JP2009190122A - Legged robot and method of controlling the same - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a legged robot which achieves a stable walking/traveling action by excellently maintaining the follow-up capability of a track, and also to provide a method of controlling the robot. <P>SOLUTION: According to the legged robot, a target torque of a joint is calculated by using "a dynamic system in which mass is concentrated in a trunk, a left knee position, a right knee position, a position in the vicinity of a right leg flat portion, and a position in the vicinity of a left leg flat portion" that is obtained by modeling the legged robot from data providing a left leg flat positional attitude, a right leg flat positional attitude, and a trunk positional attitude, and data providing a trunk position for implementing a positional attitude. Then a target rotational quantity for each joint necessary for implementing the positional attitude is calculated based on a data group. An instruction torque to be applied to a motor of an actuator is adjusted based on a deviation (62, 64) between the calculated target rotational quantity and a detected actual rotational quantity, and a calculated target torque is added (66) to the instruction torque. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a legged robot and a control method thereof.

Many techniques for generating a walking trajectory for a legged robot to walk stably have been proposed. For example, Patent Document 1 discloses a gait data creation device capable of calculating a gait data with a small amount of calculation by calculating a ZMP (zero moment point) of a walking robot at high speed. The gait data creation device described in Patent Document 1 speeds up ZMP calculation by approximating a full model of a robot to six mass points and appropriately allocating these mass points. In this specification, the “trajectory” refers to data describing a change in position / posture with time.
JP 2004-237403 A Japanese Patent No. 3148827

  However, even with the ideal trajectory data created by the prior art, there is a problem that a stable walking / running operation cannot always be realized when the trajectory data is actually reproduced by a robot. It was. When the created trajectory is actually reproduced by the robot, the following reasons are assumed as the reason why the robot cannot stably walk / run.

  First, there is a deterioration in the followability and responsiveness to the trajectory data due to friction generated in the joint portion of the robot, actuator performance, gain setting, and the like. The frictional characteristics change with the reassembly of the robot and the passage of time. Usually, since the degree of assembly of the robot differs each time, the friction also changes each time. Even with the passage of time, the friction changes due to less grease. Further, the target trajectory is discontinuous depending on the walking / running motion pattern, stabilization control, and the like. For this reason, depending on the joint, the commanded torque may exceed the specification of the actuator, and the actuator cannot be operated normally at a torque or speed exceeding the maximum specification range of the actuator.

  Another cause is the deterioration of the center-of-gravity position followability due to the bending of a low-rigidity part of the robot (such as a shock absorbing member on the sole or a speed reducer provided at a joint). Such a low-rigidity part has an impact absorbing effect, but can also cause the robot to lose its posture. If the servo tracking is good, the mechanism (link, joint, reducer, etc.) is sufficiently rigid, and there is no disturbance, the robot will be at the target value (center of gravity, joint angle, etc.). (Sequential data) can be satisfactorily followed, and stable operation is possible. However, in reality, it is difficult to adjust the servo following ability, and a portion having a low mega rigidity is likely to bend. With regard to the servo, although some adjustment is possible by experiment, when the rigidity is increased with respect to the mechanism, there is a problem that it takes time for weight increase, design change, and reassembly.

  Furthermore, there is a large deviation between the target ZMP and the actual ZMP due to high gain position control and stabilization control of the actuator. In the prior art, the torque command value required to drive the joint is calculated, and the response to the torque command value is enhanced by increasing the gain in the control in order to accurately follow the torque command value. . However, in such a situation where the torque command value is output, there is a case where the landing is made at a wrong timing due to the fact that the sole of the robot is bent. In this case, since the rigidity of the sole is controlled by locally increasing at the moment of landing, there is a possibility that the sole kicks the road surface. For this reason, in order to realize shock absorption at the time of landing, it is preferable to reduce the gain and reduce the torque command value applied to the actuator. There is also a problem that control is likely to become unstable when the gain is high. On the other hand, if the response to the torque command is inferior except for the stance when landing, the robot cannot achieve the target trajectory (posture) and tends to become unstable. is there.

  In Patent Document 1, data (mass point position and acceleration) created in the process of creating gait data is used only for speeding up the creation of trajectory data. It was not considered.

  In Patent Document 2, the joint of the leg and the grounded end of the leg are connected by a spring mechanism, and a deformation angle of the spring mechanism necessary for generating the target load is obtained. A walking control device for a legged mobile robot that corrects a joint angle command value is disclosed. In the walking control device for the legged mobile robot described in Patent Document 2, the tip of the leg is flexibly grounded to the unevenness and inclination of the floor, the traverse torque is improved and the disturbance torque is reduced, and the impact at the time of landing is also effective. The goal is to relieve and secure the ability to kick the initial floor so as not to hinder walking. However, in the technique described in Patent Document 2, a spring mechanism is provided between the leg joint and the leg grounding end, and the target load is obtained in order to obtain the deformation angle of the spring mechanism necessary for generating the target load. Complex calculation is required for detection, and the target load cannot be detected in a short time.

  As described above, in the prior art, even if the created trajectory data is ideal, when the trajectory data is actually reproduced by the robot, the trackability of the trajectory cannot be maintained satisfactorily. There is a problem that the walking / running operation cannot be realized.

  The present invention has been made to solve such a problem, and provides a legged robot capable of stably realizing walking / running operation by maintaining good track followability, and a control method thereof. For the purpose.

  The legged robot according to the present invention inputs and stores data indicating the left foot position / posture, right foot position / posture, and trunk position / posture of the legged robot, and data indicating the trunk position realizing the position / posture. And a "dynamic system in which mass is concentrated at the body, left knee position, right knee position, left foot position, and right foot position" that models the legged robot from the data group. Means for calculating a target torque of a joint of the legged robot, means for calculating a target rotation amount of each joint necessary for realizing the position and orientation from the data group, and the calculated target rotation Means for adjusting the command torque applied to the motor of the actuator based on the detected amount and the deviation of the detected actual rotation amount, and the adjusting means adds the calculated target torque to the command torque. is there.

  In this way, when feedback control is performed so that the actual rotation amount matches the target rotation amount, the calculated target torque is input (added) to the torque command value that is the operation amount as feedforward torque. Thus, even when the control gain is reduced, the followability of the target rotation amount can be maintained satisfactorily. When creating gait data, a vector indicating the left foot posture and a vector indicating the right foot posture are often input. However, as will be described later, the foot posture change pattern can be standardized. The input of the left foot posture vector and the right foot posture vector is not indispensable.

  In another aspect of the legged robot according to the present invention, data indicating a left foot position / posture, a right foot position / posture, and a trunk position / posture of the legged robot, and data indicating a trunk position that realizes the position / posture are provided. "Mechanism in which mass is concentrated at the body, left knee position, right knee position, left foot position, and right foot position" Means for calculating a target torque of a joint of the legged robot using a system, and means for calculating a target rotation amount of each joint necessary to realize the position and orientation from the data group. Further, a means for adjusting a command torque applied to the motor of the actuator based on the target rotation amount and a deviation of the detected actual rotation amount, a means for detecting the actual torque of the actuator, and the calculated target torque And before It means for comparing whether Compare command torque, or calculated the actual torque which the detected and target torque, based on the comparison result, in which comprises means for estimating the frictional characteristic value of the actuator.

  As a result, even when the friction characteristics change due to the passage of time, temperature change, overhaul, etc., the accuracy of the calculated target torque can be maintained satisfactorily by identifying the friction characteristic values at any timing. it can.

  The comparison means may perform comparison using torque data at the time of a free leg. This is because the torque data at the time of the free leg does not receive the reaction force from the floor, and the friction and inertia components become remarkable, so that the friction characteristic value can be estimated more easily.

  Another aspect of the legged robot according to the present invention is a legged robot in which the rigidity of a part of the legged robot is locally reduced, and the low-rigidity part acts as a spring element, Means for inputting and storing data indicating left foot position / posture, right foot position / posture and trunk position / posture, data indicating the trunk position realizing the position / posture, and the legged robot from the data group; Calculate the target torque of the joint of the legged robot using the modeled "dynamic system with mass concentrated at the torso, left knee position, right knee position, left foot position and right foot position" Means, a means for calculating a target rotation amount of each joint necessary to realize the position and orientation from the data group, a calculated target rotation amount, and a deviation of the detected actual rotation amount. The actuator motor On the basis of the calculated target torque, the deviation of the detected actual torque, and the spring constant of the spring element. Means for calculating a correction amount necessary for correcting the target rotation amount of the actuator that drives the rigid portion, and the adjustment means add the calculated correction amount to the target rotation amount.

  In this way, by correcting the deflection of the low-rigidity part at the joint angle level based on the relationship between the target torque and the actual torque, the position followability of the center of gravity can be improved without impairing the impact absorption effect of the low-rigidity part. Can do.

  Further, the foot sole portion of the legged robot may act as the spring element. Furthermore, a reduction gear of the actuator may act as the spring element. The actuator that drives the low rigidity portion may be an actuator that drives an ankle joint of the legged robot.

  Furthermore, the means for setting and storing a limit value for the command torque is compared with the command torque and the limit value, and when the command torque exceeds the limit value, the limit value is determined. A means for outputting a drive signal to the motor of the actuator may further be provided. This is because, by setting a limit value for the command torque, it is possible to prevent excessive torque from being generated in the joint portion and to prevent the foot from being peeled off the floor.

  Further, an allowable range in which ZMP can be allowed is calculated at the sole portion of the legged robot at the time of standing, and a limit value for the command torque may be set based on the calculated allowable range. .

  Furthermore, the target torque calculation means calculates from the data group a means for calculating a target rotation amount of each joint necessary for realizing the position and orientation, and “a trunk and a left knee modeled on the legged robot” Position, right knee position, left foot vicinity position, and right foot position near the position of the mass point from the temporal change of the target rotation amount calculated for each joint A means for calculating acceleration and calculating a target torque of the joint of the legged robot may be provided.

  Further, when the leg portion of the legged robot is a standing leg, the target torque of the target joint is calculated using another mass point excluding the mass point existing in the toe direction from the target joint of the standing leg. You may do it.

  Furthermore, when the leg portion of the legged robot is a free leg, the target torque of the target joint is calculated using the material point existing in the toe direction from the target joint of the free leg. Also good.

  Further, when the legged robot is in an aerial phase floating in the air, the target torque of the target joint is calculated using a mass point existing in the toe direction from the target joint of the leg. May be.

  Furthermore, the means for setting and storing a predetermined value capable of driving the actuator and the target torque are compared with the predetermined value, and the target torque is calculated when the target torque exceeds the predetermined value. You may make it further provide the means to change to the said predetermined value.

  Further, a maximum torque value that can be driven by the actuator may be set as the predetermined value.

  Further, the target torque calculating means calculates data indicating the left knee position and the right knee position from the data group, the torso, the left knee position, the right knee position, the position near the left foot, and the A dynamic system in which mass is concentrated in the vicinity of the right foot, the left foot position / posture, the right foot position / posture, the trunk position, the trunk position / posture, the left knee position, and the right knee position. A means for calculating a target torque of the joint of the legged robot from the data may be provided. In this target torque calculation, since it can be calculated geometrically from the dimension data of the robot, the calculation amount can be much less than the calculation of the joint angle of each joint. In calculating the target torque from the position, posture, and motion state of the robot, instead of using the conventional complex system of “12 joints 13 links”, the body, left knee position, right knee position, and left foot position A dynamic system in which mass is concentrated near the right foot is used. By using a dynamic system in which concentrated mass exists at the above position, the target torque of the robot can be accurately calculated with a small amount of calculation. Therefore, the target torque can be calculated accurately with a small amount of calculation.

  Further, the dynamic system used in the target torque calculating means concentrates the mass on the trunk, left knee position, right knee position, left foot vicinity position, right foot vicinity position, trunk lower reference point, and trunk upper reference point. The dynamic system may exist. Thus, when calculating the target torque, mass is concentrated on the trunk, left knee position, right knee position, left foot vicinity position, right foot vicinity position, trunk lower reference point, and trunk upper reference point. It is preferable to calculate data indicating the target torque using an existing six-mass dynamic system. It is possible to use a dynamic system that gives one mass point to the trunk, and it is not always necessary to divide it into a lower trunk reference point and an upper trunk reference point, but a lower trunk reference point and an upper trunk Since the trunk position / posture can be accurately reflected by dividing the reference points, the calculation accuracy of the target torque is improved.

  Furthermore, the dynamic system is such that the total mass of the left shin lower half mass, the left ankle mass, and the left foot mass is concentrated at a position near the left foot, and the right shin lower half mass and the right ankle mass are located near the right foot. And the total mass of the right foot mass is concentrated, the total mass of the upper left shin, the mass of the left knee joint and the mass of the lower left thigh are concentrated at the left knee position, and the right shin is positioned at the right knee position. The total mass of the upper half mass, right knee joint mass, and right lower leg half mass is concentrated, and the upper left thigh mass, left hip mass, right upper thigh mass, and right hip mass are The dynamic mass system may be configured such that the total mass of the lower half of the trunk is concentrated and the upper half of the trunk is concentrated at the upper reference point. As schematically shown in FIG. 2, the total mass M1 of the mass of the left lower shin half 324, the mass of the left ankle 322, and the mass of the left foot 320 is concentrated at the position near the left foot, and the position near the right foot The total mass M3 of the mass of the lower right shin half 340, the mass of the right ankle 338, and the mass of the right foot 336 is concentrated, and the mass of the upper left shin half 326 and the mass of the left knee joint 328 are located at the left knee position. The total mass M2 of the mass and the mass of the left lower leg half 330 is concentrated, and the total mass of the mass of the right upper shin half 342, the mass of the right knee joint 344, and the mass of the right lower leg half 346 is located at the right knee position. M4 is concentrated, and the mass of the upper left half 332, the mass of the left hip joint 334, the mass of the upper right thigh half 348, the mass of the right hip joint 350, and the mass of the lower trunk half 352 are located at the lower trunk reference point. The total mass M5 is concentrated and the upper half of the trunk is at the upper trunk reference point. Using a dynamical system mass M6 54 are concentrated, it is preferable to calculate data indicating the target torque. Although this dynamic system is simple, it accurately models the robot and improves the calculation accuracy of the target torque without impairing the simplification of the calculation.

  Further, the height of the trunk lower reference point and the trunk upper reference point may be set to a height that maintains the magnitude of the trunk inertia moment around the horizontal axis. In particular, the heights of the lower trunk reference point and the upper trunk reference point are preferably set to a height that maintains the magnitude of the trunk inertia moment about the horizontal axis. The trunk has a large mass, and it is preferable to consider the moment of inertia in calculating the target torque. In particular, the inertia around the horizontal axis greatly affects the target torque as compared to the inertia around the vertical axis. The height of the lower trunk reference point and upper trunk reference point is set to a height that maintains the magnitude of the trunk inertia moment around the horizontal axis, that is, the height that provides the actual trunk inertia moment. Therefore, the target torque can be calculated in consideration of the moment of inertia of the trunk around the horizontal axis.

  The control method of the legged robot according to the present invention is a model of the legged robot "dynamic system in which mass is concentrated at the body, the left knee position, the right knee position, the position near the left foot, and the position near the right foot" A step of calculating a target torque of a joint of the legged robot, a step of calculating a target rotation amount of each joint necessary for realizing the position and orientation of the legged robot, and the calculated target A step of adjusting a command torque to be applied to the motor of the actuator based on a rotation amount and a detected deviation of the actual rotation amount, and in the adjustment step, the calculated target torque is added to the command torque It is.

  Further, another aspect of the legged robot control method according to the present invention is a model of a legged robot. “Mass is concentrated on the body, left knee position, right knee position, left foot vicinity position, and right foot position. Calculating the target torque of the joint of the legged robot using the existing dynamic system, and calculating the target rotation amount of each joint necessary to realize the position and orientation of the legged robot; Adjusting the command torque to be applied to the motor of the actuator based on the calculated target rotation amount and the detected deviation of the actual rotation amount, and comparing the calculated target torque with the command torque Or comparing the calculated target torque with the detected actual torque, and estimating a friction characteristic value of the actuator based on the comparison result.

  Furthermore, another aspect of the legged robot control method according to the present invention is a legged robot control method in which the rigidity of a part of the legged robot is locally reduced and the low rigidity part acts as a spring element. The legged robot is modeled using the “dynamic system in which mass is concentrated at the body, left knee position, right knee position, left foot vicinity position, and right foot position”. Calculating a target torque of each joint, calculating a target rotation amount of each joint necessary to realize the position and orientation of the legged robot, the calculated target rotation amount, and the detected actual amount Based on the rotation amount deviation, the step of adjusting the command torque applied to the motor of the actuator, the calculated target torque, the deviation of the detected actual torque, and the spring constant of the spring element, Calculating a correction amount required to correct the target rotation amount of the actuator for driving the low-rigidity portion, the adjustment step is a calculated said correction amount intended to be added to the target rotation amount.

According to the present invention, it is an object of the present invention to provide a legged robot capable of stably realizing walking / running operation by maintaining good track followability, and a control method thereof.
That is, the first object of the present invention is to provide a legged robot that can maintain good track followability even when the control gain is reduced, and a control method therefor.
In addition, the present invention provides a leg that can maintain the accuracy of the calculated target torque satisfactorily by identifying the friction characteristic value at an arbitrary timing even when the friction characteristic changes over time. A second object of the present invention is to provide a robot and its control method.
Furthermore, a third object of the present invention is to provide a legged robot capable of improving the position followability of the center of gravity without impairing the impact absorbing effect by the low rigidity portion, and a control method thereof.

  The main features of the legged robot according to the present invention described below will be described first. FIG. 1 is a control block diagram showing main features of a legged robot according to the present invention. As shown in FIG. 1, the legged robot according to the present invention moves by driving an actuator motor 8. The motor driver 6 of the legged robot drives the motor 8 based on target torque data and target joint angle data applied to each joint.

  More specifically, the legged robot inputs and stores data indicating the left foot position / posture, right foot position / posture, and trunk position / posture, and data indicating the trunk position that realizes the position / posture. Then, from the data group, a legged robot was modeled using a "dynamic system in which mass is concentrated at the torso, left knee position, right knee position, left foot position, and right foot position". Calculate the target torque of the robot joint. Further, the legged robot calculates the target rotation amount of each joint necessary to realize the position and orientation from the data group. For example, a target joint angle is calculated as the target rotation amount. A method for calculating the target torque and target joint angle of the joint using the dynamic system described above will be described later.

  The legged robot according to the first embodiment adjusts the command torque to be applied to the motor 8 of the actuator based on the calculated target rotation amount and the detected deviations (62, 64) of the actual rotation amount. Here, the calculated target torque is added to the command torque (66). For example, an actual angle or an actual angular velocity is detected as the actual rotation amount. These actual rotation amounts are detected by the encoder 10 or the like.

  The legged robot according to the second embodiment detects the actual torque of the actuator. The actual torque is detected by an ammeter 12 that energizes the motor 8 or the like. Then, the calculated target torque is compared with the command torque, or the calculated target torque is compared with the detected actual torque. The legged robot estimates the friction characteristic value of the actuator based on the comparison result.

  The legged robot according to the third embodiment is a legged robot in which the rigidity of the part of the legged robot is locally reduced and the low-rigidity part acts as a spring element. The legged robot is required to correct the target rotation amount of the actuator that drives the low rigidity portion based on the deviation (68) between the calculated target torque and the detected actual torque and the spring constant of the spring element. Calculate the correct correction amount. Here, the correction amount is added to the target rotation amount (70).

  The legged robot according to the fourth embodiment sets and stores a limit value for the command torque. Then, the legged robot compares the command torque with the limit value, and when the command torque exceeds the limit value, outputs (72) a drive signal corresponding to the limit value to the motor 8 of the actuator.

  The robot according to the fifth embodiment presets and stores a predetermined value that can drive the actuator, compares the target torque with the predetermined value, and if the target torque exceeds the predetermined value, The target torque is changed to a predetermined value.

  According to the present invention, walking / running can be realized stably by maintaining good trackability. In particular, it is superior to the prior art in terms of servo follow-up and impact mitigation at landing. Further, in the present invention, the target torque is calculated using the dynamic system shown in FIG. By using such a simple dynamic system, the trajectory and the target torque can be calculated in a short time.

  Next, a method for calculating the target torque and target joint angle of the joint will be described. A method for creating control data (gait data) necessary for a biped robot to walk with two legs will be described with reference to FIG. This gait data creation device 290 is used prior to actually walking the biped walking robot, and creates control data necessary for actually walking the biped walking robot. The gait data created by the gait data creation device 290 is transmitted to the biped robot. The biped robot walks by operating according to the transmitted gait data. The gait data creation device 290 can create gait data at a sufficiently high speed, and can create gait data substantially in real time. While watching how the walking robot is walking, the next step is actually performed after the person specifies the next step using the joystick etc. Since gait data for one step can be created, gait data can be created substantially in real time.

  The operator inputs data such as a course that the walking robot wants to walk into the walking command data (gait data) creation device 290. Specifically, five vectors and one scalar quantity shown in the lower left (A) of FIG. 3 are input. A vector C is a left foot position vector, which indicates the left foot position of the robot as viewed from the coordinate origin O. A vector D is a left foot posture vector and indicates the direction of the left foot. A vector E is a right foot position vector and indicates the right foot position of the robot as viewed from the coordinate origin O. A vector F is a right foot posture vector and indicates the direction of the right foot. A vector R is a trunk posture vector and indicates the direction of the trunk of the robot. A direction in which a member corresponding to a human spine extends is shown. When the body of the robot is separated into the waist (lower body) and the upper body, the waist corresponds to the trunk, and the trunk posture vector R defines a direction perpendicular to the waist. Q is the trunk height, which corresponds to the height of the waist position that humans say. The operator successively inputs these five types of vectors and one type of scalar quantity in time series. Data input support software for this purpose has been developed. By specifying data at main timing, intermediate data in time series is complementarily calculated. In addition, the input amount can be reduced by instructing to repeatedly use the input data for one step.

  In addition, the operator may instruct the walking course, walking speed, etc. with a joystick or the like. By storing the standard foot posture vector change pattern, trunk posture vector change pattern, and trunk height change pattern for one step, the operator can specify the walking course and walking speed with a joystick, etc. The one-step five-type vector C (t), D (t), E (t), F (t), R (t) and one-type scalar quantity Q (t) can be generated and stored. Five types of vectors C (t), D (t), E (t), F (t), R (t) and a seed scalar quantity Q (t) of 1 necessary for creating gait data are: The data is input to the gait data creation device 290 and stored in the storage device 292.

  The operator further inputs a target ZMP position vector according to a time series. In this case, since the distinction between the left foot contact state, the two foot contact state, and the right foot contact state is known from the vectors C to F, the position in the left foot contact surface is specified in the left foot contact state, and the right foot contact surface in the right foot contact state. Specify the position within. By using the vectors C to F, a target ZMP position vector (ZMP * vector) can be designated. By controlling so that the actual ZMP observed at the time of walking coincides with the target ZMP, the robot can walk without falling down.

  The gait data creation device 290 shown in FIG. 3 calculates the ZMP that will actually occur based on the five types of vectors and one type of scalar input by the operator, determines the deviation from the target ZMP, Gait data is created by calculating a trunk position vector P that yields the target ZMP from the ZMP deviation. Furthermore, the gait data creation device 290 creates gait data by calculating the target torque of each joint based on the five types of vectors and one type of scalar input by the operator. The means 294 generates data indicating one trunk position that realizes the position and posture of the robot defined by five types of vectors and one type of scalar. Since there are a plurality of trunk positions that realize the position and posture of the robot defined by five types of vectors and one type of scalar, it cannot be uniquely determined. Here, an arbitrary one is adopted. When the trunk position vector P is generated, six types of vectors C, D, E, F, P, and R are prepared, so that the position and posture of the robot are uniquely determined.

  The trunk vector P used here is temporarily used and is corrected by the means 310 described later, and may be rough. It may be specified by a person and input to the gait data creation device 290 for storage, or may be generated by the gait data creation device 290.

  The trunk posture vector R, the toe position vectors C and E, the toe posture vectors D and F, the trunk height Q, and the target ZMP vector can be standardized with respect to the trunk position vector P. If the change pattern for one step is standardized, the person inputs the five types of vectors and one type of scalar required by the gait data creation device 290 by specifying the trunk position vector P. Can do. The trunk position vector P can be specified using a joystick or the like. It is possible to input the trunk position vector P in real time by operating a joystick or the like while observing a walking robot. The trunk position vector P can be input in real time, and the five types of vectors and one scalar quantity required by the gait data generation device 290 can be input in real time. If gait data can be created at a higher speed in time for real-time control, walking of a biped robot can be specified in real time.

  If the trunk position P and posture R are determined, the positions of the left and right hip joints are determined. If the left foot position C and the left foot posture D are determined, the position of the left ankle is determined. If the position of the left hip joint and the position of the left ankle are determined, the position G of the left knee is determined. Since the length of the left thigh and the length of the left shin are known and the left knee joint has only one degree of freedom of rotation, the position of the left knee is unique if the position of the left hip joint and the position of the left ankle are determined. To be determined. Similarly, the position of the right ankle is determined from the right foot position E and the right foot posture F. The right knee position H is determined from the position of the right hip joint and the position of the right ankle. The means 296 calculates the left knee position vector G and the right knee position vector H using the above geometric relationship. In means 298, the ZMP position of the robot is calculated. Here, the 6-mass system model shown in FIG. 2 is used.

  In the dynamic system having six mass points, the total mass M1 of the mass of the lower left shin half 324, the mass of the left ankle 322, and the mass of the left foot 320 is concentrated in the vicinity of the left foot. The total mass M3 of the mass of the lower right shin half 340, the mass of the right ankle 338, and the mass of the right foot 336 is concentrated in the vicinity of the right foot. The total mass M2 of the mass of the left upper shin half 326, the mass of the left knee joint 328, and the mass of the left lower leg half 330 is concentrated at the left knee position. A total mass M4 of the mass of the right upper shin half 342, the mass of the right knee joint 344, and the mass of the right lower leg half 346 is concentrated at the right knee position. The total mass M5 of the mass of the left upper thigh half 332, the mass of the left hip joint 334, the mass of the right upper thigh half 348, the mass of the right hip joint 350, and the mass of the lower trunk half 352 is concentrated on the lower trunk reference point. Exists. It is assumed that the mass M6 of the upper half 354 of the trunk is concentrated on the upper trunk reference point.

  The position of the mass point M1 may be in the vicinity of the left foot, may be deviated from the left ankle joint position, or may be deviated from the end point of the left foot position vector C. The position of the mass point M3 may be in the vicinity of the right foot, may be deviated from the right ankle joint position, or may be deviated from the end point of the right foot position vector E. On the other hand, it is extremely effective that the position of the mass point M2 is at the position of the left knee joint. By assuming that the mass is concentrated at the position of the left knee joint, the calculation of the ZMP position is possible. This greatly simplifies the calculation of the trunk position correction amount. Similarly, it is extremely effective that the position of the mass point M4 is at the right knee joint position. The mass points M5 and M6 exist on the human spine. The height should be set so that the value obtained by calculating the moment of inertia around the horizontal axis of the trunk using the concentrated masses M5 and M6 matches the actual moment of inertia around the horizontal axis of the trunk. Is preferred.

  The means 298 in FIG. 3 uses the velocity and acceleration of the mass points M1 to M6 and the moment of inertia around the horizontal axis due to the concentrated masses M5 and M6 based on the change in the position and posture of the robot determined by the means 296. ZMP position occurring at

  The gait data creation device 290 shown in FIG. 3 includes means 292 for inputting and storing data indicating the target ZMP position, and the ZMP position calculated by the means 298 and the target ZMP stored in the means 292. The deviation from the position can be calculated (means 300).

  The gait data creation device 290 calculates a trunk position vector correction amount necessary for eliminating the ZMP deviation calculated by the means 300, and calculates a ZMP position equal to the target ZMP position. A means 310 for calculating P is provided. The means 310 corrects how much the data indicating the trunk position is corrected, how much the data indicating the left knee position and the right knee position is changed, and how much the data indicating the ZMP position is changed. The correction amount of the data indicating the position is obtained. According to the correction means 310, it is not necessary to repeatedly calculate and converge the correction amount, and the ZMP deviation can be eliminated by a single correction. In calculating the movement amount of the left knee position and the right knee position from the correction amount of the trunk position, the movement amount of the left knee position and the right knee position is calculated by multiplying the correction amount ΔP of the trunk position by a coefficient. This further simplifies the calculation. By using the proportional coefficient between the trunk height and the knee position height, the movement amounts of the left knee position and the right knee position can be easily calculated from the trunk position correction amount ΔP.

  As described above, a ZMP position that substantially coincides with the target ZMP position is calculated. However, there is a case where the ZMP deviation is not sufficiently solved by correcting the trunk position vector P once. In this case, the ZMP deviation can be further eliminated if the means 298 and the subsequent steps are used again and corrected twice. Thus, it has been verified that gait data that closely follows the target ZMP can be calculated in a short time with a small amount of calculation.

  Then, the target torque calculation method of each joint is demonstrated. The means 312 in FIG. 3 uses the speed and acceleration of the mass points M1 to M6 and the moment of inertia around the horizontal axis due to the concentrated masses M5 and M6 based on the change in the position and posture of the robot determined by the means 310. The target torque of each joint is calculated. When calculating the target torque of each joint, the target torque can be calculated according to the walking / running motion pattern of the robot. In the present invention, the target torque of the joint is calculated according to whether the leg of the robot is standing, swinging, or in the air. From the data input to the gait data creation device 290, it can be determined whether the robot leg is standing, swinging, or in the air.

  FIG. 4A is a diagram illustrating a target torque calculation method for a joint provided on the stance side when the robot leg is in the stance state. When the leg of the robot is a stance, the target torque of the joint of the stance leg of the robot is calculated using another mass point excluding the mass point existing in the toe direction from the joint of interest of the stance. Here, the means 312 assumes that the ankle joint 338 of the leg that is the supporting leg is fixed, and is not a target torque calculation target. Therefore, the target torque of the right knee joint 344 and the right hip joint 350 provided on the stance side is calculated. Hereinafter, a joint that is a target torque calculation target is referred to as a target joint.

  More specifically, first, the means 312 uses the speed and acceleration of the mass points M1 to M6 and the moment of inertia around the horizontal axis due to the concentrated masses M5 and M6 from the change in the position and posture of the robot determined by the means 310. Then, the force F (inertial force and gravity) generated at each of the mass points M1 to M6 is calculated. Here, when the target joints are the right knee joint 344 and the right hip joint 350, the means 312 generates forces F (F1, F2, F5, F6) generated at the four mass points M1, 2, 5, 6. calculate. If the robot has a toe joint (not shown) on the right foot, the means 312 calculates forces F (F1 to F1) generated at the six mass points M1 to M6 in order to calculate the target torque of the toe joint. 6) is calculated.

Next, the means 312 calculates a moment T acting on the target joint from each mass point. The following formula is used to calculate the moment that the force F from the moving mass point (the force F from the free leg / body side mass point) brings to the joint of interest (the limb side joint). Here, T is a moment acting on the target joint, R is a (relative) position vector from the target joint to the mass point, and F represents a force (inertia, gravity) acting on the mass point. R and F are vectors having values in the x, y, and z directions, and the vector T is calculated by calculating the outer product of these vectors R and F.

Next, the means 312 calculates the sum of all the mass points to be calculated for the moment T, and converts the total value into the torque value of the motor at the joint of interest. The target torque value [N · m] at this time is shown in the following equation. Here, inrt is the inertia of the motor of the actuator and the speed reducer used for the motor, and fric is the friction of the motor and the speed reducer. For example, as shown in FIG. 4A, the target torque is a torque required to advance when the robot is moving forward. The target torque is also a torque necessary to support the robot's own weight. “inrt (inertia)” is a physical value determined by an object (its mass and shape), and can be determined from a specification table or the like for the motor of the actuator. The fric (friction) can be basically determined from a specification table or the like. Note that fric can also be calculated using a friction characteristic value estimated by an estimation method described later.

  FIG. 4B is a diagram illustrating a target torque calculation method for a joint provided on the free leg side when the leg of the robot is at the free leg. When the leg portion of the robot is a free leg, the target torque of the joint of the free leg of the robot is calculated using the mass points existing in the toe direction from the joint of interest of the free leg. The means 312 calculates target torques of the left hip joint 334, the left knee joint 328, and the left ankle joint 322 provided on the free leg side. Only two mass points are arranged on the leg portion at the time of the free leg. For example, when the target joint is the left hip joint 334, the means 312 calculates forces F (F1, F2) generated at the two mass points M1 and M2. Thereafter, the target torque of the left hip joint 334 is calculated in the same manner as in the above-mentioned standing posture.

  FIG. 4C is a diagram illustrating a target torque calculation method for each joint when the leg of the robot is in the aerial phase. When the robot is in the aerial phase floating in the air, the target torque of the target joint of the robot is calculated using the mass points existing in the toe direction from the target joint of the leg. The target torque of each joint in the aerial phase is preferably calculated from the direction in which the number of mass points existing on the toe side of the target joint is smaller. That is, the target torque is calculated in consideration of the mass point existing in the toe direction from the target joint. In the Newton-Euler method, either end is set as a fixed end, and torque is calculated in consideration of force transmission toward the free end. Since there is no fixed part in the aerial phase, it is considered that the target torque of each joint cannot be calculated if the Newton-Euler method is applied as it is. However, since the force F acting on each mass point is calculated in consideration of the dynamics using the dynamic system described above, one end is fixed even in the aerial phase as in the case of standing and swinging. It can be calculated assuming that. For this reason, even if there is no fixed end, the target torque of each joint can be calculated. The aerial phase is a situation when the robot is traveling. For example, the robot needs to largely swing the hip joint forward during the traveling. In other words, the target torque of the joint in the aerial phase is, for example, the torque required to drive the hip joint as such.

  Next, a control device for a biped robot will be described with reference to FIG. FIG. 5 shows a configuration of a control device that controls biped walking of the robot, and the trunk position vector P created by the walking command data creation device 290 (this is input by the operator using the walking command data creation device 290). ), A trunk posture vector R, a left foot tip position vector C, a left foot tip posture vector D, a right foot tip position vector E, and a right foot tip posture vector F are input, and inverse kinematics calculation is performed. A calculation device 304 is provided. However, the trunk position vector P is not only corrected by the walking command data creation device 290, but is further modified to perform “inverted pendulum control” and “follow control”, which will be described later, to the calculation device 304. Entered. Also for the trunk posture vector R, a process for correcting the target trunk posture vector is performed and input to the calculation device 304 so that the actual trunk posture vector matches the target trunk posture vector.

  The calculation device 304 inputs the trunk position vector P (which has been corrected), the trunk posture vector R (which has also been corrected), the left foot position vector C, and the left foot position vector D. Then, based on the right foot tip position vector E and the right foot tip posture vector F, the rotation angle θ of each joint necessary to realize the position and posture described by each input vector is calculated. The rotation angle is calculated by the rotation amount of the motor that rotates each joint. Although only one actuator is displayed in FIG. 5, the motor rotation amounts of all actuators are calculated. In this calculation, an inverse kinematics operation is performed. The target rotation amount θ calculated for each actuator is input to the driver 306 for each actuator. Each driver adjusts the torque applied to the motor 308 of each actuator based on the deviation between the target rotation amount and the actual rotation amount. The actual rotation amount is detected by an encoder 310 provided for each motor of the actuator. With this feedback control, the motor 308 of the actuator is feedback controlled so that the actual rotation amount matches the target rotation amount.

  The target ZMP is input to the inverted pendulum model 334. The actual trunk position vector P is also input to the inverted pendulum model 334. The actual trunk position vector P is obtained by calculating the signal of the gyro 328 provided on the trunk of the robot by the calculation device 330.

  The inverted pendulum model 334 uses the target ZMP and the actual trunk position vector P to calculate the inclination φ from the vertical of the inclination line from the target ZMP to the trunk position vector P. The calculated φ is input to the ZMP correction amount calculation device 336. The ZMP correction amount calculation device 336 calculates the ZMP correction amount according to the equation: ΔZMP = k1 × φ + k2 × dφ / dt. In other words, the ZMP correction amount is calculated by adding the value proportional to the tilt angle φ and the value proportional to the time derivative of the tilt angle φ. The calculated ZMP correction amount ΔZMP is added to the input target ZMP, and the target ZMP is corrected by ΔZMP. As described above, the process of correcting the target ZMP by ΔZMP is usually referred to as inverted pendulum control.

  When a person walks, particularly when traveling at a high speed, the human body makes it easier to travel by tilting the trunk forward. The center of gravity is moved forward by tilting the trunk forward, and the foot is moved forward so as to follow the center of gravity moved forward. Inverted pendulum control corresponds to this kind of human control, and controls to tilt the trunk position forward.

  The target ZMP (340) corrected by the inverted pendulum model is compared with the actual ZMP (326), and the deviation (342) is calculated. The actual ZMP (326) is calculated by calculating the outputs of a plurality of force sensors 322 provided on the soles of the robot by the calculation device 324. The deviation 342 is multiplied by a gain k3 and converted into a dimension of the trunk position vector P. The deviation ΔP converted into the dimension of the trunk position vector P is added to the trunk position vector P. The process of converting the deviation (342) between the target ZMP (340) and the actual ZMP (326) into a dimension of the trunk position vector P and adding it to the trunk position vector P is usually referred to as “profile control process”. .

  On the surface on which the robot walks, a trunk position vector P, a trunk posture vector R, a left foot tip position vector C, a left foot tip posture vector D, a right foot tip position vector E, a right foot tip posture vector F, At the stage of determining the target ZMP, unexpected irregularities exist, and the robot's feet may step, for example. This is equivalent to kicking a person. At that time, the human is prevented from falling by bending the knee and sending the position of the waist forward. The profile control process corresponds to this kind of human control, and controls to move the trunk forward. When the fall is prevented by bending the knee and translating the trunk forward, it is not possible to continue walking. After preventing a fall, it is necessary to extend the knee and return to the normal walking posture. The inverted pendulum model executes control corresponding to it. Figuratively speaking, the follow model is equivalent to body flexibility, and the inverted pendulum model is equivalent to the ability to maintain walking posture.

  It should be noted that the target ZMP is corrected by the inverted pendulum control system 334, and the target trunk position vector P is corrected by the deviation (342) between the target ZMP and the actual ZMP by the follow control system. The amount may be limited.

  Follow-up control corresponds to body flexibility. The greater the gain k3 (conversion device 344) for the profile control, the more flexible the limb of the robot changes its posture, and the received shock is reduced. Therefore, it is preferable that the profile control works strongly when receiving a shock such as when landing. On the other hand, if the profile control is working strongly at all times, the posture may be disturbed by inertial force when standing on one foot, and landing may occur at a timing that is not scheduled to land, resulting in a large landing shock. The setting of the value of the gain k3 for the profile control has the above-mentioned conflicting elements, and it is difficult to adjust to the optimum value. In order to solve this reciprocal relationship, the value of the gain k3 for follow-up control may be increased or decreased in accordance with the state of the robot.

Embodiment 1 of the Invention
As shown in FIG. 6, the robot according to the first embodiment gives a command to be applied to the motor 8 of the actuator based on the calculated target rotation amount and the detected deviation (62, 64) of the actual rotation amount. Adjust the torque. Here, the calculated target torque is added to the command torque (66). For example, an actual angle or actual speed is detected as the actual rotation amount. These actual rotation amounts are detected by the encoder 10 or the like.

  In the prior art, the angle deviation and angular velocity deviation from the encoder 10 and the like are fed back with respect to the target joint angle (position loop, velocity loop), and these deviations are converted into torque command values. It was given to drive. At this time, in order to realize impact absorption at the time of landing, the gain (Kp, Kv) in the figure is reduced, the command torque applied to the motor 8 is reduced (that is, the joint is softened), and the target angle and the actual angle are set. Control was performed to allow deviation. However, if the follow-up performance of the motor 8 is poor with respect to the target joint angle, except for the standing stance at the time of landing, the robot cannot achieve the target trajectory (posture) and is likely to become unstable. It was a thing. The gain Kp is a gain for converting a position deviation (angle deviation) into a torque value, and the gain Kv is a gain for converting a speed deviation (angular speed deviation) into a torque value.

  As shown in FIG. 6, the robot according to the first embodiment inputs the above target torque as a feed forward torque (addition 66) to the final output of the position / speed control loop, thereby obtaining a gain (Kp , Kv) can be maintained, the followability of the target joint angle can be maintained well. In other words, by adding the target torque, the value of the gain Kp can be set to a small value without making it larger than necessary. The fact that the gain can be set to a small value means that rigidity can be lowered and the joint can be controlled softly. This makes it easier to absorb the impact when landing. However, if the gain is set to a value smaller than necessary, the joint becomes unnecessarily soft and may fall over, so that the amount is compensated with the target torque.

Embodiment 2 of the Invention
The robot according to the second embodiment detects the actual torque of the actuator. The actual torque is detected by an ammeter 12 that energizes the motor 8 or the like. Then, the calculated target torque is compared with the command torque, or the calculated target torque is compared with the detected actual torque. The robot estimates the friction characteristic value of the actuator based on the comparison result.

  The friction characteristics of the actuator change when the robot is reassembled, overhauled, or over time. Due to the overhaul, the degree of assembly is different from the previous one, so for example, the friction is increased. Even with the passage of time, friction changes due to factors such as less grease. According to the robot according to the second embodiment, even when the friction characteristic changes due to overhaul, time lapse, temperature change, etc., the identification of the friction characteristic value is performed at an arbitrary timing, so that the normal friction is always maintained. The value (friction torque) can be determined, and the accuracy of the calculated target torque can be maintained well.

  Hereinafter, a friction characteristic value estimation method will be described with reference to FIGS. FIG. 7 is a flowchart for explaining the friction characteristic parameter estimation processing. First, the robot performs a normal walking / running operation based on the gait data created by the gait data creation device 290 (step S101). At this time, the robot acquires and stores torque data during the walking / running operation. The acquired torque data may be a command torque for driving the motor 8 or an actual torque that is actually detected. Next, the robot calculates a target torque as described above (step S103). Here, in the calculation of the target torque, the friction fric of the motor and the speed reducer described above is calculated using a nonlinear friction characteristic model shown in FIG. FIG. 8 is a diagram showing the friction torque (T) generated in the speed reducer according to the angular velocity (ω) of the motor. Kc is a static friction coefficient, and Kv is a viscous friction coefficient. Here, in the range where the angular velocity ω is 0 or more, the relationship T = Kv × ω + Kc is established. As the initial values of the friction characteristic parameters Kc and Kv, values in a catalog specification table can be used.

  Since these friction characteristic parameters Kc and Kv change under the influence of various factors as described above, the robot compares the calculated target torque with torque data (command torque or actual torque) (step S104). Based on the comparison result, the friction characteristic parameter of the actuator is determined (step S105). FIG. 9 is a diagram illustrating changes in the calculated target torque and acquired torque data. If the difference between the target torque and the torque data is due to the angular velocity ω, the viscous friction coefficient Kv may be changed. If the difference occurs when the angular velocity ω is zero, the static friction coefficient Kc may be changed. That's fine. That is, in the time series comparison of the target torque and the torque data, the difference when the angular velocity ω is 0 indicates the static friction coefficient Kc, and the difference at other time points is proportional to Kv. Parameters Kc and Kv are determined. For example, the friction characteristic parameters Kc and Kv can be determined by selecting a parameter that minimizes the sum of squared deviations.

  The comparison between the target torque and the torque data is preferably performed using data at the time of the free leg. When standing, when comparing with torque data, the external force from the floor must be taken into account, and in some cases, the component from the floor reaction force becomes larger than the angular velocity ω, and the influence of friction is relatively marked. There is a risk. For this reason, there exists a problem that calculation becomes complicated. On the other hand, since it is not necessary to consider the external force from the floor at the time of the free leg, the influence of friction and inertia is large, and those components become relatively remarkable, so the friction characteristic parameter can be determined more easily. .

  In addition, as described above, the friction characteristic parameter estimation process may be implemented in advance by installing a parameter determination program in the robot and executing the parameter determination process in real time during the operation of the robot. The friction characteristic parameter may be determined by comparing the two data offline. By configuring the friction characteristic parameter estimation process in real time, the friction parameter determination process can be executed at any timing, and the friction characteristic parameter is efficiently maintained while maintaining the target torque calculation accuracy at a high level. An estimation process can be executed. The parameter estimation process may be determined when the operator compares the torque data with the target torque and the two are greatly deviated.

Embodiment 3 of the Invention
As shown in FIG. 12, the robot according to the third embodiment is a robot in which the rigidity of the robot part is locally reduced and the low rigidity part acts as a spring element. The robot performs correction necessary to correct the target rotation amount of the actuator that drives the low rigidity portion, based on the deviation (68) between the calculated target torque and the detected actual torque, and the spring constant of the spring element. Calculate the quantity. Here, the correction amount is added to the target rotation amount (70).

  The servo tracking of the robot is good, the rigidity of the mechanism (link, joint, reducer, etc.) can be made sufficiently high, and if there is no disturbance, the robot will follow the target value (center of gravity, joint angle, etc.). Stable operation is possible. However, in reality, it is difficult to adjust the servo following ability, and there is a problem that a portion having low mechanical rigidity is easily bent. As for the servo, it can be adjusted to some extent by experiments. However, with respect to the mechanism, the improvement in rigidity leads to an increase in weight, and there is a problem that extra time is required for design change and reassembly. For this reason, the low-rigidity part has an effect of absorbing the impact, but can also cause the posture of the robot to collapse.

  For example, with respect to the target posture shown in FIG. 10 (a), as shown in FIG. 10 (b), the low-rigidity part of the robot (for example, the part of the sole, which is indicated by hatching in the figure) bends. The gravity center position G may be shifted. Therefore, as shown in FIG. 10C, the robot according to the third embodiment adds the angle correction in a feed-forward manner to a joint (for example, an ankle joint) near the low-rigidity portion, thereby This improves the followability.

  FIG. 11 is a diagram exemplifying a situation in which the sole portion of the robot (the portion indicated by hatching in the drawing) is bent. FIG. 11 shows an example of only one mass point and one foot, but the same phenomenon occurs in the case of a full model of the robot. FIG. 11A is a diagram illustrating a case where the ZMP is located substantially at the center of the sole. On the other hand, FIG. 11B is a diagram exemplifying a case where the load balance is biased and the low-rigidity portion is bent because the ZMP moves to the end portion of the sole due to the inertial force of the mass point.

When the movement of the ZMP is mainly caused by the low rigidity of the sole, as shown in FIG. 11C, the deflection of the sole can be corrected by the joint closest to the sole. That is, the ankle joint closest to the low-rigidity part is assumed to be a spring model, and the correction angle θ for correcting the deflection of the low-rigidity part is calculated by the following equation. Here, Tref is the target torque calculated as described above, Tmes is the actual torque detected as described above, and K is a spring constant. The spring constant K can be determined in advance or in real time by examination with an actual machine or simulation.

  As shown in FIG. 12, the robot according to the third embodiment adds (70) the correction angle θ corresponding to the deviation (68) between the calculated target torque and the detected actual torque to the target joint angle. . In this way, by correcting the deflection of the low-rigidity part at the joint angle level based on the relationship between the target torque and the actual torque, the position followability of the center of gravity can be improved without impairing the impact absorption effect of the low-rigidity part. Can do.

Embodiment 4 of the Invention
The robot according to the fourth embodiment sets and stores a limit value for the command torque. Then, the robot compares the command torque with the limit value, and when the command torque exceeds the limit value, outputs a drive signal corresponding to the limit value to the motor 8 of the actuator (72).

  In the first and third embodiments described above, the robot calculates a torque command value based on the target joint angle and the target torque, and drives the motor of the actuator by giving the torque command value. When driving the motor of the actuator, for example, when the body of the robot is tilted forward, the robot needs to step on the sole of the foot in order to restore its posture. However, since the inertia acting on the trunk is greater than the force pushing the sole, there is a problem that if the force pushing the sole becomes too large, the sole peels off. Therefore, the robot according to the third embodiment provides a torque limit for the torque command value given to the motor, restricts the torque given to the ankle joint in advance before the sole is peeled off, and generates more torque than necessary. To prevent that.

With reference to FIG. 13, the limit value of the torque command value applied to the ankle joint will be described. As shown in the figure, first, the balance equation shown below is established for the torque τ * applied to the ankle joint. In consideration of the moment around the ankle, when ZMP * becomes ZMPmax, the torque τmax when the foot rotates (the foot peels) with ZMPmax as a fulcrum is calculated. Here, the torque has a positive value in the counterclockwise direction and a negative value in the clockwise direction. F1 and F2 are substantially equal to the inertial force acting on the center of gravity. Since the gravity center position trajectory is known, F1 and F2 can be calculated by differentiating the value. Dmax and h represent the positional relationship between the ankle joint and ZMPmax shown in the figure, and can be determined from the mechanism of the mechanism. The allowable ZMP range (Dmax, Dmin) indicates a range in which ZMP can exist on the sole, and is determined by looking at the characteristics of the actual machine. Dmax indicates the range up to the foot rear portion where ZMP can exist when the front portion of the foot is not peeled off. Dmin indicates the range up to the front part of the foot where ZMP can exist when the rear part of the foot is not peeled off.

  FIG. 14 is a diagram illustrating calculated torque command value limit values (τmax, τmin). In the figure, the horizontal axis is the calculated target torque, and the vertical axis is the actually applied torque (torque command value). That is, the torque command value is limited so that the target torque does not exceed τmax, and the torque command value is limited so that the target torque does not fall below τmin.

  As shown in FIG. 1, the robot according to the fourth embodiment can prevent the foot from being detached from the floor by setting the torque limit 72 for the torque command value.

  In the embodiment described above, the ZMP allowable range D is calculated in advance, but the ZMP allowable range D may be changed in real time during the walking / running operation. Further, the joint for setting the torque limit is not limited to the torque command value of the joint close to the floor surface, and the torque limit may be set for the torque command value given to an arbitrary joint provided on the leg.

Embodiment 5 of the Invention
In the above-described embodiment, the calculated target torque is used as it is, but the present invention is not limited to this. That is, the robot according to the fifth embodiment sets and stores a predetermined value that can drive the actuator in advance, compares the target torque with the predetermined value, and the target torque exceeds the predetermined value. Alternatively, the target torque may be changed to a predetermined value. Here, the maximum torque value that the actuator can drive can be set as a predetermined value.

  The target trajectory may be discontinuous depending on the robot walking / running motion pattern, stabilization control, and the like. In such a case, depending on the joint, the target torque may exceed the specification of the actuator (motor). Since the actuator cannot be operated at a torque or angular velocity exceeding the maximum specification range of the motor, there is a problem that the track following ability deteriorates. The robot according to the fifth embodiment checks the magnitude of the calculated target torque and changes it so as to be within the motor specification range. The actuator specifications are generally expressed using a torque diagram shown in FIG. In the figure, the vertical axis represents the motor speed, and the horizontal axis represents the motor torque. The maximum specification of the motor is determined according to the rotation speed and torque. As shown in FIG. 15, when the calculated target torque (shown by a solid line in the figure) is within the maximum specification range, the target torque is output as it is. On the other hand, when the calculated target torque (indicated by a broken line in the figure) does not fall within the range of the maximum specification, the target torque value is changed and output.

  As a method for changing the target torque, for example, the following two methods can be adopted. First, when the target torque exceeds the maximum specification as shown in FIG. 16A, the target torque value is set to the maximum specification or a predetermined value smaller than the maximum specification as shown in FIG. Can be changed. A predetermined value smaller than the maximum specification can be given by the operator. Also, as shown in FIG. 17A, when the target torque exceeds the maximum specification, the target torque value (past data) one calculation cycle before is newly set as shown in FIG. It can be replaced with torque data (latest data).

Other embodiments.
In the above-described embodiment, a legged robot having two leg links has been described. However, the present invention can also be embodied as a legged robot having three or more leg links, or a leg having only one leg link. It can also be embodied as a type robot.

  In addition, the present invention is not limited to the above-described embodiments, and various modifications can be made without departing from the gist of the present invention described above.

It is a figure which shows the structure of the legged robot which concerns on embodiment of this invention. It is a figure which shows notionally the 6 mass point model which concerns on embodiment of this invention. It is a figure which shows the structure of the gait data creation apparatus which concerns on embodiment of this invention. It is a figure which shows the structure of the control apparatus which concerns on embodiment of this invention. It is a figure for demonstrating the target torque calculation method which concerns on embodiment of this invention. It is a figure which shows the structure of the control apparatus which concerns on Embodiment 1 of this invention. It is a flowchart of the friction characteristic parameter estimation process which concerns on Embodiment 2 of this invention. It is a figure which shows the nonlinear friction characteristic model which concerns on Embodiment 2 of this invention. It is a figure for demonstrating the friction characteristic parameter estimation process which concerns on Embodiment 2 of this invention. It is a figure for demonstrating the bending correction process which concerns on Embodiment 3 of this invention. It is a figure for demonstrating the bending correction process which concerns on Embodiment 3 of this invention. It is a figure which shows the structure of the control apparatus which concerns on Embodiment 3 of this invention. It is a figure for demonstrating the torque limit which concerns on Embodiment 4 of this invention. It is a figure for demonstrating the torque limit which concerns on Embodiment 4 of this invention. It is a figure for demonstrating the target torque change method which concerns on Embodiment 5 of this invention. It is a figure for demonstrating the target torque change method which concerns on Embodiment 5 of this invention. It is a figure for demonstrating the target torque change method which concerns on Embodiment 5 of this invention.

Explanation of symbols

6 Motor driver, 8 Motor, 10 Encoder, 12 Ammeter,
290 Gait data creation device,
328 Left knee joint, 334 Left hip joint, 344 Right knee joint, 350 Right hip joint,
320 left foot, 322 left ankle, 324 left shin lower half, 326 left shin upper half,
330 left lower thigh, 332 left upper thigh, 336 right foot, 338 right ankle,
340 Lower right shin half, 342 Upper right shin upper half, 346 Lower right thigh half, 348 Upper right thigh half,
352 Lower trunk half, 354 Upper trunk half

Claims (38)

Means for inputting and storing data indicating the left foot position / posture, right foot position / posture and trunk position / posture of the legged robot, and data indicating the trunk position realizing the position / posture;
From the data group, the leg robot is modeled using the “mechanical system in which mass concentrates at the torso, left knee position, right knee position, left foot position, and right foot position”. Means for calculating the target torque of the joint of the robot,
Means for calculating a target rotation amount of each joint necessary to realize the position and orientation from the data group;
Means for adjusting a command torque applied to the motor of the actuator based on the calculated target rotation amount and a deviation of the detected actual rotation amount;
The legged robot, wherein the adjusting means adds the calculated target torque to the command torque. Means for inputting and storing data indicating the left foot position / posture, right foot position / posture and trunk position / posture of the legged robot, and data indicating the trunk position realizing the position / posture;
From the data group, the leg robot is modeled using the “mechanical system in which mass concentrates at the torso, left knee position, right knee position, left foot position, and right foot position”. Means for calculating the target torque of the joint of the robot,
Means for calculating a target rotation amount of each joint necessary to realize the position and orientation from the data group;
Means for adjusting a command torque to be applied to the motor of the actuator based on the calculated target rotation amount and a deviation of the detected actual rotation amount;
Means for detecting the actual torque of the actuator;
Means for comparing the calculated target torque with the command torque, or comparing the calculated target torque with the detected actual torque;
A legged robot comprising means for estimating a friction characteristic value of the actuator based on a comparison result. 3. The legged robot according to claim 2, wherein the comparison means performs comparison using torque data at the time of a free leg. A legged robot that locally reduces the rigidity of a legged robot part, and the low-rigidity part acts as a spring element,
Means for inputting and storing data indicating the left foot position / posture, right foot position / posture and trunk position / posture of the legged robot, and data indicating the trunk position realizing the position / posture;
From the data group, the leg robot is modeled using the “mechanical system in which mass concentrates at the torso, left knee position, right knee position, left foot position, and right foot position”. Means for calculating the target torque of the joint of the robot,
Means for calculating a target rotation amount of each joint necessary to realize the position and orientation from the data group;
Means for adjusting a command torque to be applied to the motor of the actuator based on the calculated target rotation amount and a deviation of the detected actual rotation amount;
Means for detecting the actual torque of the actuator;
A correction amount necessary to correct the target rotation amount of the actuator that drives the low-rigidity portion based on the calculated deviation between the target torque and the detected actual torque, and the spring constant of the spring element A means of calculating
The legged robot characterized in that the adjusting means adds the calculated correction amount to the target rotation amount. The legged robot according to claim 4, wherein a sole portion of the legged robot acts as the spring element. The legged robot according to claim 4, wherein the actuator reduction gear acts as the spring element. The legged robot according to any one of claims 4 to 6, wherein the actuator that drives the low rigidity portion is an actuator that drives an ankle joint of the legged robot. Means for setting and storing a limit value for the command torque;
The apparatus further comprises means for comparing the command torque with the limit value and outputting a drive signal corresponding to the limit value to the motor of the actuator when the command torque exceeds the limit value. The legged robot according to any one of claims 1, 2, and 4. The allowable range in which ZMP can be allowed is calculated at the sole portion of the legged robot when standing, and a limit value for the command torque is set based on the calculated allowable range. 8. The legged robot according to 8. The target torque calculating means is
Means for calculating a target rotation amount of each joint necessary to realize the position and orientation from the data group;
Modeling the legged robot, using the "mechanical system in which mass is concentrated at the torso, left knee position, right knee position, left foot position, and right foot position" 5. The apparatus according to claim 1, further comprising means for calculating a target torque of a joint of the legged robot by calculating a position and acceleration of each mass point from a temporal change in the target rotation amount. The legged robot described in the section. When the leg portion of the legged robot is a standing leg, the target torque of the target joint is calculated using another mass point excluding the mass point existing in the toe direction from the target joint of the standing leg. The legged robot according to claim 10. When the leg portion of the legged robot is a free leg, a target torque of the target joint is calculated using a mass point existing in the toe direction from the target joint of the free leg. Item 12. The legged robot according to item 10 or 11. When the legged robot is in the aerial phase floating in the air, the target torque of the target joint is calculated using a mass point existing in the toe direction from the target joint of the leg. The legged robot according to any one of claims 10 to 12. Means for setting and storing a predetermined value capable of driving the actuator;
2. The apparatus according to claim 1, further comprising means for comparing the target torque with the predetermined value and changing the target torque to the predetermined value when the target torque exceeds the predetermined value. The legged robot according to any one of 2 and 4. 15. The legged robot according to claim 14, wherein a maximum torque value that can be driven by the actuator is set as the predetermined value. The target torque calculating means is
Means for calculating data indicating a left knee position and a right knee position from the data group;
A torso, a left knee position, a right knee position, a position near the left foot and a position near the right foot, a dynamic system in which mass is concentrated, the left foot position, the right foot position, and the trunk position; 5. The apparatus according to claim 1, further comprising means for calculating a target torque of a joint of the legged robot from data indicating the trunk position / posture, the left knee position, and the right knee position. The legged robot according to item 1. The dynamic system used in the target torque calculation means has mass concentrated on the trunk, left knee position, right knee position, left foot vicinity position, right foot vicinity position, trunk lower reference point, and trunk upper reference point. The legged robot according to claim 16, wherein the legged robot is a dynamic system. The dynamic system is such that the total mass of the left shin lower half mass, the left ankle mass, and the left foot mass is concentrated in the vicinity of the left foot, and the right shin lower half mass, the right ankle mass, and the right foot are positioned in the vicinity of the right foot. The total mass of the mass is concentrated, the mass of the upper left shin, the mass of the left knee joint, and the mass of the left knee joint, and the mass of the lower left thigh are concentrated, and the mass of the upper right shin is at the right knee position. And the total mass of the right knee joint mass and the right lower thigh mass is concentrated, and the upper left thigh mass, the left hip mass, the right upper thigh mass, the right hip joint mass and the lower trunk mass at the lower trunk reference point 18. The legged robot according to claim 17, wherein the half-mass is a dynamic system in which the total mass of the half mass is concentrated and the half mass on the trunk is concentrated on the upper trunk reference point. The leg type according to claim 18, wherein the height of the lower trunk reference point and the upper trunk reference point is set to a height that maintains the magnitude of the trunk inertia moment about the horizontal axis. robot. Using the "legal robot model of the legged robot" using the "mechanical system in which mass is concentrated at the torso, left knee position, right knee position, left foot position and right foot position" Calculating the torque;
Calculating a target rotation amount of each joint necessary to realize the position and orientation of the legged robot;
Adjusting the command torque applied to the motor of the actuator based on the calculated target rotation amount and the deviation of the detected actual rotation amount;
In the adjustment step, the calculated target torque is added to the command torque. Using the "legal robot model of the legged robot" using the "mechanical system in which mass is concentrated at the torso, left knee position, right knee position, left foot position and right foot position" Calculating the torque;
Calculating a target rotation amount of each joint necessary to realize the position and orientation of the legged robot;
Adjusting a command torque applied to the motor of the actuator based on the calculated target rotation amount and a deviation of the detected actual rotation amount;
Comparing the calculated target torque with the command torque, or comparing the calculated target torque with a detected actual torque;
A legged robot control method comprising a step of estimating a friction characteristic value of the actuator based on a comparison result. The method of controlling a legged robot according to claim 21, wherein in the comparison step, the comparison is performed using torque data at the time of a free leg. A legged robot control method in which the rigidity of a part of a legged robot is locally reduced, and the low rigidity part acts as a spring element,
Modeling the legged robot using the "mechanical system in which mass is concentrated at the torso, left knee position, right knee position, left foot position, and right foot position" Calculating a target torque;
Calculating a target rotation amount of each joint necessary to realize the position and orientation of the legged robot;
Adjusting a command torque applied to the motor of the actuator based on the calculated target rotation amount and a deviation of the detected actual rotation amount;
Based on the calculated deviation between the target torque and the detected actual torque, and the spring constant of the spring element, a correction amount necessary to correct the target rotation amount of the actuator that drives the low-rigidity portion is calculated. A calculating step;
In the adjusting step, the calculated correction amount is added to the target rotation amount. 24. The method of controlling a legged robot according to claim 23, wherein a sole portion of the legged robot acts as the spring element. The legged robot control method according to claim 23, wherein the actuator speed reducer acts as the spring element. The method for controlling a legged robot according to any one of claims 23 to 25, wherein the actuator that drives the low-rigidity portion is an actuator that drives an ankle joint of the legged robot. The apparatus further comprises means for comparing the command torque with the limit value for the command torque and outputting a drive signal corresponding to the limit value to the motor of the actuator when the command torque exceeds the limit value. 24. The method for controlling a legged robot according to any one of claims 20, 21, and 23. The allowable range in which ZMP can be allowed is calculated at the sole portion of the legged robot when standing, and a limit value for the command torque is set based on the calculated allowable range. 27. A method for controlling a legged robot according to 27. The target torque calculation step includes:
Calculating a target rotation amount of each joint necessary to realize the position and orientation of the legged robot;
Modeling the legged robot, using the "mechanical system in which mass is concentrated at the torso, left knee position, right knee position, left foot position, and right foot position" 24. The method according to claim 20, further comprising a step of calculating a target torque of a joint of the legged robot by calculating a position and an acceleration of each mass point from a temporal change of the target rotation amount. A control method for the legged robot according to the item. When the leg portion of the legged robot is a standing leg, the target torque of the target joint is calculated using another mass point excluding the mass point existing in the toe direction from the target joint of the standing leg. 30. A method of controlling a legged robot according to claim 29, wherein: When the leg portion of the legged robot is a free leg, a target torque of the target joint is calculated using a mass point existing in the toe direction from the target joint of the free leg. Item 29. A method for controlling a legged robot according to item 29 or 30. When the legged robot is in the aerial phase floating in the air, the target torque of the target joint is calculated using a mass point existing in the toe direction from the target joint of the leg. The method for controlling a legged robot according to any one of claims 29 to 31. The step of comparing the target torque with a predetermined value for the actuator and further changing the target torque to the predetermined value when the target torque exceeds the predetermined value. The method for controlling a legged robot according to any one of 20, 21, and 23. 34. The control method for a legged robot according to claim 33, wherein a maximum torque value that can be driven by the actuator is set as the predetermined value. The target torque calculation step includes:
Calculating data indicating left knee position and right knee position;
A torso, a left knee position, a right knee position, a position near the left foot and a position near the right foot, a dynamic system in which mass is concentrated, the left foot position / posture, the right foot position / posture, and the trunk 24. The step of calculating a target torque of a joint of the legged robot from data indicating the position, the trunk position and posture, the left knee position, and the right knee position. The method for controlling a legged robot according to any one of the preceding claims. The dynamic system used in the target torque calculation means has mass concentrated on the trunk, left knee position, right knee position, left foot vicinity position, right foot vicinity position, trunk lower reference point, and trunk upper reference point. 36. A control method for a legged robot according to claim 35, wherein the control system is a dynamic system. The dynamic system is such that the total mass of the left shin lower half mass, the left ankle mass, and the left foot mass is concentrated in the vicinity of the left foot, and the right shin lower half mass, the right ankle mass, and the right foot are positioned in the vicinity of the right foot. The total mass of the mass is concentrated, the mass of the upper left shin, the mass of the left knee joint, and the mass of the left knee joint, and the mass of the lower left thigh are concentrated, and the mass of the upper right shin is at the right knee position. And the total mass of the right knee joint mass and the right lower thigh mass is concentrated, and the upper left thigh mass, the left hip mass, the right upper thigh mass, the right hip joint mass and the lower trunk mass at the lower trunk reference point 37. The control method for a legged robot according to claim 36, wherein the half-mass total mass is concentrated and the trunk half-mass is concentrated at the upper trunk reference point. The leg type according to claim 37, wherein the height of the lower trunk reference point and the upper trunk reference point is set to a height that maintains the magnitude of the inertia moment of the trunk about the horizontal axis. Robot control method.

Images