JP4848686B2 - Legged robot and its motion adjustment method - Google Patents

Description

  The present invention relates to a technique for controlling the operation of a legged robot. In particular, the present invention relates to a technique for adjusting the operation of a legged robot following an external command or the like.

A legged robot that has a trunk and a leg link that is swingably coupled to the trunk, and that walks by changing the relative posture (sometimes referred to simply as posture) of the trunk and the leg link group (simply simply Have been developed).
In order for a legged robot to walk, data describing changes over time in the relative postures of the trunk and leg link groups is required. The relative posture of the trunk and the leg link group can be described by, for example, a change with time of the positions of the feet and the trunk. At this time, the temporal change in the trunk position needs to be an appropriate value with respect to the temporal change in each toe position, and if the value is not appropriate, the robot falls.
In order to obtain the time-dependent change in the trunk position where the legged robot does not fall down, a complicated calculation that takes into account the dynamics of the robot is required. Below, an example in the case of a bipedal walking robot is demonstrated.
(1) Designate temporal data indicating the positions of the left and right foot tips of the robot.
(2) Designate a position where the ZMP of the robot must be present in consideration of the ground contact position of the foot. ZMP (zero moment point) refers to a point on the floor where the moment of the resultant force of gravity, floor reaction force and inertial force acting on the robot becomes zero. If the ZMP is within the foot of the grounded leg, the robot will not tip over. Conversely, in order for the robot not to fall, the ZMP must be in the foot of the grounding leg. Therefore, a target ZMP that satisfies the following relationship is designated in consideration of the position and posture of the foot of the grounding leg. That is, while one leg link (for example, the left leg) is a free leg, it exists in the foot of the ground leg (right leg), and the one leg (left leg) is grounded and both legs are grounded. At the time of movement, it starts to move toward the foot of the newly grounded leg (left leg), and before the leg (right leg) that was grounded before becomes a free leg, Specify the ZMP to finish moving in the foot. The ZMP trajectory specified in this way is called a target ZMP trajectory. If the actual ZMP trajectory matches the target ZMP trajectory, the robot continues to walk without falling.
(3) When a change in the ground contact position of the toe and a target ZMP that changes following the change are specified, the dynamics of the robot is calculated on the assumption that the trunk position changes with time. Since the temporal change in the toe position is specified at the time of calculation, assuming the temporal change in the trunk position of the robot, the posture of the whole body of the robot and the change speed thereof are determined. When the posture of the whole body of the robot is determined, the position of the ZMP in that posture can be calculated. In order to calculate the position of the ZMP, in addition to static elements, the influence of inertial force acting on the robot must be taken into account. By including the temporal change of the assumed trunk position in the calculation, it becomes possible to calculate the position of the ZMP in consideration of the dynamics of the robot. Since it is possible to calculate the position of the ZMP assuming a temporal change in the trunk position, it is possible to search for a temporal change in the position and posture of the trunk that realizes the ZMP that matches the target ZMP.

The data indicating the temporal change of the trunk position searched for by the above is called trunk gait data, the data showing the temporal change of the originally specified foot position is called the foot gait data, and both Collectively called gait data. If the robot walks according to the gait data, the actual ZMP matches the target ZMP, and the robot can continue walking without falling.
Gait data is given as a change in position over time over time. Since position, velocity, and acceleration are related, and other quantities can be calculated from one of these quantities, changes over time in speed or acceleration may be handled instead of position. The positions of the toes and the trunk can be described by, for example, the position of the reference point defined on the trunk and the direction of the reference direction. The foot position is specified by the position of the reference point fixed to the foot and the direction of the reference line fixed to the foot, and the trunk position is the position and body of the reference point fixed to the trunk. It is specified by the direction of the reference line fixed to the trunk.

Research on robots that sequentially change walking motion following external commands is ongoing. For example, in Non-Patent Document 1, gait data for a predetermined period is created based on an external command, and newly created gait data is sequentially connected to previously created gait data, A technique for creating a series of gait data following external commands is described.
Koichi Nishiwaki, “Walking System Construction and Online Generation Control of Humanoid Robot”, Doctoral Dissertation, Graduate School of Engineering, The University of Tokyo, December 14, 2001

When changing from a planned walking motion to a new walking motion, it is necessary to ensure the continuity of the position and speed of the toes and the position and speed of the trunk at the time of the change. However, the position and speed of the toes at the start of the walking movement after the change and the position and speed of the trunk are forced to the position and speed of the toes and the position and speed of the trunk at the end of the walking movement before the change. Therefore, the ZMP may deviate significantly from the target ZMP immediately after the start of the walking operation after the change.
Therefore, in the technique of Non-Patent Document 1, the time point when the walking motion is changed is limited to the time zone in which the robot is stable in the state of both legs standing. Alternatively, the degree of changing the walking motion is limited.
The present invention provides a technique that allows a legged robot to change a walking motion more freely following a command given from the outside or the like.

The present invention can be embodied in a legged robot including a leg link that is swingably connected to the trunk. This legged robot has means for storing time-series data of trunk positions when carrying out a scheduled walking motion, means for determining a change time point from a planned walking motion based on the command, Based on the means for creating a target ZMP trajectory after the change time, means for assuming time-series data of the trunk position after the change time, and the trunk stored for two or more time points immediately before the change time Means for creating time series data by connecting time series data of the trunk position after the assumed change time to two or more data of positions, and calculating a temporary ZMP trajectory after the change time from the time series data Correction means for correcting the time-series data of the assumed trunk position so that the sum of squares of the ZMP deviation between the calculated temporary ZMP trajectory and the target ZMP trajectory is minimized; Trunk position And means for updating the time-series data of the torso position of the corrected time series data.

The legged robot determines a change time point from the scheduled walking motion based on the command and creates a target ZMP trajectory after the change time point. This target ZMP trajectory enables a walking motion corresponding to the command. The command here includes, for example, a command given from the outside of the robot, a command given from the inside of the robot, or the like. Examples of commands from outside the robot include commands that an operator or the like sequentially gives to the robot. The command from the inside of the robot includes, for example, that the robot itself detects the presence of an obstacle and the robot itself autonomously generates a corresponding operation change command. Alternatively, the robot itself may previously store a series of commands prepared in advance by an operator or the like, a series of commands created from information learned by the robot itself during a previous operation, and the like.
Assuming a temporal change in the trunk position after the change time, a temporary ZMP trajectory can be calculated. At this time, by using two or more data of the trunk position stored for two or more time points immediately before the change time point, the trunk position and the change speed immediately before the change time point are also taken into account. Then, a temporary ZMP trajectory is calculated. The calculation of ZMP requires the dynamic behavior of the robot to be taken into account, and the temporal ZMP trajectory after the change time is calculated by including the temporal change in the trunk position after the time immediately before the change time. Can do.
The assumed trunk position can be corrected based on the deviation between the temporary ZMP trajectory calculated from the temporal change in the assumed trunk position and the target ZMP trajectory. Usually, the work for correcting the trunk position is repeated until a change with time of the trunk position in which the ZMP deviation is zero is obtained. However, when changing the walking motion to an unexpected time, a dynamic motion for correcting the trunk position before the change from the trunk position before the change is added, so the ZMP deviation is set to zero from the time of the change to after the change. I can't do it. If the ZMP deviation immediately after the change is to be maintained at zero, the ZMP deviation at the time of the change becomes large.
Therefore, in the legged robot of the present invention, the assumed trunk position is corrected based on the sum of squares of the ZMP deviation over a predetermined period. That is, the ZMP deviation that should occur when the walking motion is changed is dispersed over time.
If the ZMP deviation is corrected to a trunk position that disperses with time by a verification experiment, even when the walking motion is changed relatively large, the actual ZMP does not deviate significantly from the target ZMP at the time of changing the walking motion. It has been confirmed. Thereby, for example, it is possible to set the change time of the walking motion more freely or to change the walking motion relatively large.
According to this legged robot, the walking motion can be changed more freely following the command.

Preferably, the correction means corrects the assumed sum of the trunk position with time so that the sum of squares of the deviation of the ZMP is minimized.
Thereby, the ZMP deviation that should occur can be more appropriately distributed.

The present invention can also be embodied in a method for adjusting the walking motion of a legged robot. This method includes a step of storing time series data of a trunk position when performing a scheduled walking motion in a legged robot, a step of determining a change time point from a planned walking motion based on a command, and a command The legged robot memorizes the step of creating the target ZMP trajectory after the change time based on the step, assuming the time-series data of the trunk position after the change time, and two or more time points immediately before the change time. Time series data is created by connecting time series data of the trunk position after the assumed change time to two or more data of the current trunk position, and the temporary ZMP trajectory after the change time is created from the time series data The legged robot stores the step of calculating the time series data of the assumed posture so that the sum of squares of the deviation between the calculated temporary ZMP trajectory and the target ZMP trajectory is minimized. Change The time series data of a point subsequent orientation, and a step of updating the time-series data of the torso position after correction.
According to this method, the walking motion of the legged robot can be changed more freely following the command.

  According to the present invention, the degree of freedom of a robot that continues walking by changing an unplanned walking motion from a planned walking motion is improved. It contributes to the realization of robots that can adapt to occasions.

First, the main features of the embodiments described below are listed.
(Embodiment 1) The robot includes an instruction device for an operator to input a walking motion change command.
(Embodiment 2) The robot creates walking pattern data describing the target grounding position, grounding direction, target grounding time, etc. of the target leg based on a command signal from the pointing device.
(Mode 3) The robot creates a target ZMP trajectory based on the walking pattern data.

A legged robot according to an embodiment of the present invention will be described with reference to the drawings.
FIG. 1 shows an outline of the robot 2 of this embodiment. The robot 2 includes a trunk 4, a left leg link 6, a right leg link 8, and a control unit 10. In addition, an instruction device 12 for providing an operation command to the robot 2 by an operator or the like is prepared.
In this embodiment, as shown in FIG. 1, an absolute coordinate system (x, y, z) fixed to the floor outside the robot 2 is defined. In the absolute coordinate system (x, y, z), the x-axis direction and the y-axis direction are in the horizontal direction, and the z-axis direction is in the vertical direction. Further, a relative coordinate system (x ′, y ′, z ′) fixed to the reference point of the support leg of the robot 2 is defined. The relative coordinate system (x ′, y ′, z ′) is fixed to the reference point R0 when the robot 2 uses the right leg link 8 as a supporting leg, and the robot 2 uses the left leg link 6 as a supporting leg. When it is fixed to the reference point L0. In the relative coordinate system (x ′, y ′, z ′), the direction extending from the heel to the toes along the foot link 48 is the x ′ axis, and is orthogonal to the x ′ axis along the foot links 38, 48. The direction to be taken is the y ′ axis, and the direction perpendicular to the foot links 38 and 48 is the z ′ axis. Normally, the x ′ axis corresponds to the direction in which the robot 2 walks, the y ′ axis corresponds to the body side direction of the robot 2, and the z ′ axis corresponds to the height direction of the robot 2.

The left leg link 6 is swingably connected to the trunk 4 at one end via a hip joint. The left leg link 6 has a knee joint and an ankle joint, and has a left foot 7 at the tip. The right leg link 8 is swingably connected to the trunk 4 via the hip joint. The right leg link 8 has a knee joint and an ankle joint, and has a right foot 9 at the tip. Each joint of the robot 2 includes actuators (not shown), and these actuators are rotationally driven by instructions from the control unit 10. A reference point L0 is provided at the approximate center of the left foot 7 of the left leg link 6. A reference point R0 is provided at the approximate center of the right foot 9 of the right leg link 8. The reference points L0 and R0 are points that serve as a reference when creating the motion pattern of the robot 2. In the figure, G is a reference point of the trunk 4 of the robot 2 and substantially coincides with the position of the center of gravity of the robot 2 when the walking motion is executed. In the robot 2, the reference point G of the trunk 4 can be handled as the center of gravity of the robot 2.
The control unit 10 is a computer device having a CPU, ROM, RAM, hard disk and the like. The control unit 10 can communicate with the instruction device 12, and a command input by the operator to the instruction device 12 is taught to the control unit 10 of the robot 2. The control unit 10 controls the operation of the robot 2 based on the command signal input from the pointing device 12. Details of the control unit 10 will be described in detail later.

  The operation performed by the robot 2 will be described with reference to FIGS. The robot 2 performs a walking motion by alternately swinging the left and right leg links 6 and 8 in accordance with a command signal from the pointing device 12. At this time, the leg link on which the foot is grounded is called a support leg, and the leg link on which the foot moves in the air is called a free leg. The robot 2 determines a target position and direction for grounding the foot of the free leg based on a command signal input from the pointing device 12, swings the free leg to the ground target position, and performs a walking motion. The direction in which the foot is grounded refers to the direction extending from the foot heel to the toe. The ground contact direction of the foot is specified by an angle formed by the x ′ axis and the x axis.

FIG. 2 illustrates a target ground contact position determined by the robot 2 based on a command signal from the pointing device 12. As the pointing device 12, for example, a joystick or the like can be used. Initially, it is assumed that the right foot 9 is grounded at the position P1, the left foot 7 is grounded at the position P2, and the robot 2 is stationary. When a command signal “move forward” is input from the pointing device 12, for example, the right leg link 8 is determined as a free leg, and the target grounding position P3, the grounding direction D3, and the grounding time of the right foot 9 are determined. When the target ground contact position P3 and the direction D3 of the right foot 9 are determined, the target of the left foot 7 of the left leg link 6 to be the next free leg along the target ground position P3 and the direction D3 of the right foot 9 is determined. A grounding position P4, a grounding direction, and a grounding time are determined. That is, the robot 2 plans and executes a walking motion that moves by one step in response to a single command signal, and stands still with the feet 7 and 9 aligned. When a command signal is input again while executing this walking motion, the target grounding position P6 and the direction D6 of the left foot 7 are newly determined, and the target grounding position P5 of the right foot 9 is determined. The robot 2 creates a walking plan for two steps based on an instruction from the operator, and connects to the running walking plan to create a series of walking plans.
The target grounding position, grounding direction, and grounding time are based on an instruction input to the pointing device 12 by the operator. The robot 2 performs a series of walking actions while changing the walking direction, the walking speed, the stride, etc. following the instructions of the operator. As will be described later, when the operator continuously instructs, the robot 2 makes a walking plan in which the right foot 9 is grounded to the target grounding position P3 in the direction D3 and the left foot 7 is grounded to the target grounding position P6 in the direction D6. Thus, the walking motion is not performed in which the feet 7 and 9 are aligned and stationary.
Data that sequentially describes the ground contact position, the ground contact direction, the ground contact time, etc. that are the target of the free leg may be referred to as walking pattern data.

Next, the control unit 10 of the robot 2 will be described. As shown in FIG. 3, the control unit 10 functionally includes a gait data creation device 54, a gait data correction device 56, a joint angle group calculation device 58, an actuator control unit 60, and an actual motion calculation device. 62 etc. are provided.
The entire control unit 10 may be physically included in one device, or may be accommodated separately for each physically separated device. Each element of the control unit 10 does not necessarily have to be mounted on the robot 2. It may be arranged outside the robot 2 and instruct the robot 2 wirelessly or by wire.

The gait data creation device 54 creates and stores gait data including foot gait data and trunk gait data based on a command signal input from the instruction device 12. The toe gait data is data describing the trajectories of the reference points L0 and R0 (toe) of the feet 7 and 9 of the left and right leg links 6 and 8 of the robot 2. The trunk gait data is data describing the trajectory of the reference point G of the trunk 4. The trajectory means a change in position over time. The trajectory of each reference point L0, R0, G may be described by absolute coordinates (x, y, z) and absolute time, or relative coordinates (x ′, y ′, z ′) for each step or multiple steps. ) And relative time. The gait data creation device 54 will be described in detail later. The gait data stored in the gait data creation device 54 is input to the gait data correction device 56.
The gait data correction device 56 corrects the gait data input from the gait data creation device 54 based on the actual motion of the robot 2. The gait data corrected as necessary is input to the joint angle group calculation device 58.
The joint angle group calculation device 58 calculates each joint angle of the robot 2 by solving so-called inverse kinematics based on the input gait data. The calculated joint angle group data is input to the actuator control unit 60.
The actuator control device 60 controls the actuator group mounted on the robot 2 based on the input joint angle group data. When the actuator control device 60 controls the actuator group, the mechanical system 64 of the robot 2 performs a walking motion.

The walking motion actually performed by the robot 2 is affected by disturbance forces due to unexpected road surface unevenness and structural deflection of the robot 2. The walking motion actually performed by the robot 2 is detected by a sensor group 66 mounted on the robot 2.
The actual motion calculation device 62 calculates the actual motion of the robot 2 based on the output signal of the sensor group 66. The actual motion calculation device 62 can calculate the actual motion of the left leg link 6, the actual motion of the right leg link 8, and the actual motion of the trunk 4, for example. The actual motion of the robot 2 calculated by the actual motion calculation device 62 is input to the gait data correction device 56.
As described above, the gait data correction device 56 corrects the gait data input from the gait data creation device 54 based on the actual motion of the robot 2 calculated by the actual motion calculation device 62. The walking motion of the robot 2 is feedback controlled based on the actual motion of the robot 2.

Next, the gait data creation device 54 will be described in detail. As shown in FIG. 4, the gait data creation device 54 functionally includes an update time determination device 102, a walking pattern creation device 104, a target ZMP trajectory creation device 106, a foot start activation creation device 108, and a body. A stem activation creation device 110 and a data storage device 112 are provided. The data storage device 112 can store walking pattern data 122, target ZMP trajectory data 124, foot gait data 126, trunk gait data 128, and the like.
The update time determination device 102 determines a time point at which the walking motion of the robot 2 is updated when a command signal is input from the pointing device 12. The update time determination device 102 determines the update time of the walking motion based on the calculation time required by the gait data creation device 54.
The walking pattern creation device 104 updates the walking pattern data 122 stored in the data storage device 112 based on the update time determined by the update time determination device 102 and the command signal from the instruction device 12.
The target ZMP trajectory creation device 106 creates a target ZMP (Zero Moment Point) trajectory based on the walking pattern data 112 stored in the data storage device 112. ZMP means a point on the floor where the sum of moments due to external forces (including inertial force) acting on the robot 2 becomes zero. If the ZMP is maintained within the polygonal range formed by the feet 7 and 9 that are in contact with the ground, the robot 2 can continue to walk without falling. The target ZMP trajectory ZMP (t) for this purpose can be determined based on the temporal change (walking pattern) of the ground contact position of the feet 7 and 9 of the leg links 6 and 8. The target ZMP trajectory may be described in absolute coordinates (x, y, z) and absolute time, or in relative coordinates (x ′, y ′, z ′) and relative time for each step or steps. You can also. The created target ZMP trajectory is stored in the data storage device 112 as discretized target ZMP trajectory data 124. Further, the target ZMP trajectory creation device 106 updates the target ZMP trajectory data 124 stored in the data storage device 112 when the walking pattern data 122 is updated.

Based on the walking pattern data 122 stored in the data storage device 112, the foot tip trajectory creation device 108 uses the trajectories (foot tip trajectories) of the reference points L 0 and R 0 of the feet 7 and 9 of the left and right leg links 6 and 8. ). The created toe trajectory is stored in the data storage device 112 as the toe gait data 126. In addition, when the walking pattern data 122 is updated, the toe trajectory creation device 108 updates the toe gait data 126 stored in the data storage device 112.
The trunk trajectory creation device 110 creates a trajectory of the reference point G of the trunk 4 based on the target ZMP trajectory data 124 and the toe gait data 126 stored in the data storage device 112. The created trunk trajectory is stored in the data storage device 112 as discrete trunk gait data 128. Further, the trunk trajectory creation device 110 updates the trunk gait data 128 stored in the data storage device 112 when the target ZMP trajectory data 124 is updated.

The robot 2 executes a walking motion by moving the toes and the trunk according to the toes' gait data 126 and the torso gait data 128 stored in the data storage device 112 of the gait data creation device 54. To do. When the operator inputs a command from the pointing device 12, the gait data creation device 54 creates the footstep data created based on the command of the operator in the currently executed foot gait data 126 and trunk gait data 128. The gait data 126 and the trunk gait data 128 are connected to create a series of gait data.
FIG. 5 shows the flow of processing executed by the gait data creation device 54. A processing procedure in which the gait data creation device 54 creates a series of gait data following the operation command will be described along the flow shown in FIG.

In step S2, it will be in a standby state until the command signal from the instruction device 12 is input. During this time, the robot 2 executes the gait data 126 and 128 stored in the data storage device 112 of the gait data creation device 54. When a command from the instruction device 12 is not input, trajectory data in which the robot 2 is kept stationary may be created and stored every predetermined time. When the command signal from the instruction device 12 is input, the process proceeds to step S4.
In step S4, the update time determination device 102 determines the update (change) time of the walking motion. As shown in FIG. 6, the update time point is determined at a time point (for example, k) delayed from the time point when the operation command signal is input by a margin time considering the time required for processing described later. K in FIG. 6 indicates the order of data in the time-series data. For example, Q _k means the k-th data in the data being executed. When the update time point k is determined, newly created update data (R ₂ , R ₃ ...) Is connected to the time-series data (Q ₀ , Q ₁ ... Q _k−1 , Q _k ) being executed. The data position (Q _{k-1 in} FIG. 6) is determined. The data after the data Q _k in the time series data in execution is discarded.
In step S6 of FIG. 5, the walking pattern creation device 104 updates the walking pattern data 122 based on the command signal from the instruction device 12 and the determined update time. For example, as shown in FIG. 2, the contact position P4 and the direction (equal to D3) of the left foot 7 of the left leg link 6 are updated to the contact position P6 and the direction D6.

In step S <b> 8 of FIG. 5, the toe trajectory creation device 108 updates the toe gait data 126 stored in the data storage device 112 based on the updated walking pattern data 122. If the time-series data shown in FIG. 6 is the toe gait data 126, the updated toe step newly created in the data Q _k-1 at the time point k-1 of the toe gait data 126 being executed. A series of foot gait data is created by connecting the condition data (R ₂ , R ₃ ,...). At this time, when the first data R ₂ of the updated gait data is determined, a plurality of data (for example, Q _k−2 and Q _k−1 ) retroactive from the update time of the gait data being executed are taken into consideration. By doing so, a series of toe gait data describing a continuous toe trajectory can be created.
In step S <b> 10, the target ZMP trajectory creation device 106 updates the target ZMP trajectory data 124 stored in the data storage device 112 based on the updated walking pattern data 122. For example, as shown in FIG. 2, when the ground contact position P4 and direction (equal to D3) of the left foot 7 of the left leg link 6 are updated to the ground contact position P6 and the direction D6, point PA-point P2-point P3- The target ZMP trajectory connecting point PB is updated to a trajectory connecting point PA-point P2-point P3-point P6-point PC.

The trunk trajectory creation device 110 updates the trunk gait data 128 stored in the data storage device 112 by the processing in steps S12 to S18 in FIG. The trunk trajectory creation device 110 updates the trunk gait data 128 so that the robot 2 can implement the walking patterns sequentially updated by the operator as a series of walking motions.
In step S12, a trunk trajectory after the update time k is assumed, and a ZMP deviation between the temporary ZMP trajectory calculated from the assumed trunk gait data and the ZMP trajectory described by the target ZMP trajectory data 124 is calculated. If the time series data shown in FIG. 6 is a discretized trunk trajectory (ie trunk gait data), trunk gait data connected after the update time point k is (R ₂ , R ₃ ,. R−2 time series data consisting of R _n−1 ). The assumed trunk trajectory may be, for example, a trajectory obtained by moving the target ZMP trajectory by a predetermined height in the z-axis direction.
In general, the ZMP trajectory (Px (t), Py (t)) for the trunk (centroid) trajectory (x (t), y (t), z (t)) can be calculated from the following ZMP equation. .

Here, m is the mass of the robot 2. fz is a floor reaction force acting on the foot of the support leg. I _x · dω _x / dt and I _y · dω _y / dt indicate the influence of inertia on the rotation of the reference point G of the trunk 4. In this embodiment, since the angular momentum of the robot 2 around the horizontal axis is set to 0 (zero), I _x · dω _x / dt and I _y · dω _y / dt become 0 (zero).
By discretizing the above ZMP equation, for example, the following ternary equation can be obtained in the x-axis direction.

Here, x _i and P _xi are discretizations of the trunk trajectory x (t) and the ZMP trajectory px (t) in the x direction, respectively. i indicates the order of the time-series data divided by the unit time Δt. Equation 2 above, ZMP in the x-direction coordinate _{P xi} at time i is the x-direction coordinate _{x i-1} immediately preceding, the x-direction coordinate _{x i} at that time, that can be calculated from the x-direction coordinate _{x i + 1} immediately after the Show. The coefficients a _i , b _i and c _i are coefficients calculated as follows. For the floor reaction force fz, the equation of motion of the robot 2 in the z-axis direction: m · d ² z / dt ² = −mg + fz holds.

As can be understood from Equation 2 above and FIG. 6, the updated time points of the assumed trunk gait data (R ₂ , R ₃ ,..., R _n-1 ) and the trunk gait data 128 being executed are updated. By using two pieces of data (Q _k−2 , Q _k−1 ) retroactive from k, the ZMP deviation d _i is
d ₁ = ZMP ₁ −a ₁ · Q _k−2 + b ₁ · Q _k−1 + c ₁ · R ₂
From
d _n−2 = ZMP _n−2 −a _n−2 · R _n−3 + b _n−2 · R _n−2 + c _n−2 · R _n−1
Up to n−2 time series data can be obtained. Here, ZMP _i is a discretized target ZMP trajectory described by the target ZMP trajectory data 124. The ZMP deviation d ₁ is a ZMP deviation with respect to the time immediately before the update time k. By using two pieces of data (Q _k−2 , Q _k−1 ) retroactive from the update time point k, it is possible to take into account the trunk position and the speed of change at the time point immediately before the update time point k as constraints.

In step S14 of FIG. 5, the trunk gait data (R ₂ ) assumed in step S12 based on the time-series data (d ₁ , d ₂ ,..., D _n−2 ) of the ZMP deviation calculated in step S12. , R ₃ ,..., R _n-1 ) are determined (r ₂ , r ₃ ,..., R _n-1 ). However, since it is necessary to stabilize the robot 2 at the final time point n−1 (with the position and speed being continuous), the data R _{n−2 of the} final two time points n−2 and n−1. , R _n-1 is not corrected, and the correction amounts r _n-2 , r _n-1 are set to zero.
From the ZMP equation shown in Equation 1, in general, the ZMP deviation d (t): (d _x (t), _dy (t)) and the trunk (center of gravity) trajectory required to reduce the ZMP deviation to zero. The relationship of the horizontal correction amount r (t) :( r _x (t), r _y (t)) can be expressed by the following equation.

The ZMP equation relating to the correction amount r (t) can also be rewritten into a ternary equation by discretization. The trunk trajectory creation device 110 substitutes the ZMP deviation (d ₁ , d ₂ ,..., D _n−2 ) calculated in step S12 for the ternary equation obtained by discretizing the mathematical formula 4, and the correction amount (r ₁ , r ₂ ,..., Rn _-3 ), the following simultaneous equations are derived. As described above, the final two correction amounts r _n−2 and r _n−1 are already set to zero. In the following formula, the x and y axis directions are not distinguished, but in actuality, the x and y axis directions are derived.

If the correction amount (r ₁ , r ₂ ,..., R _n-3 ) that satisfies the above mathematical formula 5 can be derived, the actual ZMP of the robot 2 can be the target even at the time of change of operation (update time point k). Trunk gait data consistent with ZMP can be obtained. However, in Equation 5 above, n−2 conditional expressions are given for the correction amount (r ₁ , r ₂ ,..., R _n−3 ) where the number of data is n−4. It is an indefinite simultaneous equation. This is because in this embodiment, the ZMP deviation at the time point k−1 immediately before the update time is taken into consideration.
As is well known, when a trunk (center of gravity) trajectory that follows a target ZMP trajectory is created, if the position and speed of the trajectory end point are constrained, the ZMP of the trajectory end point cannot be compensated. In the conventional technique, when a newly created center of gravity trajectory is connected to a trunk trajectory prepared in advance, ZMP is not compensated at the time of connection (update). In other words, the ZMP deviation d ₁ is ignored (set to zero) when solving the above Equation 5. In this method, when the operation is greatly changed, the ZMP deviation at the time of the change exceeds the allowable range, and the robot 2 may fall down in some cases. For this reason, in the prior art, the time point at which the motion is changed is limited to the time when both legs are standing, or the degree of change in the motion is limited.

In the robot 2 of the present embodiment, based on the idea that the ZMP deviation that maximizes before and after the motion change is dispersed during the motion after the time of the change, the indefinite simultaneous equations shown in the above formula 5 are expressed by the least square method. Use to solve the problem. As a result, a significant change in the operation that is impossible with the conventional technique can be suppressed to an acceptable level without increasing the ZMP deviation.
The trunk trajectory creation device 110 derives the following formulas 6 and 7 from the above formula 5 using the least square method.

In Equation 6 above, the sign of T attached to the right shoulder of the matrix indicates that it is a transposed matrix. By solving Equation 6, a correction amount (r ₁ , r ₂ ,..., Rn _-3 ) that minimizes the sum of squares of the ZMP deviation (d ₁ , d ₂ ,..., Dn _-2 ) is calculated. can do. Note that Equation 6 can be uniquely solved because the number of data of the correction amount (r ₁ , r ₂ ,..., Rn _-3 ), which is a variable, matches the number of conditional expressions.

In step S16, torso trajectory assumed in step _{S12 (R 1, R 2,} ··, R n-3) a correction amount calculated in step _{S14 (r 1, r 2,} ··, r n-3 ) Only by correcting, the updated trunk trajectory is created.
In step S18, the trunk gait data 128 stored in the data storage unit 112 is connected to the updated trunk trajectory (trunk gait data) created in step S16, thereby creating a new series of body gait. Update to trunk gait data 128.
The robot 2 repeatedly executes the process between steps S2 to S18 described above until an operation end command is input. The gait data 126 and 128 stored in the data storage device 112 are sequentially updated and connected based on an external command while being executed by the robot 2.

  As shown in FIG. 7, in the robot 2 of the present embodiment, the actual ZMP trajectory B follows the target ZMP trajectory A while maintaining an allowable level deviation after the operation update time. Unlike the conventional robot (A in the figure), the ZMP deviation does not increase abruptly when the operation is changed (data update). The operation can be changed more freely following the instructions of the operator. At this time, the continuity of the posture of the robot 2, its changing speed, and its changing acceleration is ensured.

Specific examples of the present invention have been described in detail above, but these are merely examples and do not limit the scope of the claims. The technology described in the claims includes various modifications and changes of the specific examples illustrated above.
In the above description, the reference point G of the trunk 4 is treated as the center of gravity of the robot 2, but if the reference point G cannot be approximated to the center of gravity, the trajectory of the center of gravity is obtained by the above-described method. The trajectories of the reference point G of the trunk 4 and the reference points L0 and R0 of the left and right foot 7 and 9 may be obtained.
Further, the instruction device that inputs an operation change command to the robot may be changed to, for example, a sensor that detects an obstacle or the like present on the walking road surface. Accordingly, the robot itself can generate an operation change command corresponding to the walking environment. That is, an operation change command may be given from the inside of the robot. A robot that can walk in a flexible manner can be realized in response to various environments.
In the robot of this embodiment, the walking pattern data is sequentially generated based on an external command (or an internal command). Therefore, it can be said that the walking pattern data describes a series of commands given from the outside. Also, preparing walking pattern data in advance and teaching the robot corresponds to teaching a series of commands to the robot in advance. In any case, the robot according to the present embodiment can sequentially generate gait data for a predetermined number of steps based on the walking pattern data and generate a series of gait data.

  The technical elements described in this specification or the drawings exhibit technical usefulness alone or in various combinations, and are not limited to the combinations described in the claims at the time of filing. In addition, the technology illustrated in the present specification or the drawings achieves a plurality of objects at the same time, and has technical utility by achieving one of the objects.

FIG. 3 is a diagram schematically illustrating the robot according to the first embodiment. The figure explaining the walking pattern of a robot. The functional structure of the computer apparatus with which the robot is mounted is shown. The figure which shows the functional structure of a gait data creation apparatus. The flowchart which shows the flow of the process which a gait data creation apparatus performs. The figure which shows typically the temporal relationship between the data in execution and the data after an update. The figure which compares and shows a target ZMP track and an actual ZMP track.

Explanation of symbols

2: Robot 4: Trunk 6: Left leg link 7: Left foot 8: Right leg link 9: Right foot 10: Control unit 12: Instruction device 54: Gait data creation device 56: Gait data correction device 58: Joint Angle group calculation device 60: Actuator control device 62: Actual motion calculation device 64: Robot mechanical system 66: Various sensors 102: Update point determination device 104: Walking pattern creation device 106: Target ZMP trajectory creation device 108: Toe trajectory creation Device 110: Trunk trajectory creation device 112: Data storage device 122: Walking pattern data 124: Target ZMP trajectory data 126: Toe gait data 128: Trunk gait data

Claims (2)

A legged robot having a leg link that is swingably connected to the trunk,
Means for storing time-series data of trunk positions when carrying out scheduled walking movements;
Means for determining a change point from the scheduled walking motion based on the command;
Means for creating a target ZMP trajectory after the change point based on the command;
Means for assuming time-series data of the trunk position after the time of change;
Create time-series data by connecting the time-series data of the trunk positions after the assumed modification time to two or more data of the trunk positions stored for the two or more time points immediately before the modification time. , A means for calculating a temporary ZMP trajectory after the change time from the time series data ;
Correction means for correcting time series data of the assumed trunk position so that the sum of squares of the deviation between the calculated temporary ZMP trajectory and the target ZMP trajectory is minimized;
Means for updating the time-series data of the trunk position after the stored change time to the time-series data of the trunk position after correction;
Legged robot with A method for adjusting the walking motion of a legged robot,
Storing the time-series data of the trunk position in the legged robot when performing the scheduled walking motion;
A step of determining a point of change from the scheduled walking motion based on the command;
Creating a target ZMP trajectory after the change time based on the command;
Assuming time series data of the trunk position after the change time,
Time series by connecting time series data of trunk positions after the assumed change time to two or more data of the trunk positions stored by the legged robot for two or more time points immediately before the change time A process of creating data and calculating a temporary ZMP trajectory after the change time from the time series data ;
Correcting the time-series data of the assumed trunk position so that the sum of squares of the deviation between the calculated temporary ZMP trajectory and the target ZMP trajectory is minimized;
Updating the time series data of the trunk position after the change time stored in the legged robot to the time series data of the trunk position after correction;
For adjusting the movement of a legged robot.

Images