention for example as shown in claim 1 wherein, provided with a body joint you driving at least the upper body, is connected to the body that includes a plurality of joints to the legs, the legged mobile robot to walk by driving the relationship <br/> clause of the legs, the
The inertial force generated by driving the legs is
The drive control amount of the upper body joint is determined so as to cancel at least partially by the movement of the.

[0008]

[Operation] The inertial force generated by driving the legs is
At least partially cancel out by upper body movement
In addition, since the drive control amount of the body joint is determined, it becomes easy to determine the drive control amount of the upper body, and when the determination is made using the microcomputer mounted, a relatively low level is sufficient, and The development efficiency of gait data can be improved. In addition, the frictional force is insufficient on the walking surface
Even if the stability decreases, it is easy to secure or recover.
You can

[0009]

EXAMPLES Specific examples will be described in detail below. still,
In the following, the gait not only means a time series change of the joint angles of the legs, but also means a time series change of all joint angles including joints of the upper body and arms. On the other hand, the term "leg gait" is used to mean only the movement of the legs and the term "body gait" is used to mean only the movement of the upper body including the arms.

FIG. 1 is a diagram showing a skeleton and a degree-of-freedom arrangement of a bipedal robot having a preferable mode for carrying out the present invention. As shown, the robot has a structure similar to a human body. The robot has a total of 21 degrees of freedom,
Arms have 2 degrees of freedom in the shoulders, 1 degree of freedom in the elbows, 1 in each of the three orthogonal axes in the center of the waist, 3 in the legs and 1 in the knees, 2 are placed on each foot. These degrees of freedom are configured by a combination of a joint that freely rotates around one axis, a motor that drives this joint, and an encoder that detects the rotation angle of the motor.

More specifically, the central joints 4, 5, 6 and the left and right joints 7, 8, 9 (corresponding to hip joints) constitute a waist portion, and the joints 7, 8, 9 form the thigh link 13. It is connected to the knee joint 10 via the. The knee joint 10 is connected to the ankle joints 11 and 12 via a lower leg link 14, and a foot 15 is provided on the lower portion thereof. Joints in the center of the waist 4, 5, 6
Is connected to a third link 16 which extends vertically above it and a fourth link 17 which is arranged orthogonal to the third link. The joints 1 and 2 are connected to both ends of the fourth link 17 to form a shoulder, and the joints 1 and 2 are connected to the upper arm link 1.
8 is connected to the elbow joint 3, and the elbow joint 3 is connected to the hook-shaped hand 20 via the lower arm link 19. Although not shown, the third and fourth links 1 are provided on the upper portions of the joints 4, 5 and 6 (the portions are referred to as “upper body” in this specification).
6, 17 are covered with a housing.

The symbol m indicates the sum of the inertial mass distributed over the arm, and when the arm is rotated about the shoulder,
A moment of inertia acts on the robot due to this mass.
Since such a configuration is well known in this field and is described in, for example, Japanese Patent Application Laid-Open No. 3-184,782 proposed by the present applicant, detailed description thereof will be omitted. For convenience of explanation, these degrees of freedom are numbered as shown in the figure.
Here, the numbers are serial numbers from the top, and R and L mean right and left. In the following description, this numeral and the symbol R or L will be commonly used in the subscript of the symbol (in principle, the number on the left side of the subscript and the R or L on the right side).

Above the foot 15 at the tip of the leg of the robot, a six-axis force sensor 21 for detecting six pieces of information on forces and moments in the X, Y, and Z axis directions is provided. As shown in FIG. 1, the arrangement of these three axes with respect to the spatial coordinates is such that the traveling direction is set to X, the lateral direction is set to Y, and the vertical direction is set to Z. Moments are also displayed on this axis, and the directions of MX, MY, and MZ are also promised as shown by the right-hand thread display method.

In addition, a tilt sensor 22 for detecting the tilt angle of the robot is provided on a member connecting both legs in the waist region. Inside the sensor, a tilt angular velocity in the front-rear direction, a lateral direction, and a vertical axis rotation. A gyro 22 that detects the rotational angular velocity (yaw rate) when the robot rotates.
a, 22b, 22c are provided at positions orthogonal to each other. These three gyros output an angular velocity signal as an output signal, but if this signal is integrated over time, it can also be extracted as an angle signal. Therefore, for convenience, six signals of θX, θY, θZ, θX dot, θY dot, and θZ dot are provided. Signal is obtained
(In this specification, "dot" stands for differential value). As a specific example of the sensor, a sensor using a vibrating gyro is commercially available at a low cost, and the structure of the sensor itself does not form an important part of the present invention, and therefore further description will be omitted.

FIG. 2 is a block diagram showing an optimum control device for carrying out the present invention. The reference gait generator 30 is a part that generates a gait that can walk a known road surface in a stable manner. When the speed of the computer mounted on the robot is sufficiently high, the computer generates a gait in real time. If the computer speed is not enough to generate in real time, it can be generated offline in advance and the calculation result can be stored in the gait memory 31, and this memory can be used when actually walking. One of the desired gaits may be selected and executed. The gait of a robot is different when walking on level ground or when climbing stairs. Also, when walking on the same level ground, the walking speed and the way it bends are different, so that a practical robot can walk. At times, multiple gaits are generally required. In addition to stable walking, gait setting must also take into consideration walking that minimizes energy consumption and other design criteria, so gait design in real time depends on the ability of the computer. However, in this description, an off-line method that does not have such a problem is illustrated.

In this embodiment, at least two types of gait data are stored: a gait on the assumption that the road surface has high friction for walking on the same route and a gait on the assumption that the road surface is slippery. . The simplest gait for slippery roads is a gait with a slightly reduced walking speed. Other than this, like a human being walking on the ice, a gait in which the hips are slightly dropped and the sole is in contact with the road surface in a wide area as long as possible is also considered, but will not be discussed here.

The environment knowledge base 32 stores data on the environment of the object to be walked by the robot. For example, the arrangement of rooms, the arrangement of stairs, the number of stairs, and the height of one step. -The depth is stored in an easy-to-use form. In this data, with respect to the location of walking, two coefficients called feedforward gain KFF and feedback gain KFB are entered as an index representing the average magnitude of frictional force at the location together with the address of the location. , As will be described later, the robot uses this coefficient to decide how to swing the first arm. Further, in another embodiment, this coefficient is set in association with the gait, which will be described later.

Since the output of the inclination sensor 22 is analog in this example, it is converted into digital once and then input to the central processing circuit 35. Since the output signal of the 6-axis force sensor 21 is also analog, it is converted to digital and also input to the central processing circuit. The MM interface 36, which is an interface between a human and a robot, includes a CRT that displays the input / output status of the robot, a keyboard that transmits commands, and a gain setter that sets the gain. When setting the gain with software, you may enter it with the keyboard.

The central processing circuit 35 selects an optimum gait on the basis of all these inputs, and issues an appropriate current command to the drive motors arranged at each joint in order to execute the gait. Two motor drive circuits 37, 38 are shown in FIG.
, And the remaining motor drive circuits are shown in black box form. The number is 21 in total in this embodiment. Display drive circuit is shoulder joint 1R / L
This is for the freedom of swinging back and forth. To briefly explain one of the drive circuits, the current command value is output as a digital signal, so this is converted to an analog value by the D / A conversion circuit, then amplified by the amplifier Amp, and the motor jML / R Is rotated (j means a joint number). The rotation angle of the motor is detected by the encoder Enc, counted by the counter, and fed back to the central processing circuit 35.

FIG. 3 and FIG. 4 are flow charts for explaining the operation of the above-mentioned control device. Before the description of FIG. 3 is started, the walking control according to the present invention will be outlined.

The feature of this control is that it recovers from an unstable condition by utilizing inertial force and continues to walk stably again. Although there is a free leg as an inertial force available, it is desirable. If you have a joint that is not directly related to walking, such as your torso or arm, driving this joint when it is unstable can reposition itself like humans do. However, this human behavior is reflexive, and it is not clear what kind of control law is used. By using the technology already known, it is preferable that the robot behaves like a human as a result, the stability during walking is ensured, and the stability is threatened by disturbance. A technical method for restoring stability by driving the joints in a desired direction at a desired speed and in a desired order has been awaited.

Therefore, one of the characteristics of this control is to increase the stability of the free joint in the direction of restoring the stability when the stability of walking is about to be impaired. It is to drive at such speed to restore stability.

Further, as to the inertial force, as described above, the frictional force against the spin force generated by the drive of the free leg is determined by the road surface on which it moves, so the only way to walk is with the inertial force that matches the frictional force. One method is to adjust the walking speed, and the other method is to walk with the same speed but with a reduced inertial peak value.

By the way, when observing the walking of a person, both hands are oscillated in the opposite directions in synchronization with the movements of both legs.
Due to the inertial force caused by the movement of both hands, a part of the inertial force of the both legs is canceled. From this finding, as a third method for dealing with the fixed frictional force, there is a method of canceling a part of the inertial force of the leg by utilizing the inertial force generated by moving the upper body. Even in a legged robot, if there is a mass that can cancel the moment generated by the movement of the legs and a joint that can drive this mass, the mass is driven in synchronization with the walking cycle, and The same effect can be obtained by configuring the parts to be canceled.

In this case, what is important is that it is not necessary to cancel all of the inertial force of the leg. If the inertial force of the leg can be reduced and the frictional force of the road surface falls within the range, the purpose is achieved. It is possible. Generating an inertial force that cancels more than necessary becomes a drawback in a robot that relies on an on-board energy source for all energy.

When reproducing the movements of both hands of a human on a robot, the muscles that are human actuators,
It is also a point to consider that the operating principle is different from that of a robot, and that imitating a human being is not preferable in terms of energy consumption. That is, since the motor, which is the actuator of the robot, is of a high speed type, a reducer having a considerably high reduction ratio is interposed in use. Shaking both hands with this combination of motor and reducer means continuously driving the motor at high speed repeatedly in forward and reverse directions.In addition to the net output for moving both hands, energy loss consumed internally is high. Too much. In addition, human muscles naturally return their hands when they lift their power, but with a combination of a motor and a speed reducer, it is necessary to drive the motor even when returning. It turns out that we cannot imitate movement.

When swinging the upper body of the robot, especially both arms in conjunction with the legs, the purpose is to cancel the inertial force generated by the movement of the legs, so it is most reasonable to swing the arms. That is, the phase of the inertial force generated by the movement of the arm is opposite to that of the leg, and the shape of the inertial force is similar. It is also desirable that the arm can be easily moved and the burden on the on-board computer is small.

Furthermore, when reconsidered from the viewpoint of minimizing the energy consumption due to shaking the arm, the energy consumption becomes the smallest when the arm freely vibrates around the shoulder.
At this time, the motor needs to rotate itself while following the free vibration of the arm, but the output to the outside can be minimized.
This method may raise the question that it may not be possible to respond to changes in the walking speed (cycle), but fortunately there is an elbow joint, so bend this joint appropriately and shoulder By adjusting the inertial mass of the entire arm viewed from the joint of, the period of free vibration can be made to match the period of walking.

It is not desirable from the viewpoint of energy consumption that the arm is swung regardless of the condition of the road surface to decide how to swing the arm. Shaking your arms is nothing more than a waste of energy if you have high enough friction. Further, when the road surface is slightly slippery, it is not desirable to swing the arm with a large amplitude for the same reason.

Specifically, if the person is walking while swinging his or her arms and if slippage occurs, it is judged that the current swinging method is insufficient for the frictional condition of the road surface, and the swinging is performed with a larger amplitude. To configure. On the other hand, if there is no slip when walking with the once determined amplitude, the arm amplitude is slightly decreased after walking a predetermined distance to save energy consumption. More preferably,
Initially, the user can walk without swinging his arms, and by swinging his arms only when slippage occurs, he can walk to further save energy. Even in this case, when the user can walk the predetermined distance without shaking by swinging his arm, the swinging method of the arm is changed again to the direction in which energy consumption is reduced, and the validity of the current swinging method of the arm is verified. This is because the coefficient of friction of the road surface is changing, and it is not always true that the previous decision to shake the arm due to slippage is correct.

Further, when the robot walks while carrying the luggage in its hand, the load of the luggage strongly presses the sole of the foot against the road surface, so that the slip does not naturally occur. In particular, when carrying the luggage in both hands, the center of gravity of the luggage is located on the outermost side of the robot, so that the inertial mass around the vertical axis of the robot also increases and it becomes difficult to rotate. In response to this change, it is necessary to change the way of swinging the arm.

Furthermore, when a person's gait including the upper body is walked and a slip occurs, it is necessary to restore the posture of collapse due to the slip as an emergency measure. The above-mentioned gait, in which the arm is steadily shaken, is a behavior that places emphasis on reducing or eliminating slippage from now on, and what is used to make an emergency avoidance behavior and restore the collapsed posture is control. Must be changed. Regarding the behavior of the upper body for emergency avoidance, how to connect to a steady arm swing after the emergency avoidance exercise becomes a problem.

Furthermore, what is important in deciding how to swing the arm is what timing to change the swinging method of the arm. The reason why this is a problem is, for example, when changing the swinging direction of the arm to a direction that decreases, if the change command is issued when the arm is raised high, the next target angle becomes smaller. The reason is that, although the movement of the arm becomes sharp and the robot walks at a stable angle, it creates an unstable upper body, which deviates from the original purpose of the present invention.

Furthermore, in order for the robot to intelligently move around the environment, a method of utilizing a knowledge base about the environment has been proposed. This knowledge base includes indoors, along with shapes about the environment, such as flatlands, stairs, steps, doorways, etc.
If the items related to the coefficient of friction such as outdoor items related to the weather and slippery plastic style or non-slip concrete road surface are entered, the robot can be operated even if the operator does not teach how to swing the first arm. You can decide for yourself.

Further, although the above-mentioned coefficient is entered by the designer at our office, since the robot actually walked to optimize the coefficient, it is desirable to use this coefficient thereafter. Of course, there are changes over time in the environment, and apart from changes over time,
The friction coefficient may change due to splashing of water or accumulation of dust, but if the latest data is recorded as the coefficient, the credibility of the coefficient increases.

Referring to FIGS. 3 and 4, an example of an algorithm for driving the joints of the robot is shown in FIG. Here, it is assumed that the robot is instructed by the operator to walk and move from point A to point B in a known environment. The robot takes out map information from the environment, searches for a route based on its current position, and divides the entire route into parts that can be walked with a single gait. For example, stairs and thresholds require different gaits, so they are divided into different areas. The illustrated algorithm exemplifies a case in which, for example, the ascending / descending of stairs is completed, a flat ground section is entered, and the section is walked with a single planned gait.

In FIG. 3, the operation starts in step 1, the device is initialized in step 2, and the process proceeds to step 3 to read the setting gains K1 and K2 used in step 21 later. Subsequently, in step 4, what kind of environment the walking environment is like, and the attribute value of the environment is recognized by referring to the environment knowledge base. Geometrical environment maps and brightness are entered as the attribute values, but as the values related to the present invention, the magnitude of friction on the road surface and the presence or absence of cross winds at locations where blowers are installed, etc. Is described together with the travel route information. However, in order to avoid complications, only those relevant to the invention will be described here. In the case of FIG. 3, a feedforward correction gain KFF is prepared according to the magnitude of friction, and a certain number is entered as the gain KFF for each walking target section. The gain K FF is large on a road surface with low friction and small on a road surface with high friction. Initially, the designer gives this gain empirically and fills it in. That is, in the case of this embodiment, it is premised that the degree of friction is known or can be predicted. This is the case for many indoor work spaces and the like.

Then, in step 5, gait data for walking on a flat ground is read from the gait memory. Here, of the two types of gait data described above, gait data that is premised on high road friction is read. The gait data is a time series table of each joint angle for both two types, and the joint angle at the time t of the joint j is represented as jαt. In step 6, the gain (shoulder joint angle correction gain) KFF is selected (set),
In step 7, the actual encoder value jCR / of each joint motor
L is read, and the measured values of the tilt sensor 22 and the 6-axis force sensor 21 are read in step 8.

In step 9, it is determined whether the yaw rate has exceeded a predetermined first threshold value THθZ1 dot which is considerably large from the state where one foot floats in the air while walking to the present. The yaw rate should not occur when the robot is moving straight ahead, and if a large yaw rate occurs, it must be expected that the stability will be impaired. However, if the gait itself is designed to drive the joint 7R / L to generate the yaw rate, such as when the robot turns a corner, the stability is determined only by the information from the tilt sensor 22. It is not accurate to do.
Therefore, specifically, in step 9, the difference (absolute value) between the yaw rate DSRDθZ dot that should naturally be generated from the gait and the detected yaw rate θZ dot is calculated, and compared with the first threshold value THθZ1 dot. Judge.

When it is determined in step 9 that the yaw rate difference value is less than or equal to the first threshold value, the process proceeds to step 10. Similarly, in step 10, the difference value (absolute value) between the yaw rate DSRD θZ dot and the actual yaw rate θZ dot
Is calculated, and the difference value is compared with the first threshold value to make a small second value.
The threshold value THθZ2 dot is compared and it is determined whether or not the threshold value is exceeded. If the difference is equal to or less than the second threshold value, even if there is a slight difference between the expected value and the measured value, it is due to mechanical play such as reduction of the motor, minute slip due to vibration during walking, etc. The road friction is not lower than expected. However, if the difference value exceeds the second threshold value in step 10, it is considered that significant slippage has occurred, the frictional force of the road surface has lost the margin, and it is determined that stable walking cannot be continued as it is.

Then, after proceeding to step 12 (FIG. 4) through step 11 (described later), the amount increased by 10% to compensate for the shortage of the previously selected correction gain KFF is substituted so as to be newly used as the correction gain KFF. Do the work. this
If the yaw rate difference value exceeds the second threshold in the next program startup, the 10% increase is still considered insufficient, and the gain increased by 10% is used again. However, if you continue to increase the number of times, there is a possibility that the value will become endlessly large, and such a situation should not occur in reality, so the correction gain KFF should be stopped with the upper limit of 50% of the initial setting value. There is.

However, the timing is required to start swinging the arm with the changed correction gain, and if it is changed carelessly, it will be an unstable state, so either the left or right leg may be used, but the angle of the waist joint 8R / L We decided to change it only when is sufficiently small (that is, when the arm is at the lowest position). Step 11 is a determination step for this purpose, and while this condition is not satisfied, the routine proceeds to step 14, and the gain KFF (t-1) used at the time of the previous control is set as the current gain (at the time of the first program start, at step 6 When this condition is satisfied for the first time together with the selected gain KFF (t-1), the process proceeds to step 12.

If it is determined in step 10 above that the yaw rate difference value is less than the second threshold value and a significant yaw rate does not occur, it is not necessary to change the correction gain for the time being. Although it follows also used (step 14), after walking step count N, since it is unknown whether proper from the viewpoint of energy saving to continue swinging the arms gain really remain current, the gain KFF to try
Reduce by 10%. Step 13 is the decision step for this trial, and if so, step 15
After going through (similar to step 11), the process proceeds to step 16.

If the gain KFF is simply decreased by 10%, the gain KFF becomes infinitely small and eventually becomes 0. Therefore, in order to prevent it, the minimum value of KFF is set to the initial value (selected in step 6). It is set to 1/2 and not to be less than that. Even if the gain is reduced by 10%, if it is not determined that significant slippage has occurred in the next control cycle in step 10, this change is determined to be correct, and then N steps are walked with the same gain. However, if it is judged that this change causes a significant slip, the gain is increased again by 10%, and the same level as the original gain is restored.
Even when the gain is reduced, step 15 is prepared because it is necessary to change the timing.

In step 17, the correction gain KFF thus determined is used to calculate the shoulder joint correction value ₁ αt to be given to the motor of the shoulder joint 1 R / L. In this embodiment, this correction value is obtained by multiplying the swing target angle of the hip joint 8R / L by the above gain. Since the arms and legs are out of phase, the signs of the angles are reversed between the arms and legs, and the left hip angle is used to calculate the right arm angle.

Next, at step 18, the encoder target value ₁ COMPCR / L corresponding to the shoulder joint angle correction value of this arm is calculated. This is because the origin of the encoder does not always match the zero point of the joint angle. Then step 1
In 9, add the target value ₁ COMPR / L obtained by the above calculation to the encoder command value DSRDCR / L (constant when not inevitable) when driving the arm for a reason other than the cancellation of the leg inertial force. Use as command value. In step 20, the difference value ΔjCR / L from the previous time of the encoder is calculated, and in step 21, the setting gains K1 and K2 read in step 3 are used, and the torque value (current) proportional to this difference value and the joint angle The total torque value proportional to the difference from the previous time is used as the torque command value for the shoulder joint 1R / L motor. Only the torque value proportional to the difference value of the joint angle from the previous time is used as the motor torque command value for joints other than the shoulder joint 1R / L. Then, in step 22, a command is issued to each joint motor. If the user wants to continue walking, the procedure returns to step 7, but at the end, step 24
Then, the program ends by updating the correction gain KFF in the knowledge base of the environment to the latest value. Therefore, when walking in the same section at the time of starting the program after the next time, this latest correction gain KFF is used from the beginning.

On the other hand, when it is determined in step 9 that the yaw rate difference value exceeds the first threshold value, that is, when a large slip has occurred, the routine proceeds to step 26, and when the two-leg support period is reached, step 27 Then, the gait data for low friction is changed from the two types of gait data. That is, when the threshold value is less than the first threshold value and exceeds the second threshold value, the correction gain is changed to change the amplitude of the arm swing, and when the threshold value is exceeded, the gait data itself. I changed it. In the example, the gait data for low friction (specifically, the gait data for low-speed walking) is set in advance, but as previously proposed by the applicant, It can also be realized by reducing the output speed of the gait of the above, and is technically easy to realize.
Further, the fact that both legs are in the supporting period can be easily discriminated because the output FZ of the 6-axis force sensor 21 increases.

As described above, this embodiment is constructed so that both arms are shaken from the start of walking to cancel a part of the inertial force of the legs. The amount of cancellation is planned and decided in advance based on the environmental knowledge base. That is, it is possible to determine a gait for continuing stable walking on the assumption that the degree of friction is previously recognized or can be predicted, and further energy can be obtained even when walking stably. To reduce the consumption of. Moreover, since the swing of the arm is determined based on the swing of the leg (specifically, the waist joint), it is possible to easily create a body gait with a small load on the computer. As a result, it is only necessary to sufficiently design the gait that pays attention to the movement of the legs, and the development efficiency of gait data can be improved.

Further, with the above structure, an optimum solution can be obtained for the trade-off problem between ensuring stability by shaking the arm and reducing energy by not shaking the arm. The robot can judge whether the current selection is correct, increase the amount of swing if the arm is not swinging enough, and even if it is enough, it can be judged whether it is enough or not too much, It is possible to walk with stable and low energy. Further, the yaw rate difference value is compared with the threshold values of two types, large and small, and when it is determined that a large slip has occurred, the gait data can be changed to have a high margin of stability such as decreasing the walking speed.

Even when the correction gain KFF is changed according to the current situation, the range of change is limited (± 50% in this example), so that the gain may increase or decrease indefinitely. If there is not this, especially the braking in the decreasing direction becomes 0 at the end, and the updated value of the environmental knowledge database becomes 0 at the end of the program, but there is a possibility that the arm cannot be shaken at the next walk. , Such inconvenience does not occur. Furthermore, since it has been decided to change the timing when the upper body gait is least likely to be a disturbance, it is possible to avoid an unintentional situation in which the stability is rather compromised with the intention of ensuring stability.
Furthermore, the gain K that converges as a result of walking in the knowledge base of the environment
By rewriting the FF, it became possible to always preset the optimal way of arm swing. In other words, the designer can set the initial value of the gain KFF with insufficient accuracy, and the burden on the designer can be reduced.

5 and 6 are flow charts showing the algorithm of the second embodiment according to the present invention. The mechanism and control device targeted by the algorithms of the second embodiment and the following embodiments are basically the same as those of the first embodiment.

In FIG. 5, steps up to step 105 are the first
There is no difference from the example. In step 106, shoulder joint 1R /
The first feedforward correction gain ₁ KFF for determining the angle target value of L and the second feedforward correction gain ₃ KFF for determining the angle target value of the elbow joint 3R / L.
And are read from the environmental knowledge base (where subscripts 1 and 3 are
And are matched with the joint numbers, respectively. ). Next, at step 107, the present counter value of the encoder jCR /
L is read, and the sensor output is read in step 108.

In step 109, the yaw rate difference value is compared with the second threshold value to determine whether significant slippage has occurred since the one-leg support period, and if it is determined that no significant slippage has occurred, the correction is made. It is determined that the gain KFF is appropriate, and the process proceeds to step 111 through step 110 and the previous gain KFF.
Use FF (t-1). Here, KFF without a subscript shall represent the 1st and 2nd gain as a representative. However, as in the first embodiment, the gain may be too large, so it is judged in step 110 whether the predetermined time T has elapsed, and if it is judged that the predetermined time T has elapsed, step 1 is executed.
Go to 13 and try with a gain that is 20% less than that.
However, if the gain is already 80% of the initial value, the previous value is held. In the second embodiment, as in the first embodiment, the timing of change is selected in step 112. In the second embodiment, the shoulder joint angle ₁ αt is used as the timing judgment index instead of the leg waist joint angle ₈ αt.

On the other hand, when it is determined in the previous step 109 that significant slippage has occurred, it is determined that the current way of swinging the arm is not sufficient, and step 11 is executed via step 114.
Go to 5 and try again with a gain 20% higher than the previous gain. Also in this case, when the previous gain has already used the largest gain, the previous gain is used as it is. In the control cycle after the next time, step 109
If it is determined that the stability has been restored without the occurrence of a significant slip in (1), the same gain is used again (step 111). When the gain is changed in step 115, the timing is adjusted in step 114.

Then, in step 116 (FIG. 6), the correction values of the shoulder joint angle and the elbow joint angle are determined using the two gains thus determined. In steps 117 and 118, this correction value is converted into an encoder value to be the target value of the encoder, in step 119 the difference value between the current value of the encoder counter value and the target value is obtained, and in step 120 the shoulder joint, elbow joint, etc. The torque command value to the drive motor of is calculated (the torque command value to the drive motors of the remaining joints is only the first term of the illustrated formula). Then it is step 12
Output at 1. Updating the value of the gain of the knowledge base of the environment in step 123 prior to the end of walking is the same as in the first embodiment.

In the second embodiment, as compared with the first embodiment, the elbow joint is also moved in reverse phase in conjunction with the knee joint, so that the inertial force generated by the movement of the leg joint can be more accurately measured. You can cancel with your upper body. This method can improve the stability more than in the first embodiment in walking in which the knee joint is greatly moved, such as when going up and down stairs.

Further, the correction gain is adjusted according to the occurrence of slippage.
Since the upper and lower portions are prepared by 20%, it is not necessary to set a limit to increase or decrease the correction gain as in the first embodiment. Further, in the first embodiment, the control is performed by the geometric series that is changed by the "ratio", but the drawback of the geometric series is that the effect of the change becomes smaller gradually as it is reduced. is there.
In the present embodiment, since the change is made in the arithmetic series, even if the number of changes is increased, a certain effect can be expected. In the second embodiment, an example is shown in which the original gain and the correction gains 1.2 KFF and 0.8 KFF, which are increased and decreased by 20% respectively, are set. You can change to.

Further, it has been shown that the same effect can be obtained not only by using the movement of the hip joint as an index but also by the movement of the shoulder joint in determining the change timing in steps 112 and 114. Further, in the first embodiment, whether or not the user has walked N steps is used as the trigger for the next trial, but in the present embodiment, the trial is triggered at time T. Everyone gets the same result. Another candidate for the trigger may be the moving distance. Although the first threshold value is not used and the gait data itself is not changed, the gait data itself may be modified as necessary.

FIG. 7 and FIG. 8 are flow charts showing the algorithm of the third embodiment of the present invention, and show an example most suitable for application when carrying luggage (cargo) in at least one hand.

When the robot carries a baggage (cargo) in its hand, if the arm is shaken without accurately recognizing the weight of the baggage, the stability may be deteriorated. Considering the stability problem when carrying luggage, the inertial mass of the legs is not relevant, since the inertial mass of the legs does not change at all. However, since the weight definitely increases, the increased weight works to hold the sole of the foot firmly on the road surface, so it is more slippery when carrying luggage than when not carrying luggage. I understand. In other words, when carrying luggage, the stability originally increases, so it is necessary to change the control of swinging the arms as compared to when carrying luggage.

Explaining with reference to the above, in FIG. 7, the process proceeds to step 205 after performing the same steps as those in the previous embodiment, where the gain R / L KFF for determining the joint angle correction values of the left and right shoulders is set to 2 Read one. The two gain values are the same, but here, for the purpose of the subsequent processing, here are shown cases prepared separately. In step 206, it is determined whether or not the weight of the package (payload) is greater than the negligible amount ΔW, and if it is determined that the weight is less than the negligible weight, the subsequent processing is basically the same as in FIG. 3 of the first embodiment. The same as step 7 in step 7 above. If the amount exceeds the non-negligible amount, the process proceeds to steps 207 and 210 to determine whether the user is carrying both hands or one of the left and right hands.

Here, the above-mentioned discrimination method will be explained. The output of the 6-axis force sensor 21 of the ankle is read while the robot is standing on both feet. The posture of the robot at this time is arbitrary, but in order to save the trouble of calculating the distances S and L shown in FIG. 9 from the joint angles, in the following calculation, the width of the foot is the same as the width of the hip joint, and Is assumed to be standing vertically down. Then, the distance between the two reaction forces R / L FZ acting vertically upward on the ankle becomes the space 2L between the waist joints, as shown in FIG. Weight of luggage hanging on both arms
The R / L W interval is also the shoulder joint interval 2S.

In addition to luggage, the robot has its own weight WRO.
Since there is B, the position of the center of gravity at this time is assumed to be in the center of the robot for the sake of convenience in the following calculation, assuming that the robot is weighted evenly on the left and right sides (of course, it can be easily calculated even if they are not equal). LFZ + LFZ = WR because the downward force due to own weight and luggage balances the reaction force acting on the ankle.
OB + RW + LW Also, considering the balance of the moment around the right ankle (or left ankle), RW (S-L) + LFZ (2L) = WROB x L + LW
(L + S). From these two equations, the two unknowns R / L W are RW = 1/2 {RFZ (1 + L / S) + LFZ (1-L
/ S) -WROB} LW = 1/2 {RFZ (1-L / S) + LFZ (1 + L
/ S) -WROB}.

In order to determine the hand that is not carrying the luggage, the calculated value of RW or LW should be substantially 0,
It does not necessarily have to be 0. RW calculated from measured values
Alternatively, the accuracy of LW depends on the measurement accuracy of the 6-axis force sensor. It may be negative depending on the accuracy. Specifically, the smaller value of RW and LW is compared with the reference value ΔW (appropriate setting), and if smaller than that,
It is determined that the corresponding hand is substantially carrying no baggage. Specifically, if both RW and LW are ΔW or less, the process proceeds to step 7 in FIG. 3, and if either of them exceeds ΔW, the process proceeds to step 207 and thereafter. And step 2
If both RW and LW exceed ΔW at 07, it is judged that both hands are carrying luggage. When the result is negative, the routine proceeds to step 210, where a hand that is not carrying luggage is determined.
The operator may teach this determination through the interface. It is also possible to attach a 6-axis force sensor to the wrist of the robot and then judge.

Subsequent processing differs depending on which hand is carrying the luggage. Step 20
The gait data of the legs separately set for carrying luggage is read at 8, 211 and 213.

Step 2 when carrying luggage in both hands
At 09, the gain is replaced with 0 so that the arm is not shaken as a result. The reason for this is that when carrying a bag of significant weight in both hands, the weight of the bag is added to the sole of the foot to make the friction force sufficient, and the bag is located on the outermost side of the robot, This is because the inertial mass around the vertical axis is sufficiently large and the amount of rotation is reduced even when the same spin force acts.

When it is judged to be carried with one hand, the arm carrying the luggage is not shaken. Instead, the correction gain is doubled in steps 212 and 214, and the free arm is doubled. Shake with the amplitude of. Thereafter, according to this, basically the same processing as that described in the first or second embodiment is sequentially performed in steps 215 to 235 (FIG. 8). If it is determined in step 217 that significant slip has occurred, the process proceeds to step 218. If it is determined that both hands are carrying luggage, the process proceeds to step 220 as soon as it is confirmed in step 219 that both legs are supported, and the gait data itself is obtained. To change. This is because slippage occurred despite the increased stability of carrying the luggage with both arms.

As described above, the third embodiment is configured to change the way of swinging the arm when carrying luggage in the hand. That is,
The arm on the side carrying the luggage should not be shaken, and when only one arm is shaken, it should be shaken with double amplitude. As a result, the number of swinging arms is reduced, and energy consumption can be reduced accordingly. Also, when significant slippage occurs while carrying luggage in both hands, gait data itself is changed to ensure stabilization. Note that the gait data for carrying luggage is not set separately,
The same gait data may be read by 8, 211 and 213.

10 and 11 are flow charts showing the algorithm of the fourth embodiment of the present invention.

As described above, if there is a sufficient friction coefficient and stable walking is possible, it is better not to shake the arm which consumes energy. The fourth is based on this idea
This is an example.

Explaining below, in FIG. 10, the same steps 301 and 302 as those of the previous embodiment are performed, and then step 3 is executed.
In step 03, the three setting gains including the third gain K3 are read, and in steps 304 and 305, the process proceeds to step 306, in which the feedback correction gain KFB (described later) is used as the shoulder joint angle correction gain in addition to the feedforward correction gain KFF. ) Is selected (set).

Subsequently, the process proceeds to step 310 through steps 307, 308 and 309, and if it is determined that significant slippage occurs, the process proceeds to step 311 to determine whether or not the arm is currently swung. If it is judged in step 310 that significant slippage has occurred while shaking,
In order to increase the amplitude, the routine proceeds to step 312, where the previous correction gain is increased by 10% and a new gain is set (the upper limit is 50% of the initial gain. When changing the gain, the timing is set as in the previous embodiment. May be).

However, if it is determined in step 311 that the arm is not currently swung, the arm is swung, but before that, the process proceeds to step 313, where the current free leg lands and the first two-leg support period. To determine if you have reached. If it is determined that the period for supporting both feet has not yet been reached, the process proceeds to step 314, where the coefficient β is set to 0.8. This coefficient will determine the following speed of the motor as will be described later, and setting 0.8 here results in the arm swing following the target angle relatively slowly, and the impact at the time of change is set. This is to alleviate. If it is determined in step 313 that the two-leg support period has been reached, there is a margin of stability even if the impact at the time of change is somewhat large.
Set to 1.0 to increase the tracking speed. When the selection of the coefficient β is completed, the process proceeds to step 316, and the value of the gain KFF is set to the previously set amount.

If it is determined in step 310 that a significant yaw rate has not occurred, the process proceeds to step 317.
Therefore, it is determined whether or not the arm is currently shaken. If the result is negative, the stability is maintained without shaking, so the process proceeds to step 324 (FIG. 11) to set the shoulder joint target angle correction value to 0.
To On the contrary, if the affirmative determination is made in step 317, the current setting value is understood to be proper, and therefore steps 318, 31
During the subsequent N steps through 6, the gain uses the previous gain as it is. When it is determined in step 318 that the number of steps N has been exceeded, the process proceeds to step 319, in which the gain is slightly reduced as in the previous embodiment, and the state is observed.

The difference of this embodiment from the previous embodiment is that in the process of decreasing the correction gain, if the gain becomes less than half the initial value read from the environmental knowledge base, the arm swing is stopped. That's what I did. That is, while the result of reducing the previous gain by 10% in step 319 exceeds half of the initial value, this 10 is the same as in the previous embodiment.
The% reduction amount is set as a new gain. If the reduction result is reduced to less than half, the process proceeds to step 321 to fix the gain to the previous value, and at step 322, the shoulder joint angle 8R /
Wait for the timing when L becomes the smallest (along the body with the arm lowered) and proceed to step 323 (Fig. 11) to set the coefficient β to 0.8, then proceed to step 324 and set the correction value ₁ R /
Let L αt be 0. At the same time, the angle cαt for correcting and driving the shoulder joint in proportion to the yaw rate is also set to zero.

On the other hand, after determining the necessary correction gain KFF in steps 312, 316 and 320, step 325
The joint angle correction value of the shoulder is calculated in (FIG. 11). Specifically, the angle correction value when the target angle of the hip joint 8R / L is multiplied by the feedforward correction gain KFF and the arm is steadily shaken
₁ R / L αt is calculated, the yaw rate difference value is multiplied by the feedback correction gain KFB, and two types of correction values of the correction angle cαt for swiftly swinging the arm with a size proportional to the yaw rate are calculated. The latter loses its substantial effect when the yaw rate disappears.

Subsequently, the process proceeds from step 326 to step 328 to step 329 to calculate the motor torque command value.
As shown in the figure, the command value for the shoulder joint 1R / L is obtained by multiplying the target joint angle difference value of this time and the previous time by the gain K1, and the encoder difference value by the coefficient β and the gain K2. It is calculated by obtaining the sum of the shoulder joint angle correction value multiplied by the gain K3 (the positive and negative sides are reversed from the phase relationship). In this way, the motor torque command value is multiplied by the coefficient β, so if β is 1.0, current is supplied so as to follow the measured counter value, and β is 0.
When it is 8, the following speed becomes slow and the movement of the arm becomes slow. The motor torque command value for the remaining joint is the sum of the first and second terms. The steps after step 330 are basically the same as the previous embodiment.

As described above, the fourth embodiment is basically constructed so as to walk without swinging the arm to reduce energy consumption. That is, only when significant slippage is detected, the arm is shaken to restore stability. In that case, the emergency response (steps 311 and 315) and the subsequent recurrence prevention response (step 312) are taken, and the emergency response is swift and the recurrence prevention response is step 317. Through 324 through slowly. Thereby,
It is possible to reduce the side effects caused by corresponding measures.

Further, in the fourth embodiment, when the gait becomes stable, the recurrence preventive measures are sequentially reduced to reduce energy loss, and finally the arm shaking is stopped. When discontinuing, make the position where the arm stops along the body side,
Be prepared to respond to the next disturbance again. In addition, the movement to the stop position is also performed slowly to minimize the impact caused by the stop. Furthermore, a new concept coefficient, β, was introduced to simplify the algorithm and maintain the computational speed of the computer.

12 and 13 are flow charts showing the algorithm of the fifth embodiment of the present invention.

The fourth embodiment has shown a case where, based on the fact that the arm is not shaken, an emergency measure and a permanent measure for preventing the recurrence thereafter are taken from the time when the phenomenon caused by the disturbance occurs. What should be done when a disturbance occurs and an emergency measure needs to be added while swinging the arm as in the first embodiment (FIGS. 3 and 4)? The fifth embodiment shown in FIGS. 12 and 13 describes a countermeasure therefor.

Explaining below, if it is determined in step 410 in FIG. 12 that significant slippage has occurred, step 4 is performed.
After step 15, the process proceeds to step 416 to perform the recurrence prevention process, and the yaw rate difference value is multiplied by the feedback correction gain KFB to obtain the correction angle cαt related to the emergency process.
17 (FIG. 13) and step 4 based on them
At 21, the motor torque command value is calculated.

The fifth embodiment is obtained by adding a shoulder joint angle correction value in proportion to the yaw rate to the structure of the first embodiment (FIGS. 3 and 4). From the four examples and the fifth example, it can be easily understood that the recurrence prevention strategy and the first aid strategy can coexist and coexist without conflicting with each other.

14 and 15 are flow charts showing the algorithm of the sixth embodiment of the present invention.

The control when carrying the luggage is described above, but a method is proposed in which even if the luggage is not carried in the hand, from the viewpoint of energy consumption, it is decided whether to shake both hands or only one hand. To do. The algorithm illustrated in FIGS. 14 and 15 is an example in which the designer decides in advance which of the right hand and the left hand to shake when shaking one hand. In the example shown, the right arm is first shaken, and then both arms are shaken if it is still insufficient (slippage occurs).

Explaining below, the steps up to step 508 in FIG. 14 are basically the same as those in the previous embodiment. In steps 509 and 510, it is determined whether or not the arm is currently waving, and if so, whether one hand or both hands are being waving. The following processing differs depending on the result of this judgment. That is, when the answer in step 510 is negative, the arm is not swinging at all,
If it is not determined in step 511 that significant slippage has occurred, the process proceeds to step 516 (FIG. 15), and the coefficient R / L γ (determines arm swing as described later) is set to 0 in order to control not to swing the arm. Set.

If it is determined in step 511 that significant slippage has occurred, the process proceeds to step 512, and until it is confirmed that the both-feet support period has been reached for the first time, in step 514 the coefficient β is set by recurrence preventive measures. Is set to be smaller than 1, for example 0.8. After it is confirmed in step 512 that the first both feet support period has been reached, the process proceeds to step 513, where the coefficient β
Raise to 1.0 to increase the tracking speed to the target joint angle. After that, the process proceeds to step 515, where the value R γ for the right arm is set to 1 among the coefficients R / L γ that determine to swing the arm, and the value for the left arm is set because the left arm does not swing.
Set Lγ to 0.

If it is determined in step 510 that one arm is currently waving, the current control is considered to be appropriate unless it is determined in step 517 that significant slippage has occurred. Therefore, steps 522 and 515 (see FIG. In step 15), the state is maintained up to the number of steps N. However, when it is determined in step 522 that the number of steps N has been exceeded, the arm swing is stopped,
After proceeding to step 524 through step 523, both the left and right arm coefficients γ are set to 0 (at this time, it is desirable to set the coefficient β to 0.8 by taking into consideration the concept of the fourth embodiment, but the explanation is simplified. This method is omitted because of).

When it is determined in step 517 that significant slippage has occurred, it is not possible to deal with the environment by simply swinging one of the current arms, and both arms must be swung, but the coefficient γ
Is increased to 1 for both the left and right arms, the coefficient Lβ for the left hand of the person who is swinging is reduced to 0.8 until both feet support period is reached (steps 518, 519, 520). If it is determined in step 509 that both arms are currently waving, if it is determined in step 525 that significant slippage still occurs, the control is changed as a whole, and the gait data with a slower speed is changed ( Steps 526, 527). When it is not determined that significant slippage has occurred, the state is maintained up to the number of steps N, and thereafter, one arm is shaken to save energy consumption (steps 528, 529, 530).

With the above processing, it is determined whether the coefficient γ for determining the swinging style of the arm is 0 or 1. Therefore, after that, step 53 is executed.
1 (FIG. 15), the shoulder joint angle correction value is calculated from the coefficient γ and the feedforward correction gain KFF. Subsequent steps are basically the same as in any of the previous embodiments, so description will be omitted. In this embodiment, changing the amplitude of the arm is not described for simplification of the description, but it can be easily understood that such a modification can be made by using the technique disclosed so far.

As described above, the sixth embodiment is basically constructed such that the arm does not swing and the arm swings only when a phenomenon due to a disturbance occurs. Regarding how to swing the arms to prevent recurrence, instead of swinging both arms immediately, swing from one arm first, and then swing both arms only when it is still insufficient. The reason why the arm is shaken as a first aid measure is that ensuring the stability has a higher priority than energy saving, and from that intention, the arms are shaken to ensure the stability. Even when reducing the recurrence prevention arm swinging method, it is possible to reduce energy consumption while confirming safety by sequentially changing from both arms to one arm and from a mode in which one arm does not swing at all.

In addition to the coefficient β used previously, a new coefficient γ
By introducing and using the following concept, the control algorithm can be made simple and easy to design. As a result of the simple organization of the algorithm, there is an advantage that the possibility of making mistakes at the time of design is reduced, the cause can be easily investigated even when a problem occurs, and there is an advantage that the speed of the computer can be maintained for execution.

16 and 17 are flow charts showing the algorithm of the seventh embodiment of the present invention.

The purpose of shaking the shoulder joint is to cancel a part of the leg inertial force, not to completely erase it. From this viewpoint, the method of setting the shoulder joint angle to the most economical vibration mode will be described below. As described above, when it is assumed that the arm can freely vibrate about the shoulder joint using the free pendulum, the joint motor only needs to be supplied with energy for driving itself, and does not need to perform work to the outside. Actually, since the vibration is attenuated by the friction of each part of the joint, only the energy that guarantees the amount of the attenuation is consumed, but this state requires the least energy consumption.

Explaining below, after reading the recommended leg gait data in step 605 in FIG. 16, the number of times of sampling n per cycle (two steps) is calculated from this data in step 606. For example, when walking in 0.5 seconds per step, the computer clock time is 5 msec.
, The number of samplings per cycle is 200
Times. Since it is necessary to make the number of samplings of the upper body and the number of samplings of the legs match, this processing is performed after step 607. First, in step 607, the cycle mT of the model free pendulum is divided by the above n to obtain a sampling cycle, and the angle ₁ αt of the model free pendulum is calculated for each sampling cycle. Here, the model free pendulum is a pendulum in which a weight of mass M is hung with a string of length L. If its amplitude 2α0 and length L are close to those of the arm,
Any constant will do.

In step 608, it is necessary to bend and adjust the elbow to match the vibration cycle of the arm with the walking cycle.
Therefore, the elbow angle ₃ α0 is selected. This angle ₃ α0
Is calculated in advance and set together with the leg gait, and can be easily selected by searching from the determined leg gait.

In step 609, the time required to reach this elbow joint angle is set to t _0, and a passing point on the way to reach ₃ α 0 within that time is created at each sampling time. Since a quadratic curve is used in order to make the transition smoothly and a complementary method between two points such as a method of suppressing jerk below a certain value is already known and is not a main element of the present invention, ₃ αt reproduction is performed. The details of the method are not touched upon. In step 610, the whole gait data is created by superimposing the time series data of the leg gait with the time series data of ₁ αt and ₃ α5 so that the leg and the upper body do not move separately. .

Steps 611 to 618, 62
Up to 2, 626 and 627 (FIG. 17), there is basically no difference from the fourth embodiment. Step 6 is different from the fourth embodiment.
Substituting 1 or 0 into the coefficient γ depending on whether the arm is shaken or not at 30 and 631, respectively, and the next step 63
2 is that the angle correction value of the shoulder joint can be calculated by a common calculation formula. The steps after this are basically the 4th
There is no difference from the example. If it is determined in step 614 that a large slip has occurred, the process proceeds to steps 644 and 645 to change the gait data. However, when this change is made, the gait cycle is different, so the upper body gait design is redone. There is a need.

In the above explanation, how to set the angle of the elbow in the case of extremely fast walking which does not match the walking cycle even if the elbow is adjusted or in the case of extremely slow walking. Although problems remain, in such cases, there is no choice but to bend the elbow to the maximum angle to minimize the inertial mass or extend the elbow to maximize the inertial mass. It is one of the features of this embodiment to adopt the most effective method under the given conditions. Even with such a configuration, the energy consumed by the motor of the shoulder joint can be minimized under given conditions.

FIG. 18 shows this relationship. In the figure, ● indicates a practically usable range of speed, and ○ indicates a faster walking speed, and in some cases, running rather than walking. The figure is conceptual only, and the saturation point at which the dash-dotted line shifts horizontally is determined by how the arm weight and leg weight are designed.

As described above, in the seventh embodiment, the free-vibration of the upper body gait is used to match the cycle of swinging the arm with the walking cycle, so that energy consumption can be further reduced. If the landing position and landing speed are changed depending on the disturbance, the walking cycle may be slightly deviated from the initial set value, but even in such a case, the upper body cycle changes according to the leg cycle. Can be controlled as follows. It is of course possible to arbitrarily combine the configurations of the previous embodiments in which the amplitude of the arm is changed according to the phenomenon of disturbance. Further, by using the coefficient γ, the algorithm can be simply described as compared with the fourth embodiment.

19 and 20 are flow charts showing the algorithm of the eighth embodiment of the present invention.

The feature of the eighth embodiment is that the feedforward gain KFF is discontinuously changed step by step as in the previous embodiment, and the coefficient k gradually approaches the gain KFF.
This is to use the FF to gradually change the swing amplitude of the arm for each control cycle so as to approach the gain KFF.

Explaining below, in FIG. 19, as in the case of the first embodiment, the procedure proceeds to step 705 through steps 701 to 704, where the shoulder joint angle correction gain KFF is replaced with a coefficient kFF, and the procedure proceeds to step 706 and the value of the time counter t. Is reset to 0. Subsequently, the routine proceeds to step 711, where it is judged whether or not the yaw rate difference value has exceeded the first threshold value during the current one step. When the determination is negative, the routine proceeds to step 712, where the yaw rate difference value is compared with the second threshold value, and the gain is changed or held at steps 712 to 718 according to the comparison result.

Next, in step 719, it is checked whether or not the coefficient kFF matches the gain KFF. If they do not match, the coefficient kFF is brought slightly closer to the gain KFF, and the process proceeds to step 720 to set the target angle of the shoulder joint. Is multiplied by a coefficient kFF to calculate the correction value. Then, the process proceeds to step 726 through steps 721 to 725 similar to the first embodiment, and if it is determined that the walking is not completed, step 73
The process proceeds to 0 to increment the time counter value and returns to step 708. When the result in step 711 is affirmative, the process proceeds to step 728, where it is confirmed that the gait is the first change, and the gait data is changed in step 729. It does not differ from the first embodiment if the switching from the one-leg supporting period to the two-leg supporting period is selected at the change of the gait.

In the eighth embodiment, the coefficient kFF is matched with the gain KFF in step 705, but when the gain is changed in step 714 or the like, step 71 is executed.
At 9,720, as a result, the gain is gradually changed toward the target value. As a result, the target gain KFF itself is 1
Although it may be discontinuously changed for each step, the actual gain kFF can be changed gradually in each control cycle, and the amplitude of the arm swing is gradually changed in each control cycle. It is possible to change the body gait smoothly.

21 and 22 are flow charts showing the algorithm of the ninth embodiment of the present invention.

The ninth embodiment is a modification of the eighth embodiment, and the correction gain KFF is finely changed. That is,
Instead of using the coefficient kFF, when it is determined in step 811 that the yaw rate difference value exceeds the second threshold value, the process proceeds to steps 812 to 817 to gradually increase the gain KFF within its upper and lower limit value KFFmax, min. It was decided to increase or decrease. Then, in step 818, the shoulder joint angle correction value is calculated based on the gain KFF. The remaining steps are the same as in the eighth embodiment.

Since the ninth embodiment is constructed as described above, the gain can be changed more finely than in the eighth embodiment, so that the body gait can be changed more smoothly.

23 and 24 are flow charts showing the algorithm of the tenth embodiment of the present invention.

The tenth embodiment secures stability when the frictional force obtained on the walking road surface cannot be predicted and it is not possible to reliably see that continuous walking is possible due to insufficient frictional force. For that purpose, the stability is restored by driving the joints that become free when the stability of gait is about to be impaired, in the direction of restoring the stability.

Explaining below, starting from step 901 in FIG. 23, through step 902 similar to the previous embodiment, the process proceeds to step 903 to read in three types of setting gains as in the fourth embodiment, and step 904. Go to and read the gait data. In the tenth embodiment as well, similar to the previous embodiments, different gait data at different walking velocities of high and low are preset, and in this step, the gait data on the high speed side is read. Then Step 905
After that, the process proceeds to step 906, where the feedback correction gain KFB is selected / set. In the tenth embodiment, since it is assumed that the frictional force obtained on the walking road surface cannot be predicted, the feedback correction gain KFB is a value set in association with the gait.

Then, through steps 907 to 909, the routine proceeds to step 910, where the yaw rate difference value is compared with the first threshold value, and if negative, step 913 (see FIG. 2).
Proceed to 4) and compare with the second threshold. Therefore, when it is determined that the yaw rate difference value exceeds the second threshold value, the process proceeds to step 914, and the yaw rate difference value is multiplied by the feedback correction gain KFB to calculate the shoulder joint angle correction value. When it is determined in step 913 that the yaw rate difference value is less than the second threshold value, the process proceeds to step 915 and the correction value is set to 0.

Next, in step 916, as shown in the figure, as in the fourth embodiment, the difference between the previous target value and the current target value is multiplied by the gain K1, and the encoder difference value is multiplied by the gain K2. And the correction value multiplied by the gain K3 are added to obtain a torque command value for the shoulder joint motor. As in the fourth embodiment, the left and right arms need to swing in opposite phases, so the sign of the third term is different between the left and right, and the torque command values for the other joint motors do not take into account the third term, and are simple. Is the sum of only the first and second terms.
When it is determined in step 910 that the yaw rate difference value exceeds the first threshold value, it is confirmed in step 911 that both legs are in the supporting period, and the process proceeds to step 912 to change the gait data to the low speed side. The remaining steps are no different from the previous embodiment.

As described above, in the tenth embodiment, the gait data itself is changed when the frictional force is insufficient and stable walking is continuously difficult even on a road surface where the frictional force cannot be predicted. Or, the stability is restored by changing the gain. That is, in response to various disturbances,
If you prepare a small number of gaits, try to walk with a small number of gaits, and if the influence of the disturbance is large, shake your arms and torso to cancel the influence of the disturbance, and change the actual environment. I was able to respond.

As described above, when a sufficient frictional force can be expected, it is desirable to walk with both hands and the body kept stationary without moving as much as possible in order to save energy. Therefore, it is desirable to prepare a gait in which the upper body is stationary and only both legs are moving as a reference gait. From that intention, similarly to the fourth embodiment and the like, the swing of both hands is added to the reference gait only when a slip actually occurs on a slippery road surface. Then, when the slippage is relatively large, it is determined that the stability cannot be recovered simply by shaking the arms, and the gait data for low friction with improved stability is changed. In the second gait data, the walking speed is set to be slightly slower as in the previous embodiment for safety. There is no difference from the previous embodiment in that the joint of the waist may be swung instead of swinging the arm to utilize the mass of the entire upper body.

In the first to tenth embodiments described above, the arm is driven by the shoulder joint in a phase opposite to the phase of the leg movement, but the waist joint (for example, joint 4R / L) is driven. It will be obvious that the whole may be driven. Further, in the first embodiment, an example in which the elbow joint is driven in accordance with the shoulder joint is shown, but in other embodiments, even if the elbow joint that is lighter and consumes less energy is driven instead of the shoulder joint, It is arguable that the same effect can be obtained if the degree of instability is small.

In the description of the first to tenth embodiments, the yaw rate of the tilt sensor is used to determine the presence or absence of slip, but angle information obtained by integrating the yaw rate may be used. Alternatively, the same effect can be obtained by using the differential value of the yaw rate. When using this kind of yaw rate-derived information, there is a change rate of the angle or a change rate of the yaw rate that may be obtained from a gait, and therefore it is necessary to take the difference from the measured value.

Also, the type of sensor is not limited to the inclination sensor, and the maximum value of this amount is smaller than the maximum value of the expected value by using, for example, the lateral component FY of the output of the 6-axis force sensor. In this case, it may be determined that slippage has occurred. When slipping, the coefficient of static friction shifts to a coefficient of dynamic friction smaller than this, so that the grip force of the foot on the road surface decreases. Alternatively, by taking the idea of sensor fusion into consideration, it may be determined that slip occurs when either the yaw rate or the horizontal component force of the six-axis force is different from the expected value. The engineering circuit of such a judgment circuit can be easily realized by forming a judgment circuit for each piece of information and connecting the results with an OR circuit.

Further, in the above-mentioned first to tenth embodiments, the case of forming a gait in advance and storing the result in the gait memory has been described. However, if the computer has a sufficient calculation speed, it is mounted. A computer may calculate the gait in real time. In that case, initialization and termination may be performed at the starting point and the destination, and the algorithm may change a little, but in certain sections where the attribute value is almost constant, such as stair climbing and slope walking, walking on slippery roads on level ground. For example, the above algorithm can be applied to each section.

FIG. 25 is a flow chart showing the algorithm of the eleventh embodiment of the present invention.

The eleventh embodiment and the following embodiments are the same as the first embodiment.
It is intended to recover the stability of the posture when the frictional force is insufficient and the stability of the posture is lost in walking in which the frictional force on the road surface is unpredictable as in the 0th embodiment.

As described above, when the frictional force of the sole of the support leg is negative to the moment about the yaw axis, the robot turns.
When the robot turns, the traveling direction is deviated even if the robot does not fall, and the robot deviates from the intended route. It is possible in principle and practically to correct this deviation again by using the inclination sensor 22 or by attaching a visual means and using the information, but the smaller the deviation, the easier the correction work. Is. In order to reduce the deviation, when the spin phenomenon is detected, it is desirable to rotate the support leg and make some correction at this time before the free leg lands. The eleventh embodiment is based on such an intention.

Explaining below, starting from step 1001 in FIG.
After tracing up to 9 as in the tenth embodiment, step 101
The yaw rate difference value is calculated at 0 and the threshold value THθ dot (first
If the yaw rate difference value exceeds this threshold value, the yaw rate difference value is multiplied by the feedback correction gain KFB (also set according to the walking speed). Calculate the correction value cαt.

Next, in step 1011 the motor torque command value is calculated for the lumbar joint 7R / L as in the tenth embodiment, and the remaining joints are calculated as in the tenth embodiment. Next, in step 1012, the command value is output, and unless it is determined in step 1013 that the walking has ended, the process returns to step 1007 to continue the same work.

In the eleventh embodiment, as described above, the yaw rate is monitored, and if the difference from the expected value DSRDθZ dot is detected, the hip joint 7L or 7R is driven in the direction in which the difference decreases. The orientation of the entire robot can remain substantially unchanged, even though the supporting points of the standing leg have slipped and changed direction. As a result,
The robot is less likely to get out of the way, and the amount of work to be corrected is reduced even if the path goes out of order. In the eleventh embodiment, the joints 7L and 7R are driven by the same amount in the opposite directions. However, even if one of the joints, for example, the supporting leg is focused and only the joint of the supporting leg is moved, the same result is obtained. The effect is obtained.

FIG. 26 is a flow chart showing the algorithm of the twelfth embodiment of the present invention.

The twelfth embodiment is intended to lift the arm in the direction in which the robot leans down when the robot leans in the lateral direction due to a side wind or the like and keep the robot on the stable side by the inertial force at that time. .

Explaining below, in FIG. 26, after step 1101 to step 1109, step 101
0, where the tilt angle θX detected by the tilt sensor 22
Is determined to determine whether it exceeds a predetermined value Δ (significant minimum value), and when the result is affirmative, the process proceeds to step 1111 to identify the positive or negative of the detected tilt angle θX, and a positive value, that is,
When it is determined in the promise of FIG. 1 that the vehicle is inclined to the right in the traveling direction, the detected value is multiplied by the feedback correction gain KFB (read in step 1106) to calculate the correction value cRαt (for the right side). At this time, cLαt
(Correction value for left side) is 0. Also step 1110
When the answer is NO, the process proceeds to step 1112, and the left and right correction values are set to zero.

Then, the process proceeds to step 1113, and the elbow joint 3
The target angle of R / L is set to 0 (that is, the elbow joint is fully extended), and the process proceeds to step 1114 to calculate the motor torque command value for the shoulder joint 2R / L and other joints as shown in the figure. The remaining steps are the same as in the eleventh embodiment.

In the twelfth embodiment, as described above, in order to increase the inertial force, the arm is extended without bending at the elbow. So if your elbow is bent for any reason, lift your arm while straightening it. In this case, the arm is lifted outward because of the influence of the crosswind, but when the user leans to the right, the right arm is lifted. The left arm is convenient because it does not interfere with the body. For this purpose, prepare a step to check whether the sign of the angular velocity around the X axis is positive or negative,
The algorithm is such that either the joint 2R or 2L is moved in the lifting direction according to the sign.

FIG. 27 is a flow chart showing the algorithm of the 13th embodiment of the present invention.

The thirteenth embodiment is a technique capable of recovering stability by using joints other than arms when the robot hits another object with a shoulder and receives a lateral external force or when a side wind is encountered. Indicates. Consider the case where the robot is about to fall to the right side due to a crosswind from the left while walking in FIG. If a human encounters a scene like this, and if his hands are blocked at that time, he is leaning in the direction in which he is likely to fall (in this case, to the right) and using the inertial force at that time to restore stability. . This is reproduced by the robot in the algorithm of the thirteenth embodiment.

Explaining below, the square of the tilt angle θX detected in step 1210 is compared with a predetermined value Δ, and it is confirmed that it exceeds it, and the process proceeds to step 1211.
The detected tilt angle θx is multiplied by the feedback correction gain KFB to calculate the correction value c6αt of the joint 6 at the center of the waist, and in step 1213 a dummy for shifting the hip position to the left to the target angle of the hip joint 9,12R / L. Is added to correct the target angle, and in step 1214 the motor torque command value is calculated for the joints including the joint 6R / L at the center of the waist as shown. The remaining steps are the same as in the eleventh embodiment.

In the thirteenth embodiment, as described above, a correction amount for driving the motor of the joint 6 at the center of the waist in the direction of correcting the posture collapse due to the side wind is added, and as a result, the output of the tilt sensor decreases (posture. To recover). However, since the balance is dynamically restored, the upper body is tilted in the direction in which it is likely to fall, and at the same time, the hip joint 9R / L and the foot joint 12R / L are driven by a small amount, and a crosswind blows over the entire waist. It is necessary to move horizontally in the direction of coming. For this reason, the amount obtained by adding the minute angle Δα to the gait data is given to these four motors as a new target angle.

Therefore, in the thirteenth embodiment, the robot that encounters a disturbance such as a crosswind shifts the waist position to the direction in which the crosswind is blowing, just as when a human being responds. Stability can be restored by pushing down in the same direction as the crosswind.

FIG. 28 is a flow chart showing the algorithm of the 14th embodiment of the present invention.

The fourteenth embodiment is a modification of the twelfth embodiment shown in FIG. 26. After proceeding to step 1309 and then proceeding to step 1210, the absolute value of the moment MX detected there is determined by a predetermined value Δ squared (from gait). (Expected maximum value) and confirm that the maximum value is exceeded.
In step 1, it is determined whether the detected moment MX exceeds a positive threshold value (Δsquare) or a negative threshold value (−Δsquare), and a correction value is calculated by a method similar to that of the twelfth embodiment. When the result in step 1310 is negative, the correction value is set to 0 in step 1313, and the elbow joint is returned to its original gait in step 1314. The remaining steps are the same as in the twelfth embodiment.
Also, the effect is similar to that of the twelfth embodiment. Thus, in addition to the information of the tilt sensor, MX or the like can be used with the information of the 6-axis force sensor. This is because when leaning, all of the weight is applied to the outermost periphery of the foot of the standing leg, and MX shows the maximum value.

The fourteenth embodiment is an example in which the twelfth embodiment is modified to use the moment MX, but other embodiments can be modified in the same manner. In addition to the moment MX, the force FY may be used. This is because the external force due to the cross wind appears in FY in the form of shear force. The concept of the algorithm such as the twelfth embodiment shown in FIG. 26 can be applied to not only lateral disturbance but also longitudinal disturbance in the same manner. With respect to the disturbance in the front-back direction, both arms can be lifted in the same direction, and the restoring force can be increased as compared with the case of dealing with the disturbance in the lateral direction.

The effects described in the tenth embodiment are also applicable to the eleventh embodiment and thereafter. That is, it is sufficient to prepare a small number of gaits in response to various disturbances. Try walking with a small number of gaits, and if the influence of the disturbance is large, the arms and torso should be adjusted to cancel the influence of the disturbance. By shaking, it is possible to respond to actual environmental changes.

Furthermore, as described above, the first to fourteenth embodiments have been described, but they can be variously modified and can be partially or wholly combined with each other.

[0142]

According to the first aspect of the present invention, the upper body gait (more
For details , it becomes easier to determine the drive control amount of the upper body joint ,
The calculation can be easily performed using the on-board microcomputer, and the gait data development efficiency can be improved. Furthermore
In addition, the frictional force on the pedestrian surface is insufficient and stability deteriorates.
However, it can be easily secured or recovered.

According to the second aspect, in addition to the effect of the first aspect, it becomes easy to determine the upper body gait in a bipedal walking robot whose gait design is particularly difficult.

According to the third aspect, in addition to the effect of the second aspect, it becomes easier to determine the upper body gait.

According to the fourth aspect, in addition to the effect of the second aspect, it becomes easier to determine the upper body gait.

According to the fifth aspect, the body gait can be easily determined, and the energy consumption can be minimized.

According to the sixth aspect, in addition to the effect of the fifth aspect, it becomes easier to determine the body gait that causes free vibration.

According to the seventh aspect, even if the frictional force is insufficient on the walking road surface and the stability deteriorates, it can be easily secured or recovered.

According to claim 8, a gait with a large margin of stability in which the frictional force is not inherently insufficient on the walking road surface in consideration of the past experience and the present situation comprehensively. In addition to achieving the above, it is possible to easily secure or recover the stability even if the frictional force is insufficient and the stability is reduced.

According to the ninth aspect, in addition to the effect of the seventh aspect, it becomes easier to secure or recover the stability.

According to the tenth aspect, in addition to the effect of the seventh aspect, the stability can be more surely ensured or recovered more easily.

According to the eleventh aspect, in addition to the effects of the seventh aspect, it is possible to surely secure or restore the stability in a bipedal walking robot in which it is difficult to secure the stability. .

[0153]

[0154] According to claim 1 2, wherein, in addition to the effect of claim 8 wherein, can be controlled in consideration of such past experience, is possible to more reliably ensure to ensure stability it can.

[0155] According to claim 1 3, wherein, in addition to the effect of claim 8 wherein, the frictional force of the walking road surface is unknown, it is possible to reliably and stably Ensuring or recovery.

[0156] In the claims 1 to 4, wherein, in addition to the effect of claim 8, wherein, suitably determine the control amount.

[0157] In the claims 1 to 5, wherein, in addition to the effect of claim 1 6 wherein, the control amount can be more suitably determined.

[0158] In the claims 1-6 wherein, in addition to the effects of such claim 7 wherein the control amount can be more suitably determined, also a smooth change of gait of the upper body.

[0159] In the claims 1-7 wherein, in addition to the effect of such claims 1 to 4, wherein, when walking the next same road, it is possible to walk be further stabilized.

[0160] In the claims 1-8, wherein, in addition to the effects of such claim 7 wherein, the determination of the upper body gait is facilitated in bipedal walking robot.

According to the nineteenth aspect , in addition to the effects of the seventh aspect and the like, the body gait can be easily determined in the bipedal walking robot.

[0162] According to claim 2 0 term can be in addition to the effect of such claim 9 wherein, to achieve the stability of the securing or recovered more reliably.

According to the twenty- first aspect , in addition to the effects of the seventh aspect and the like, it is possible to reliably determine whether or not the frictional force is sufficient to maintain the stability of walking.

[0164] According to claim 2 second term, in addition to the effects of such claim 7 wherein the friction force is reliably determine whether sufficient or not to maintain the stability of the gait.

According to the twenty- third aspect , in addition to the effects of the seventh aspect, it is possible to more reliably restore the stability of walking.

According to the twenty- fourth aspect , in addition to the effect of the tenth aspect, the energy consumption can be reduced.

According to the twenty- fifth aspect , in addition to the effect of the tenth aspect, the energy consumption can be reduced.

[0168] In the claims 2 to 6 wherein, in addition to the effect of such claim 2 paragraph 4, it is possible to reduce the energy consumption in the two-legged walking robot.

[0169] In the second aspect 7 wherein, in addition to the <br/> effect of claims 2 to 6 wherein, achieving maximum limit the reduction in energy consumption in the two-legged walking robot.

[0170] According to claim 2 8 wherein can be achieved in addition to the effect of claim 2 7 wherein, the reduction of energy consumption when carrying the load in the biped walking robot with a preferred.

According to the twenty- ninth aspect, in addition to the effects of the seventh aspect and the like, it is possible to reduce energy consumption while ensuring stability in the bipedal walking robot.

[0172] In the Claim 3 0 wherein, can be stopped in addition to the effect of claim 7, wherein, the effect of changing the gait to a minimum.

[Brief description of drawings]

FIG. 1 is a skeleton diagram showing an overall configuration of a robot in a walking control device for a legged mobile robot according to the present invention, including a degree of freedom arrangement.

FIG. 2 is an explanatory view showing the details of the walking control device in the walking control device of the legged mobile robot according to the present invention in terms of hardware.

FIG. 3 is the first half of the flow chart showing the first embodiment of the present invention.

FIG. 4 is the latter half of the flow chart showing the first embodiment of the present invention.

FIG. 5 is the first half of the flow chart showing the second embodiment of the present invention.

FIG. 6 is the latter half of the flow chart showing the second embodiment of the present invention.

FIG. 7 is the first half of the flow chart showing the third embodiment of the present invention.

FIG. 8 is the latter half of the flow chart showing the third embodiment of the present invention.

FIG. 9 is an explanatory diagram showing a work of discriminating luggage carrying in the flow chart of FIG. 7;

FIG. 10 is the first half of the flow chart showing the fourth embodiment of the present invention.

FIG. 11 is the second half of the flow chart showing the fourth embodiment of the present invention.

FIG. 12 is the first half of the flow chart showing the fifth embodiment of the present invention.

FIG. 13 is the latter half of the flow chart showing the fifth embodiment of the present invention.

FIG. 14 is the first half of the flow chart showing the sixth embodiment of the present invention.

FIG. 15 is the latter half of the flow chart showing the sixth embodiment of the present invention.

FIG. 16 is the first half of the flowchart showing the seventh embodiment of the present invention.

FIG. 17 is the second half of the flowchart showing the seventh embodiment of the present invention.

FIG. 18 is an explanatory diagram showing a relationship between an arm vibration cycle and a walking cycle in the flow chart of FIG. 16;

FIG. 19 is the first half of the flow chart showing the eighth embodiment of the present invention.

FIG. 20 is the latter half of the flow chart showing the eighth embodiment of the present invention.

FIG. 21 is the first half of the flowchart showing the ninth embodiment of the present invention.

FIG. 22 is the second half of the flowchart showing the ninth embodiment of the present invention.

FIG. 23 is the first half of the flowchart showing the tenth embodiment of the present invention.

FIG. 24 is the latter half of the flow chart showing the tenth embodiment of the present invention.

FIG. 25 is a flow chart showing an eleventh embodiment of the present invention.

FIG. 26 is a flow chart showing a twelfth embodiment of the present invention.

FIG. 27 is a flow chart showing the thirteenth embodiment of the present invention.

FIG. 28 is a flow chart showing a fourteenth embodiment of the present invention.

[Explanation of symbols]

1,2,3,4,5,6,7,8,9,10,11,1
2 joints 12, 13, 14, 16, 17, 18, 19 link 15 foot 20 hand 21 6-axis force sensor 22 tilt sensor 30 reference gait generator 31 gait memory 32 environment knowledge base 35 central processing circuit 37, 38 Motor drive circuit

Claims (30)

(57) [Claims] 1. A legged mobile robot comprising at least an upper body joint for driving an upper body, a plurality of joints on a leg connected to the upper body, and walking by driving the joint of the leg. The walking control of the legged mobile robot is characterized in that the drive control amount of the upper body joint is determined so that the inertial force generated by driving the legs is at least partially canceled by the movement of the upper body. Control device. 2. The legged mobile robot is a bipedal walking robot having a structure similar to that of a human body, the legged mobile robot includes a waist joint for driving the legs, and an arm is attached to the upper body connected to the waist joint. The walking control device for a legged mobile robot according to claim 1, wherein a driving shoulder joint is connected as the upper body joint. 3. The drive control amount of the shoulder joint is determined based on the movement of the lumbar joint so that the inertial force is at least partially canceled by the movement of the arm. Control device for a legged mobile robot. 4. The elbow joint is provided on the leg and the elbow joint is provided on the arm as the upper body joint, and the elbow joint is driven so that the inertial force is at least partially canceled by the movement of the arm. The walking control device for a legged mobile robot according to claim 2 or 3, wherein the control amount is determined based on the movement of the knee joint. 5. A legged mobile robot comprising at least an upper body joint for driving the upper body, a plurality of joints in a leg connected to the upper body, and walking by driving the joint of the leg. A walking control device for a legged mobile robot, which determines the drive control amount of the upper body joint so that the movement of the upper body is free vibration. 6. The walking control device for a legged mobile robot according to claim 5, wherein a cycle of the free vibration is matched with a walking cycle of the robot. 7. A legged mobile robot comprising at least an upper body joint for driving the upper body, a plurality of joints on a leg connected to the upper body, and walking by driving the joint of the leg. A. Detecting means for detecting a parameter representative of the frictional force acting between the road surface and the tips of the legs, b. Judging means for judging the stability margin of the robot by comparing the detected parameter with a predetermined value; and c. Determination result driving control amount changing means for changing the drive motion control of at least previous SL upper body joint in accordance with the walking legged mobile robot control system characterized by comprising a judgment means. 8. A legged mobile robot comprising at least an upper body joint for driving an upper body, a leg connected to the upper body having a plurality of joints, and walking by driving the joint of the leg. A. Means for storing parameters representative of frictional forces acting between the road surface and the tips of the legs, and b. Stored parameters walk controller of a legged mobile robot which is characterized in that it comprises at least before SL drive control amount changing means for changing the drive dynamic control of the upper body joints, in accordance with. 9. The drive control amount changing means changes the drive control amount of the upper body joint so as to increase a margin of stability of the robot according to a determination result of the determination means or a stored parameter. 9. The walking control device for a legged mobile robot according to claim 7, wherein the walking control device is modified. 10. The drive control amount changing means, in accordance with a judgment result of the judging means or a stored parameter,
9. The walking control of a legged mobile robot according to claim 7, wherein the drive control amount of the upper body joint is changed so as to at least partially cancel out the inertial force generated by driving the leg portion. apparatus. 11. The legged mobile robot is a bipedal walking robot having a structure similar to a human body, and a shoulder joint for driving an arm to the upper body is provided as the upper body joint, and the drive unit is configured to perform the determination. 9. The drive control amount of the shoulder joint is determined so as to increase the stability margin of the robot according to the determination result of the means or the stored parameter. A walking control device for a legged mobile robot. 12. The walking control device for a legged mobile robot according to claim 8, wherein the stored parameter is a value set based on a frictional force predicted on a walking road surface. 13. The stored parameter is a value set in association with a gait.
A walking control device for a legged mobile robot according to item. 14. The leg type according to claim 9, wherein the drive control amount changing unit increases or decreases the stored parameter according to a determination result of the determining unit. A walking control device for mobile robots. 15. The drive control amount changing means increases or decreases the stored parameter stepwise at least every predetermined number of steps or predetermined time according to the judgment result of the judgment means. A walking control device for a legged mobile robot according to item 14. 16. The drive control amount changing means has an adjustment rule for smoothly changing the drive control amount when a gait is changed, according to any one of claims 7 to 15. A walking control device for a legged mobile robot. 17. The legged mobile robot according to claim 14, wherein the drive control amount changing unit updates the stored value with the parameter increased or decreased at the end of walking. Walking control device. 18. The legged mobile robot is a bipedal walking robot having a structure similar to a human body, wherein a shoulder joint for driving an arm is provided as the upper body joint on the upper body, and the drive control amount changing means is provided. When the margin of stability of the robot is judged to be low by the judgment result of the judgment means or the stored parameter, the drive control amount of the shoulder joint is
9. The walking control device for a legged mobile robot according to claim 7, wherein a drive control amount that is in a phase opposite to that of the movement of the leg is added. 19. The legged mobile robot is a bipedal walking robot having a structure similar to a human body, and is provided with an upper body joint capable of driving the upper body, and the drive control amount changing means is provided in the judging means. When it is determined that the stability margin of the robot is low based on the determination result or the stored parameter, the drive control amount of the upper body joint is a rotation control in a phase opposite to the movement of the leg. The walking control device for a legged mobile robot according to claim 7, wherein the amounts are added. 20. The upper body joint comprises at least two joints, a first joint and a second joint, which are connected in series with each other, and the drive control amount changing means includes the first joint based on a determination result of the determining means. 11. The drive control amount is changed so that the second joint is driven in a direction in which the inertial force viewed from the first joint is increased when is driven. A walking control device for a legged mobile robot according to item. 21. The detection means is a sensor for detecting rotation of the robot about a vertical axis, and the determination means compares the sensor output with a predicted value determined from a gait of the robot, and uses the comparison value as a comparison value. The walking control of a legged mobile robot according to any one of claims 7 and 9 to 20, wherein the drive control amount changing unit changes the drive control amount when a predetermined difference occurs. apparatus. 22. The detection means is force information acting on a ground contact point of the robot, and the determination means compares the sensor output with a predicted value determined from a gait of the robot, and uses the comparison value as the comparison value. The walking control of a legged mobile robot according to any one of claims 7 and 9 to 20, wherein the drive control amount changing unit changes the drive control amount when a predetermined difference occurs. apparatus. 23. The robot selects one of a plurality of predetermined gait data and walks on the basis of the selected gait data, and the judging means compares the sensor output with the predicted value to obtain a larger second value. 7. The drive control amount changing means changes the selected gait data when a predetermined difference occurs in the comparison value.
23. A walking control device for a legged mobile robot according to any one of items 22 to 22. 24. The drive control amount changing means drives the upper body joint to walk, and stops the driving of the upper body joint in the middle of walking according to the judgment result of the judging means. The walking control device for a legged mobile robot according to claim 10, wherein the drive control amount is changed. 25. The drive control amount changing means causes the body joint to walk without driving the upper body joint, and drives the upper body joint in the middle of walking according to the determination result of the determining means. The walking control device for a legged mobile robot according to claim 10, wherein the drive control amount is changed. 26. The legged mobile robot is a bipedal walking robot having a structure similar to a human body, and is provided with shoulder joints for driving arms to the left and right of the upper body as the upper body joints,
The walking control device for a legged mobile robot according to claim 24 or 25, wherein at least one of the left and right shoulder joints is driven. 27. The arm is capable of carrying luggage, and the drive control amount changing means changes the drive control amount of the left and right shoulder joints depending on whether or not the luggage is carried. A walking control device for a legged mobile robot according to claim 26. 28. The walking control device for a legged mobile robot according to claim 27, wherein the drive control amount changing means does not drive a shoulder joint of an arm carrying the luggage. 29. The legged mobile robot includes a plurality of the body joints, and the drive control amount changing means is to drive the body joints according to the stability margin determined by the determining means. The walking control device for a legged mobile robot according to claim 7, wherein the number of the robots is increased or decreased. 30. The drive control amount changing means changes the drive control amount by selecting a time with a high margin of stability. Control device for a legged mobile robot.