JP2008049458A - Leg type robot - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a leg type robot capable of smoothly walking on a walking surface where its irregularity is unknown. <P>SOLUTION: Data of a target position attitude angle for a foot sole part of a leg section 11 to an imaginary walking surface is included in walking data. A target distance calculating section 106 calculates a target distance between the foot sole part and imaginary walking surface from the walking data. A distance sensor 50 measures a distance between the real foot sole part and real walking surface. A distance deviation calculating means 106 calculates a deviation between the target distance and measured distance. A compensated amount calculating means 110 calculates an amount of compensation for compensating the target position attitude angle of the foot sole part in a direction where its deviation is lessened. A motor driver 116 drives a joint of the leg section so that the foot sole part can follow the target position attitude angle which adds the amount of compensation. Motions of the foot sole part to the real walking surface can be brought near the motions of the foot sole part to the walking surface in the walking figure data by reducing the distance deviation. Then, walking motions close to smooth walking motions in the walking data can be realized. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a legged robot that has a leg portion connected to a body portion and walks on a walking surface by driving a joint of the leg portion. In particular, the present invention relates to a legged robot that can smoothly walk on a walking surface whose steps and unevenness are unknown.

The legged robot has a trunk and a leg connected to the trunk. The leg part has many links, and adjacent links are connected via joints. The legged robot walks by driving the joints of the legs. While walking, each leg repeats leaving, swinging forward and landing. Legs in the period from getting out of bed to landing are called swing legs, and legs in the period of landing on the walking surface are called standing legs. Further, a link at the tip of the leg that comes into contact with the walking surface is referred to as a foot.
It is more convenient to express the movement of the leg in the walk of "getting out-swinging forward-landing" by the change in the position of the foot and the posture angle over time rather than the change in the joint angle of each joint over time. It is. Therefore, in a legged robot, gait data describing changes over time in the target position and target posture angle of the foot are often created in advance. During walking, each joint is driven so that the position and posture angle of the foot portion at each time follow the target position and target posture angle of the gait data. Hereinafter, the target position and the target posture angle are collectively referred to as a target position and posture angle.
Gait data is created in advance on a computer (in other words, in a simulation model) while simulating a situation in which a virtual legged robot walks on a virtual walking surface. However, in many cases, the actual walking surface does not match the virtual walking surface on the computer, such as steps and unevenness. That is, the actual state of the walking surface is often unknown. Therefore, even if the joint is driven according to gait data created on the assumption of a virtual walking surface, there is a possibility that smooth walking is not realized.
For example, Patent Document 1 discloses a technique for correcting gait data in real time according to a step or unevenness of a walking surface while a legged robot is walking. The technique of Patent Document 1 is a technique for measuring a floor reaction force when a foot portion contacts a walking surface and correcting gait data based on the measured reaction force.

JP-A-5-305585

The technique of patent document 1 measures the floor reaction force when the foot part of the leg is landed. Therefore, in the technique of Patent Document 1, gait data is corrected after the foot has landed. Gait data is not corrected before the foot is landed. That is, in the technique of Patent Document 1, the movement of the foot portion is controlled based on the gait data set in advance until landing. For this reason, when the actual walking surface is different from the virtual walking surface, the foot will land at a timing different from the timing of landing on the gait data. When landing early, a large disturbance force is received from the walking surface, and when landing slowly, the leg is swung, and in either case, the legged robot tends to fall. Then, the walking of the actual legged robot becomes awkward.
In order to achieve smooth walking, before the foot link lands on the walking surface, the target position / posture angle of the foot part in the gait data is set according to the level difference or unevenness of the actual walking surface. It is better to be able to correct it.
Leg type that walks smoothly by correcting the target position and posture angle of the foot part in the gait data in real time according to the steps such as steps and unevenness of the walking surface before the foot part landing on the walking surface A robot is desired.

A legged robot according to the present invention is a legged robot having a leg portion connected to a body portion and walking on a walking surface by driving a joint of the leg portion. This legged robot has a storage device for storing gait data describing a temporal change in a target position / posture angle of a foot portion of a leg tip with respect to a virtual walking plane in a simulation model, and a gait data based on a gait data. A target distance calculation unit for calculating a target distance between at least three points that are set on the flat part and are not aligned on a straight line and a virtual walking surface; and a real foot part during walking A distance sensor for measuring a distance between each of the at least three points and the actual walking surface; a distance deviation calculating means for obtaining a target distance of each point and a deviation of the measured distance corresponding to each point; A correction amount calculating means for obtaining a correction amount for correcting the target position / posture angle of the foot in a direction to reduce the deviation, an adding means for adding the obtained correction amount to the target position / posture angle of the foot, Is the target position and posture angle with the correction amount added A drive control section for driving the joints of the leg portion so as to follow.
The drive unit is, for example, a motor and a speed reducer. The joint may be a rotary joint or a linear motion joint.
The target position / posture angle of the foot included in the gait data may be described directly or indirectly about the relationship of the target position / posture angle to the virtual walking surface. . In the case of direct description, for example, an absolute coordinate system is defined in a virtual space, a virtual walking plane is defined for the absolute coordinate system, and a foot portion is defined for the absolute coordinate system. This is a case where the target position / posture angle is set. Typically, for example, if the XY plane in the absolute coordinate system is defined as a virtual walking plane, the target position / posture angle of the foot part described in the absolute coordinate system is directly the virtual walking plane. It represents the relative position and relative attitude angle with respect to. In the case of indirectly describing, for example, a virtual walking plane is defined with respect to the absolute coordinate system, and the target position / posture angle of the foot portion of the stance leg is set in the absolute coordinate system. The target position / posture angle of the torso is set for the coordinate system fixed to the foot part of the foot, and the target position / posture angle of the foot part is set for the coordinate system fixed to the torso part. This is the case.

  According to the above legged robot, the gait data describes the relative positional relationship between the target position / posture angle of the foot and the virtual walking surface. From the gait data (that is, in the simulation model), a target distance between the virtual foot and the virtual walking surface is obtained. The calculated distance is a distance between at least three points that are set on the foot portion and are not arranged in a straight line, and the virtual walking surface. On the other hand, the distance between each of the at least three points and the actual walking surface is determined by the distance sensor while the legged robot is walking and before the foot of the free leg reaches the walking surface. It is measured. Based on the deviation between the target distance and the measured distance, a correction amount of the target position / posture angle of the foot is obtained. Since the relative position and posture angle of the foot with respect to the virtual walking surface can be obtained from the distance between at least three points that are not aligned on the straight line and the walking surface, The correction amount of the target position and orientation can be obtained. This correction amount corrects the target position / posture angle of the foot so that the distance between the actual walking surface measured by the distance sensor and the foot approaches the target distance obtained from the gait data. It will be a thing. That is, even if the actual walking surface does not coincide with the virtual walking surface when the gait data is created (in other words, even if the state of steps or unevenness is unknown), the foot portion is Control is performed so as to follow the change over time of the target distance between the virtual walking surface and the foot. The fact that the actual foot part is controlled to follow the temporal change in the target distance between the virtual walking surface and the foot part means that the foot part moves with respect to the virtual walking surface in the simulation. This means that a motion close to is obtained when walking with an actual legged robot. The landing timing of the foot part and the posture angle at the time of landing with respect to the actual walking surface are close to the landing timing on the simulation and the posture angle at the time of landing with respect to the virtual gait surface. As a result, it is possible to realize walking that is close to smooth walking simulated on a computer when creating gait data. That is, in a legged robot that walks on a real walking surface with an unknown unevenness or the like, it is possible to realize walking close to a smooth walking simulated on a computer based on a virtual walking surface.

In particular, it is known that a so-called bipedal legged robot having two legs has severe dynamic restrictions required for the legged robot so as not to fall. For example, in the case of dynamic walking, the constraint condition is that the ZMP exists in a convex hull (generally called a support polygon) in the contact area between the foot of the stance and the walking surface.
The target position / posture angle of the foot in the gait data is set so as to satisfy such a constraint condition. Here, there is a possibility that the target position / posture angle of the foot portion to which the correction amount is added as described above may not satisfy the constraint condition. The inventors of the present invention have hypothesized that the relative positional relationship between the foot portion and the torso portion in the absolute coordinate system (also referred to as the global coordinate system or the world coordinate system) during actual walking is simulated when creating gait data. If you follow the relative positional relationship between the foot and torso in the absolute coordinate system of a simple legged robot, the position and posture angle of the foot during actual walking can be calculated from the target position and posture angle on the gait data. It has been found that the above constraint condition is easily satisfied even if the deviation occurs. Therefore, the legged robot according to the present invention preferably has the following configuration. That is, the legged robot includes posture angle detection means for detecting the posture angle of the body part. The gait data includes data describing the change over time of the target posture angle of the trunk. Based on the posture angle detected by the posture angle detecting means, the drive control unit is configured such that the relative positional relationship between the foot portion and the torso portion in the absolute coordinate system is the foot portion and the torso in the absolute coordinate system of the virtual legged robot. The joint is driven to follow the relative positional relationship with the part. With such a configuration, it is possible to realize a legged robot that is difficult to fall over and can smoothly walk on a walking surface whose steps, unevenness, and the like are unknown.

Here, attention is focused on the target posture angle among the target position and posture angles of the foot. The target posture angle to which the correction amount is added needs to be set so that the foot can stably land. The stable landing of the foot means that the lower surface of the foot and the walking surface come into contact with each other at three or more points (three points that are not aligned). When landing with the free leg in an unstable state, that is, when the lower surface of the foot and the walking surface are in a one-point contact or two-point contact state, the other legged robot as a whole when the other leg becomes a free leg This is because there is a risk of wobbling. In particular, in a biped robot, the wobbling of the entire legged robot is likely to lead to a fall. In order for the target posture angle to which the correction amount is added to be set so that the foot can stably land, it is necessary to consider the unevenness of the walking surface at the planned landing position. This is because it is necessary to set a target posture angle of the foot portion for the lower surface of the foot portion to contact the walking surface at three or more points in consideration of the unevenness of the walking surface. For this purpose, the above-described distance sensor requires a resolution comparable to the pitch of the unevenness of the walking surface. When the uneven pitch of the walking surface is small, the distance sensor that measures the distance between the walking surface and the foot part is a distance sensor that measures the distance between two points arranged in a grid, or What uses image processing is required. A distance sensor arranged in a grid or a distance sensor using image processing is expensive.
Therefore, the legged robot according to the present invention preferably includes a sole portion that faces the lower surface of the foot portion and is elastically supported by the foot portion. And the target distance calculation part mentioned above is a point set to the foot part, and is between each of at least three points which do not line up on a straight line, and the virtual sole part which touches a virtual walking surface. Find the distance as the target distance. At the same time, the distance sensor described above measures the distance between at least three points that are set on the foot and are not arranged in a straight line, and the sole that is in contact with the actual walking surface. That is, the distance sensor measures the distance between the sole and the foot contacting the walking surface as the distance between the walking surface and the foot.

When the foot part approaches the landing position, the foot part comes into contact with the walking surface first. Since the bottom of the foot is elastically supported on the lower surface of the foot, the foot rests at a position and posture angle that matches the unevenness of the walking surface regardless of the posture angle of the foot portion. Accordingly, the posture angle of the sole becomes the posture angle that allows the foot to stably land. Since the correction amount is set so as to reduce the distance deviation between the foot portion and the sole portion, the target position / posture angle to which such a correction amount is added becomes the position / posture angle of the sole portion. Therefore, by driving the leg joint so that the foot follows the target position / posture angle to which the correction amount is added, the foot can be stably landed.
Since the sole distance sensor only needs to be able to measure the distance of at least three points, it can be manufactured at a lower cost than a distance sensor arranged in a grid or a distance sensor using image processing. A legged robot that smoothly walks by correcting gait data in real time according to the state of unevenness or the like of the walking surface before the foot reaches the walking surface can be realized at low cost.

  According to the present invention, the target position / posture angle of the foot part in the gait data is corrected in real time according to the state of the step or unevenness of the walking surface before the foot part reaches the walking surface. A legged robot that can walk smoothly can be realized.

The main features of the examples are listed.
(First feature) When the measurement distance of the sole distance sensor becomes a predetermined value or less, it is determined that the sole touches the walking surface.

First Embodiment FIG. 1 is a schematic diagram of a legged robot 1 according to a first embodiment. FIG. 1A is a schematic front view of the legged robot 1, and FIG. 1B is a schematic side view of the legged robot 1.
The legged robot 1 according to the present embodiment is a two-legged robot having a body part 10 and two leg parts 11L and 11R connected to the body part 10.
The body unit 10 includes a controller 100, an acceleration sensor 52, and an attitude angle sensor 54. The attitude angle sensor 54 is specifically a three-axis tilt sensor, and detects the tilt angle around the three axes of the body portion 10. Outputs of the acceleration sensor 52 and the attitude angle sensor 54 are sent to the controller 100.
The right leg 11R includes a right first link 12R, a right second link 16R, a right third link 20R, a right fourth link 24R, a right fifth link 28R, a right sixth link 32R, a right seventh link 36R, and a right foot. A flat link 40R is provided. The right leg 11R includes the right first joint 14R, the right second joint 18R, the right third joint 22R, the right fourth joint 26R, the right fifth joint 30R, the right sixth joint 34R, and the right seventh joint 38R. Have
One end of the right first link 12R is connected to the body part 10, and the other end is connected to the right first joint 14R. One end of the right second link 16R is connected to the right first joint 14R, and the other end is connected to the right second joint 18R. Similarly, both ends of the right third link 20R are connected to the right second joint 18R and the right third joint 22R, respectively. Both ends of the right fourth link 24R are connected to the right third joint 22R and the right fourth joint 26R, respectively. Both ends of the right fifth link 28R are connected to the right fourth joint 26R and the right fifth joint 30R, respectively. Both ends of the right sixth link 32R are connected to the right fifth joint 30R and the right sixth joint 34R, respectively. Both ends of the right seventh link 36R are connected to the right sixth joint 34R and the right seventh joint 38R, respectively. One end of the right foot link 40R is connected to the right seventh joint 38R.
Four distance sensors 50Ra, 50Rb, 50Rc, and 50Rd are installed on the lower surface of the right foot link 40R. The four distance sensors are respectively attached near the four corners of the right foot link 40R, and detect the distance between the lower surface of the right foot link 40R and the walking surface g during the operation of the legged robot 1. The output of each distance sensor is sent to the controller 100. The distance sensor 50Rd is not shown in FIG. 1 because it is located on the back side of the paper surface of the distance sensor 50Rb in FIG. 1A and on the back side of the paper surface of the distance sensor 50Rc in FIG. 1B. Absent.
The left leg 11L includes a left first link 12L, a left second link 16L, a left third link 20L, a left fourth link 24L, a left fifth link 28L, a left sixth link 32L, a left seventh link 36L, and a left foot It has a flat link 40L. The left leg 11L includes a left first joint 14L, a left second joint 18L, a left third joint 22L, a left fourth joint 26L, a left fifth joint 30L, a left sixth joint 34L, and a left seventh joint 38L. Have Four distance sensors 50La, 50Lb, 50Lc, and 50Ld are installed on the lower surface of the left foot link 40L. Since the links, the joints, and the four distance sensors in the left leg portion 11R are the same as the right leg portion 11R, the description thereof is omitted.
Each joint incorporates a motor (not shown) and an encoder (not shown), and the relative rotation angle of the connected link can be changed by driving the motor. A drive command signal to the motor is sent from the controller 100. In FIG. 1, the rotation axis of each joint is expressed as follows. A joint having a rotation axis extending in a direction perpendicular to the paper surface in the drawing is represented by a circle like a right first joint 14R shown in FIG. A joint having a rotation axis extending in the left-right direction on the drawing is represented by a rectangle with diagonal lines in the up-down direction of the drawing, like a right second joint 18R shown in FIG. A joint having a rotation axis extending in the vertical direction on the drawing is represented by a rectangle with diagonal lines in the horizontal direction of the drawing, like a right third joint 22R shown in FIG.

  FIG. 2 shows a block diagram of the legged robot 1. In FIG. 2, the right leg portion 11R and the left leg portion 11L are collectively referred to as the leg portion 11. Further, the four distance sensors 50Ra, 50Rb, 50Rc, and 50Rd attached to the right foot link 40R and the four distance sensors 50La, 50Lb, 50Lc, and 50Ld attached to the left foot link 40L are combined. This is referred to as sensor 50. Further, the motors provided at the joints are collectively referred to as a motor 51. Hereinafter, the right foot link 40R and the right foot link 40L are collectively referred to as the foot link 40.

The controller 100 has a storage device 102. Gait data 103 is stored in the storage device 102. The gait data 103 includes time-dependent data of the target position / posture angle of the torso portion 10 and time-dependent data of the target position / posture angle of the foot link 40. The gait data 103 is created on a computer in advance. By simulating the movement of the legged robot 1 when walking on the walking surface g on the computer, the position and orientation angle of the foot link 40 and the position of the body part 10 when the legged robot 1 walks smoothly A trajectory of the attitude angle is obtained. The obtained trajectory becomes the time-dependent data of the target position / posture angle of the foot link 40 and the time-dependent data of the target position / posture angle of the trunk portion 10 (that is, the gait data 103).
In this embodiment, an absolute coordinate system xyz is defined as shown in FIG. The walking surface g is a plane coinciding with the xy plane. The target positions of the foot link 40 and the torso part 10 are given as coordinate values on the xyz coordinate system. The target posture angle is given by a rotation angle (referred to as roll angle, pitch angle, or yaw angle) around each axis in the xyz coordinate system. Since the walking plane g is defined with respect to the xyz coordinate system, the target position / posture angle of the foot link 40 expressed in the xyz coordinate system is the relative target position / posture angle of the foot link 40 with respect to the walking plane g. Will be expressed.

First, the process performed in the controller 100 will be outlined.
The target position / posture angle of the foot link 40 of the gait data and the target position / posture angle of the body part 10 are sent to the joint angle conversion unit 114 after various correction amounts are added. The correction amount will be described later. The joint angle conversion unit 114 converts the target position / posture angle of the foot link 40 and the target position / posture angle of the body unit 10 into joint angles of the joints of the leg unit 11. This conversion is called so-called inverse conversion. The joint angle obtained by the joint angle conversion unit 114 is sent to the motor driver 116 as a joint angle command value. The motor driver 116 drives the motor 51 attached to each joint of the leg 11 and controls the joint angle of each joint so as to match the joint angle command value sent from the joint angle conversion unit 114. As a result, the legged robot 1 walks while following the target position / posture angle of the foot link 40 and the torso 10 of the gait data.

Each process performed in the controller 100 will be described.
Of the gait data 103 stored in the storage device 102, the data on the target position / posture angle of the foot link 40 is sent to the target distance calculation unit 104. The target distance calculation unit 104 obtains the target distance between the walking surface g and the foot link 40 on the gait data from the target position / posture angle data of the foot link 40. More specifically, the target distance calculation unit 104 calculates a distance (target distance) from a point on the foot link to which the distance sensor 50 is attached to the walking surface. The target position / posture angle of the foot link 40 is a target position and posture angle at a reference point set on the foot link 40. Since the positional relationship between the reference point and each distance sensor 50 is known, the distance between the attachment point of each distance sensor and the virtual walking surface g can be calculated from the target position / posture angle of the reference point. In this embodiment, since four distance sensors are attached to one foot link, the target distance calculation unit 104 calculates four target distances for one foot link.
In the present embodiment, as shown in FIG. 1, the walking plane g is a plane that coincides with the xy plane of the absolute coordinate system. Accordingly, among the target position / posture angle data of the foot link 40 expressed in the absolute coordinate system xyz, the position value in the z-axis direction is directly between the walking surface g and the foot link 40 at the reference point of the foot link 40. This is the target distance.
The target distance obtained by the target distance calculation unit 104 is sent to the distance deviation calculation unit 106. The distance deviation calculation unit 106 receives the output of each distance sensor 50 attached to the foot link 40. The distance deviation calculation unit 106 calculates, for each distance sensor, a deviation between the target distance and the distance between the actual walking surface measured by the distance sensor and the foot link. The obtained distance deviation is sent to the PID control unit 108. The PID control unit 108 is constructed so as to properly converge control by the following feedback loop. The feedback loop means that the measured value of the distance sensor 50 attached to the leg 11 is a distance deviation calculation unit 106, a PID control unit 108, a correction amount calculation unit 110, an adder 112, a joint angle conversion unit 114, and a motor. This loop affects the motor 51 that drives the joint of the leg 11 via the driver 116. Since the specific function of the PID control unit 108 is the same as that of a general PID controller, the description thereof is omitted. The value of the distance deviation output from the distance deviation calculation unit 106 is corrected by the PID controller. Hereinafter, the output of the PID controller is also referred to as the distance deviation.
The distance deviation output from the PID control unit 108 is sent to the correction amount calculation unit 110. The correction amount calculation unit 110 calculates the correction amount of the target position / posture angle of the foot link 40. The correction amount is set so that the distance deviation becomes smaller. The calculated correction amount is added to the target position / posture angle data of the foot link 40 of the gait data 103 by the adder 112 and input to the joint angle conversion unit 114.

The processes of the target distance calculation unit 104, the distance deviation calculation unit 106, and the correction amount calculation unit 110 described above will be described with reference to FIG. In FIG. 3, for ease of explanation, reference numerals different from the foot links 40 </ b> R and 40 </ b> L shown in FIG. 1 are attached to the foot links such as 40 a and 40 </ b> A.
The left side of FIG. 3 is a diagram showing a relative positional relationship between the foot link 40a and the walking surface g on the gait data at a certain timing (that is, in the virtual space). The right side of FIG. 3 is a diagram showing the positional relationship between the foot link 40A and the walking surface G in the real space at the same timing as the left side of FIG. That is, when the target position / posture angle of the foot link 40a is the vector r shown on the left side of FIG. 3, the actual legged robot is controlled based on the target position / posture angle, so that the walking with the actual foot link 40A is performed. The positional relationship of the plane G is the positional relationship shown on the right side of FIG.
In FIG. 3, to simplify the description, attention is paid to the difference between the position of the foot link 40a in the virtual space and the position of the foot link 40A in the real space, and it is assumed that there is no difference in the posture angle. Regarding the posture angles of the foot links 40a and 40A, it is assumed that the lower surface of the foot link is parallel to the walking surface in both the virtual space and the real space. The target position / posture angle of the foot link in the virtual space can be expressed by a vector r from the origin of the absolute coordinate system xyz set in the virtual space to the reference point P set in the foot link 40a. The absolute coordinate system xyz in the virtual space is set so that the xy plane coincides with the walking plane g.

A coordinate system XYZ shown on the right side of FIG. 3 represents an absolute coordinate system in the real space corresponding to the absolute coordinate system xyz in the virtual space.
Here, the walking plane G in the real space is shifted by ΔS in the Z-axis direction from the XY plane of the absolute coordinate system XYZ. That is, the position of the walking surface g in the virtual space does not match the walking surface G in the real space. This inconsistency occurs when the actual unevenness of the walking surface cannot be accurately reflected in the virtual space.
In FIG. 3, a position vector r (ie, target position) from the origin of the absolute coordinate system xyz in the virtual space to the reference point P of the foot link 40a, and the foot link from the origin of the absolute coordinate system XYZ in the real space. The position vector R to the reference point P of 40A does not match. That is, the foot link 40A of the actual legged robot does not accurately follow the target position. This is because, for example, there are elements that are not reflected in the legged robot in the virtual space, such as an actual legged robot assembly error.

The target distance calculation unit 104 obtains the target distance between the walking surface g and the foot link 40a in the virtual space from the gait data. The target distance is indicated by the symbol h _ref on the left side of FIG.
The distance sensor 50 measures the distance between the walking surface G and the foot link 40A in real space. On the right side of FIG. 3, the measured distance is indicated by the symbol h _mes .
The distance deviation calculation unit 106 obtains a distance deviation e _h between the target distance h _ref and the measured distance h _mes . In the example of FIG. 3, it is represented by e _h = h _ref −h _mes .
The PID control unit 108 corrects the distance deviation e _h based on the proportional control law, the differential control law, and the integral control law. In this embodiment, the output of the PID control unit 108 is also expressed by the distance deviation e _h .
Correction amount calculation unit 110 calculates a correction amount in the direction to reduce the distance deviation e _h correcting the target position r of the foot link 40a. In the example of FIG. 3, the correction amount is a distance e _{h in} the positive direction of the z axis in the absolute coordinate system xyz.
The correction amount calculated by the correction amount calculation unit 110 is added to the target position vector r on the gait data 103 by the adder 112. As shown on the left side of FIG. 3, the target position vector r + e _h the correction amount is added is a vector for moving upward the foot link 40a by a distance deviation e _h. On the left side of FIG. 3, showing the position of the foot link represented by the target position vector r + e _h by reference numeral 40b.
When each joint of the legged robot is driven based on the target position to which the correction amount is added, the position of the foot link 40A of the legged robot in the real space is the position indicated by reference numeral 40B on the right side of FIG. . That is, the target position vector r on the gait data 103 moves by a distance deviation e _h above the walking surface g by the correction amount. Moving Thus, the position of the foot link in a real space, just above the e _h from the position of the foot link (the position of the foot link 40A) when the legged robot is controlled based on the target position vector r (The position of the foot link 40B).

As described above, in this embodiment, the walking surface in the virtual space (that is, in the gait data) and the walking surface in the real space do not coincide with each other and the foot link does not follow the target position. Each joint of the legged robot is controlled so that the distance between the surface and the foot link is maintained even in the real space. Although the posture angle of the foot link is assumed to be constant in FIG. 3, the distance to the walking surface is measured at each of the three points on the lower surface of the foot (three points that are not aligned on a straight line), and the distance at each of the three points. By obtaining the deviation, the same processing can be performed for the posture angle. By such processing, the posture angle of the foot link with respect to the walking surface in the virtual space can be matched with the posture angle of the foot link with respect to the walking surface in the real space.
The process illustrated in FIG. 3 is performed for each control sampling. Therefore, the foot link in the real space can be landed at the timing when the foot link is landed in the virtual space. Furthermore, the posture angle of the foot link with respect to the walking surface when landing is also able to match the posture angle in the virtual space with the posture angle in the real space. That is, with the above correction amount, the operation of the foot link with respect to the walking surface in the real space can be brought close to the operation at the time of simulation performed when creating the gait data in the virtual space. By doing so, it is possible to realize a legged robot that smoothly walks on a walking surface whose surface such as unevenness is unknown.
The processes of the target distance calculation unit 104, the distance deviation calculation unit 106, and the correction amount calculation unit 110 have been described using the simplified example shown in FIG. More specific processing will be described later.

Returning to FIG. 2, the description of the processing of the controller 100 will be continued.
Of the gait data 103, the target position / posture angle of the body 10 is sent to the adder 118. Of the target position / posture angles of the body unit 10, the target posture angle data is sent to the adder 120.
The adder 120 receives the detection value of the attitude angle sensor 54 attached to the body unit 10. The adder 120 calculates a posture angle deviation between the target posture angle of the body unit 10 and the posture angle detected by the posture angle sensor 54. The calculated attitude angle deviation is sent to the PID control unit 122. The PID control unit 122 performs feedback control of the posture angle of the body unit 10. Hereinafter, this feedback control is referred to as attitude angle feedback.
Of the target position / posture angle of the body unit 10, the target position data is sent to the differentiator 126 twice. In the second differentiator 126, the target position data is differentiated twice, and the acceleration of the target position, that is, the target acceleration of the body portion 10 is output. The output target acceleration is sent to the adder 128. The adder 128 receives a detection value of the acceleration sensor 52 attached to the body unit 10. The adder 128 calculates an acceleration deviation between the target acceleration of the body unit 10 and the acceleration detected by the acceleration posture angle sensor 52. The calculated acceleration deviation is sent to the PID control unit 130. The PID control unit 130 performs feedback control of the acceleration of the body unit 10. Since the functions of the PID control unit 122 and the PID control unit 130 are the same as those of a general PID controller, the description thereof is omitted.
The output of the PID control unit 122 and the output of the PID control unit 130 are sent to the adder 118, added to the target position / posture angle of the body unit 10 on the gait data 103, and sent to the joint angle conversion unit 114.

Posture angle feedback is particularly important for the legged robot 1 of the present embodiment, which is a biped robot that moves dynamically. As described above, the correction amount of the target position / posture angle of the foot link 40 is calculated according to the distance between the walking surface and the foot link 40. The operation of the body part 10 is not taken into account in the correction amount calculation process. That is, the stability of walking is not taken into consideration in the correction amount calculation process. In dynamic walking, it is known that a robot falls if a predetermined condition is not satisfied. The condition is, for example, that the ZMP exists in a convex hull (generally called a support polygon) in the contact area between the foot of the stance and the walking surface. Such a condition is hereinafter referred to as a constraint condition.
When creating gait data by simulation on a computer, gait data is created so as to satisfy the constraint conditions. If an amount of correction is added to the target position / posture angle of the foot link on the walking data regardless of the action of the torso during an actual walking motion, there is a possibility that the constraint condition is not satisfied. The inventors of the present application simulated the relative positional relationship between the foot and the torso in the absolute coordinate system (sometimes called the global coordinate system or the world coordinate system) during actual walking when creating gait data. If you follow the relative positional relationship between the foot and torso in the absolute coordinate system of the virtual legged robot, the foot link position and posture angle during actual walking We found that it is easy to satisfy the constraints even if it deviates from the target position and posture angle. Therefore, the legged robot of this embodiment is provided with posture angle feedback. As a result, a legged robot that does not fall over even if the correction amount is added to the data of the target position / posture angle of the foot link has been realized.

The processing of the distance deviation calculation unit 106 and the correction amount calculation unit 110 will be described more specifically. Here again, the symbols (measurement distance h _mes : distance between the foot link 40 and the walking surface measured by the distance sensor 50 attached to the foot link 40, target distance h _ref : attachment of each distance sensor 50. The distance on the gait data between the foot link 40 and the walking surface at the position, the distance deviation e _h : the deviation between the target distance h _ref and the measured distance h _mes ) is used.
The correction amount of the foot link 40 described above need not be calculated over the entire free leg period. Therefore, when the distance between the foot link 40 and the walking surface is a predetermined distance or more, the deviation is set to zero. Further, when the foot link 40 is landing, if a small steady-state deviation occurs in the distance deviation due to, for example, a rounding error during the calculation, the integral amount of the steady-state deviation by the integral term of the PID control unit 108 becomes the correction amount. I will join. Even if the rounding error at the time of calculation is small, the integration amount may become so large that it cannot be ignored for control. Therefore, it is necessary to prevent such a steady deviation from affecting the correction amount. Therefore, in this embodiment, a dead zone is provided so that the correction amount is not calculated when the distance between the foot link 40 and the walking surface is a predetermined value or less. That is, the correction amount is calculated only when the distance between the foot link 40 and the walking surface is within a predetermined range. Such a predetermined range is referred to as a control distance range. FIG. 4 shows a method for calculating the distance deviation e _h when the control distance range is set. As the control distance range, the same range is set for both the target distance h _ref in the virtual space and the measurement distance h _mes in the real space. The upper limit value of the range is expressed as h _{th_H,} and the lower limit value of the range is expressed as h _{th_L} .
The case where both the target distance h _ref and the measured distance h _mes exceed the upper limit value h _{th_H} corresponds to the column (1) in FIG. In this case, the distance deviation e _h = 0. Further, the case where both the target distance h _ref and the measured distance h _mes are smaller than the lower limit value h _{th_L} corresponds to the column (9) in FIG. Also in this case, the distance deviation e _h = 0. By setting e _h = 0, the correction amount for the target position / posture angle of the foot link 40 is also zero.
The case where the target distance h _ref is larger than the upper limit value h _{th_H} and the measurement distance h _mes is within the control distance range corresponds to the column (2) in FIG. In this case, the distance deviation e _h = h _ref −h _mes is set. The case where the target distance h _ref is larger than the upper limit value h _{th_H} and the measurement distance h _mes is smaller than the lower limit value h _{th_L} corresponds to the column (3) in FIG. In this case, the distance deviation e _h = h _ref −h _{th_L} . According to the values of the target distance h _ref and the measured distance h _mes , e _h is calculated as shown in the columns (1) to (9) of FIG.

FIG. 5 and FIG. 6 illustrate the change with time of the distance deviation e _h accompanying the change with time of the target distance h _ref and the measurement distance h _mes . FIG. 5A shows changes over time in the target distance h _ref and the measurement distance h _mes . FIG. 5B shows the change over time of the distance deviation e _h corresponding to FIG.
The example of FIG. 5 is a case where the measurement distance h _mes is larger than the target distance h _ref . In the period a shown in FIG. 5, both the target distance h _ref and the measurement distance h _mes are larger than the upper limit value h _{th_H of the} distance. Therefore, in the period a, the calculation formula shown in the column (1) of FIG. 4 is adopted, and the distance deviation e _h = 0. In the period b, the measurement distance h _mes is larger than the upper limit value h th — _H, but the target distance h _ref is within the control distance range. In this period b, the calculation formula shown in the column (4) in FIG. 4 is adopted. In the period c, both the measurement distance h _mes and the target distance h _ref are within the control distance range. In this period c, the calculation formula shown in the column (5) of FIG. 4 is employed. In the period d, the measurement distance h _mes is within the control distance range, but the target distance h _ref is smaller than the lower limit value h _{th_L} . In this period d, the calculation formula shown in the column (8) of FIG. 4 is adopted. In the period e, both the target distance h _ref and the measurement distance h _mes are smaller than the lower limit value h _{th_L} . Therefore, in the period e, the calculation formula shown in the column (9) of FIG. 4 is adopted, and the distance deviation e _h = 0. Thus, the distance deviation e _h is as shown in FIG.
The example of FIG. 6 is a case where the measurement distance h _mes is smaller than the target distance h _ref . FIG. _6A shows the change over time of the target distance h _ref and the measurement distance h _mes . FIG. 6B shows a change with time of the distance deviation e _h corresponding to FIG. The graph shown by a broken line in FIG. 6 (B) is an output when applying a low-pass filter to the distance deviation e _h.
In the period a shown in FIG. 6, both the target distance h _ref and the measurement distance h _mes are larger than the upper limit value h _{th_H of the} distance. Therefore, in the period a, the calculation formula shown in the column (1) of FIG. 4 is adopted, and the distance deviation e _h = 0. In the period b, the measurement distance h _mes is within the control distance range, but the target distance h _ref is larger than the upper limit value h _{th_H} . In this period b, the calculation formula shown in the column (2) in FIG. 4 is adopted.
In the period c, both the measurement distance h _mes and the target distance h _ref are within the control distance range. In this period c, the calculation formula shown in the column (5) of FIG. 4 is employed. In the period d, the measurement distance h _mes is smaller than the lower limit value h _{th_L,} but the target distance h _ref is within the control distance range. In this period d, the calculation formula shown in the column (6) of FIG. 4 is adopted. In the period e, both the target distance h _ref and the measurement distance h _mes are smaller than the lower limit value h _{th_L} . Therefore, in the period e, the calculation formula shown in the column (9) of FIG. 4 is adopted, and the distance deviation e _h = 0. As described above, the change with time of the distance deviation e _h is a graph indicated by a solid line in FIG. Here, according to the calculation formula of FIG. 4, the distance deviation e _h changes stepwise at the boundary between the period a and the period b and the boundary between the period d and the period e. Therefore, it is also preferable to apply a low pass filter to the distance deviation eh. The output of the low-pass filter is a graph indicated by a dotted line in FIG. As can be seen from the dotted line in the graph of FIG. 6 (B), the can be removed stepwise change in distance deviation e _h. As a result, the correction amount can be prevented from changing stepwise.

ここで、（１）式の左辺［Δφ_Ｒ、Δθ_Ｒ、Δｚ_Ｒ］は、右足平リンク４０Ｒの目標位置姿勢角の補正量ベクトルである。Δφ_ＲとΔθ_Ｒは、歩行面に平行な面を規定する２直線回りの姿勢角の補正量である。Δｚ_Ｒは、歩行面に直交する方向の位置の補正量である。同様に、（２）式の左辺［Δφ_Ｌ、Δθ_Ｌ、Δｚ_Ｌ］は、左足平リンク４０Ｌの目標位置姿勢角の補正量ベクトルである。Δφ_ＬとΔθ_Ｌは、歩行面に平行な面を規定する２直線回りの姿勢角の補正量である。Δｚ_Ｌは、歩行面に直交する方向の位置の補正量である。
（１）式の右辺の［Ｅ_Ｒ１、Ｅ_Ｒ２、Ｅ_Ｒ３、Ｅ_Ｒ４］は、右足平リンク４０Ｒの４つの距離偏差を表す。ここで、左足平リンク４０Ｒには４つの距離センサ５０Ｒａ、５０Ｒｂ、５０Ｒｃ、及び５０Ｒｄが取り付けられている。目標距離算出部１０４は、足平リンクにおける夫々の距離センサの取り付け点から仮想空間内の歩行面までの距離（目標距離）を算出する。距離偏差［Ｅ_Ｒ１、Ｅ_Ｒ２、Ｅ_Ｒ３、Ｅ_Ｒ４］は、距離センサの取り付け点に対応した目標距離と距離センサが計測した距離の差である。同様に（２）式の右辺の［Ｅ_Ｌ１、Ｅ_Ｌ２、Ｅ_Ｌ３、Ｅ_Ｌ４］は、左足平リンク４０Ｌの４つの距離偏差を表す。
（１）式右辺のＸ_Ｒｉｊ（ｉ＝１，２，３，４、ｊ＝１，２，３，４）は、右足平リンク４０Ｒの幾何学的形状と、４つの距離センサの取り付け点から決定される定数である。ここで、（１）の左辺に示されているように、右足平リンク４０Ｒの目標位置姿勢角の補正量ベクトルは３つの変数を有している。従って、右足平リンク４０Ｒの下面の３点（一直線上に並ばない３点）と夫々の点から歩行面までの距離が計測できれば補正量ベクトルの３つの変数を決定できる。即ち、３点の距離偏差が計測できれば補正量ベクトルの３つの変数を決定できる。本実施例では、右足リンク４０Ｒの距離偏差［Ｅ_Ｒ１、Ｅ_Ｒ２、Ｅ_Ｒ３、Ｅ_Ｒ４］は、４つの変数を有している。即ち、左辺の３つの変数に対して右辺の距離変数は冗長である。本実施例ではこの冗長性を活用するようにＸ_Ｒｉｊが決定される。具体的には次の通り決定する。補正量［Δφ_Ｌ、Δθ_Ｌ、Δｚ_Ｌ］が決定されたと仮定したときの４つの距離偏差を［ＥＥ_Ｒ１、ＥＥ_Ｒ２、ＥＥ_Ｒ３、ＥＥ_Ｒ４］とする。Ｘ_Ｒｉｊは、誤差＝（ＥＥ_Ｒ１−Ｅ_Ｒ１）^２＋（ＥＥ_Ｒ２−Ｅ_Ｒ２）^２＋（ＥＥ_Ｒ３−Ｅ_Ｒ３）^２＋（ＥＥ_Ｒ４−Ｅ_Ｒ４）^２が最小となるように決定される。即ち、Ｘ_Ｒｉｊは、計測距離に基づく距離偏差（Ｅ_Ｒ１等）と、補正量を決定したと仮定したときの距離偏差（ＥＥ_Ｒ１等）との誤差が最小となるように決定する。（２）式のＸ_Ｌｉｊ（ｉ＝１，２，３，４、Ｊ＝１，２，３，４）も同様に決定する。 Next, the processing of the correction amount calculation unit 110 will be described more specifically. The correction amount calculation unit 110 calculates the correction amount of the target position / posture angle of the right foot link 40R and the left foot link 40L by the following equations (1) and (2).

Here, the left side [Δφ _R , Δθ _R , Δz _R ] of the equation (1) is a correction amount vector of the target position / posture angle of the right foot link 40R. Δφ _R and Δθ _R are correction amounts of posture angles around two straight lines that define a plane parallel to the walking surface. Δz _R is the correction amount of the position in the direction orthogonal to the walking surface. Similarly, the left side [Δφ _L , Δθ _L , Δz _L ] of the expression (2) is a correction amount vector of the target position / posture angle of the left foot link 40L. Δφ _L and Δθ _L are correction amounts of posture angles around two straight lines that define a plane parallel to the walking plane. Δz _L is a correction amount of a position in a direction orthogonal to the walking surface.
[E _R1 , E _R2 , E _R3 , E _R4 ] on the right side of the equation (1) represents four distance deviations of the right foot link 40R. Here, four distance sensors 50Ra, 50Rb, 50Rc, and 50Rd are attached to the left foot link 40R. The target distance calculation unit 104 calculates the distance (target distance) from the attachment point of each distance sensor on the foot link to the walking surface in the virtual space. The distance deviation [E _R1 , E _R2 , E _R3 , E _R4 ] is a difference between the target distance corresponding to the attachment point of the distance sensor and the distance measured by the distance sensor. Similarly, [E _L1 , E _L2 , E _L3 , E _L4 ] on the right side of the expression (2) represents four distance deviations of the left foot link 40L.
X _Rij (i = 1, 2, 3, 4, j = 1, 2, 3, 4) on the right side of equation (1) is determined from the geometric shape of the right foot link 40R and the attachment points of the four distance sensors. A constant to be determined. Here, as shown on the left side of (1), the correction amount vector of the target position / posture angle of the right foot link 40R has three variables. Therefore, if the distance from the three points on the lower surface of the right foot link 40R (three points that are not aligned on a straight line) and the respective points to the walking surface can be measured, the three variables of the correction amount vector can be determined. That is, if the three-point distance deviation can be measured, the three variables of the correction amount vector can be determined. In the present embodiment, the distance deviation [E _R1 , E _R2 , E _R3 , E _R4 ] of the right foot link 40R has four variables. That is, the distance variable on the right side is redundant with respect to the three variables on the left side. In this embodiment, X _Rij is determined so as to utilize this redundancy. Specifically, it is determined as follows. The four distance deviations when it is assumed that the correction amounts [Δφ _L , Δθ _L , Δz _L ] are determined are [EE _R1 , EE _R2 , EE _R3 , EE _R4 ]. X _Rij is determined such that error = (EE _R1 −E _R1 ) ² + (EE _R2 −E _R2 ) ² + (EE _R3 −E _R3 ) ² + (EE _R4 −E _R4 ) ² is minimized. . That, X _Rij is the distance deviation based on the measurement distance (E _R1 etc.), the error between the distance deviation, assuming that to determine the correction amount (EE _R1 etc.) are determined so as to minimize. X _Lij (i = 1, 2, 3, 4, J = 1, 2, 3, 4) in equation (2) is determined in the same manner.

左辺のΔＺ_Ｒ、ΔＺ_Ｌが、足平リンクの歩行面に垂直方向の補正量の最終的な値となる。（３）式は次の意味を有する。立脚の足平リンクにおいては、足平リンクと歩行面との距離偏差はゼロであるため、図４の（４）の欄に示す場合に相当する。従って、補正量はゼロとなる。即ち、Δｚ_Ｒ、Δｚ_Ｌのいずれか一方は常にゼロとなる。仮に左脚が立脚であるとして、Δｚ_Ｒ＝Δｚ、Δｚ_Ｌ＝０とする。（３）式によりΔＺ_Ｒ＝Δｚ／２、ΔＺ_Ｌ＝−Δｚ／２となる。立脚側である左足平リンク４０Ｒの補正量は下方へΔｚ／２となる。立脚側である左足平リンク４０Ｌは接している歩行面より下方へずれることはできないので、補正量のΔｚ／２の大きさだけ胴体部１０の位置が上方へ移動することになる。一方、遊脚側の右足平リンク４０Ｒの目標位置が上方へΔｚ／２へ補正されるので、現実の右足平リンク４０Ｒも上方へΔｚ／２だけ移動する。胴体部１０が上方へΔｚ／２の大きさだけ移動し、遊脚側の右足平リンク４０Ｒが胴体部１０に対して上方へΔｚ／２の大きさだけ移動するので、結果として遊脚側である右足平リンク４０Ｒは歩行面に対してΔｚだけ上方へ移動する。（３）式によって、遊脚の足平リンクの補正量を半分にすることができる。残りの半分は立脚側の足平リンクの目標位置の補正量とする。そうすることで、遊脚側の足平リンクの目標位置の補正量を小さくすることができる。歩容データに対する補正量の大きさを小さくすることによって、本来の歩容データによる歩行動作に近い歩行を行わせることができる。（３）式による作用は、次のように換言することもできる。補正量算出部１１０は、距離偏差算出部１０６が算出した距離偏差を１／２にして遊脚の足平リンクの目標位置姿勢角に対する補正量のうち、歩行面と交差する方向の補正量を算出し、算出した補正量の符号を逆転させた補正量を立脚の足平リンクの補正量に設定する。 The correction amount conversion unit 110 further converts the correction amounts Δz _R and Δz _L in the direction perpendicular to the walking surface of the foot link 40 obtained by the equations (1) and (2) according to the following equation (3).

ΔZ _R and ΔZ _{L on the} left side are final values of the correction amount in the direction perpendicular to the walking surface of the foot link. The formula (3) has the following meaning. In the foot link of the standing leg, the distance deviation between the foot link and the walking surface is zero, which corresponds to the case shown in the column (4) in FIG. Accordingly, the correction amount is zero. That is, one of Δz _R and Δz _L is always zero. Assuming that the left leg is a standing leg, Δz _R = Δz and Δz _L = 0. From the equation (3), ΔZ _R = Δz / 2 and ΔZ _L = −Δz / 2. The correction amount of the left foot link 40R on the stance side is Δz / 2 downward. Since the left foot link 40L on the stance side cannot be displaced downward from the walking surface that is in contact with it, the position of the body part 10 moves upward by the amount of the correction amount Δz / 2. On the other hand, since the target position of the right foot link 40R on the free leg side is corrected upward to Δz / 2, the actual right foot link 40R also moves upward by Δz / 2. The trunk portion 10 moves upward by a magnitude of Δz / 2, and the right foot link 40R on the free leg side moves upward by a magnitude of Δz / 2 with respect to the trunk portion 10, resulting in the free leg side as a result. A certain right foot link 40R moves upward by Δz with respect to the walking surface. The correction amount of the foot link of the free leg can be halved by the equation (3). The other half is the correction amount of the target position of the foot link on the stance side. By doing so, the correction amount of the target position of the foot link on the free leg side can be reduced. By reducing the magnitude of the correction amount for the gait data, it is possible to perform walking that is close to the walking motion based on the original gait data. The action of the equation (3) can be rephrased as follows. The correction amount calculation unit 110 divides the distance deviation calculated by the distance deviation calculation unit 106 by half, and among the correction amounts for the target position / posture angle of the foot link of the free leg, the correction amount in the direction intersecting the walking surface is calculated. The correction amount obtained by reversing the sign of the calculated correction amount is set as the correction amount for the foot link of the stance.

Second Embodiment Next, a second embodiment will be described. This embodiment differs from the first embodiment in the structure of the foot link of the legged robot. The configuration of the legged robot excluding the foot link is the same as that of the legged robot 1 of FIG.
FIG. 7A shows a schematic diagram of the right foot link 140R of the legged robot of this embodiment. Since the left foot link has the same structure, the description thereof is omitted.
The right foot link 140R is opposed to the lower surface thereof, and has a sole 160 that is elastically supported by the right foot link 140R by the elastic member 152. The elastic member 152 is typically a spring. The right foot link 140R has four sole distance sensors 150Ra, 150Rb, 150Rc, and 150Rd on the lower surface thereof. Each sole distance sensor measures the distance between the lower surface of the right foot link 140R and the sole 160. The four sole distance sensors 150Ra, 150Rb, 150Rc, and 150Rd are attached to the four corners of the right foot link 140R. In other words, four sole distance sensors are attached to the lower surface of the right foot link 140R in such an arrangement that at least three of them do not line up in a straight line. FIG. 7A shows only two sole distance sensors 150Ra and 150Rc. The sole distance sensor 150Rb is located on the back side of the sole distance sensor 150Ra in the drawing, and the sole distance sensor 150Rd is located on the back side of the sole distance sensor 150Rc in the drawing, and is not shown.
The right seventh joint 38R and the right seventh link 36R shown in FIG. 7A are the same as the schematic robot 1 shown in FIG.

  Since the sole 160 is positioned below the right foot link 140R, the foot 160 comes into contact with the walking surface G before the right foot link 140R when the right leg is landing. Since the foot portion 160 is elastically supported by the right foot link 140R by the elastic member 152, the foot portion 160 stops at a stable posture angle with respect to the walking surface G regardless of the posture angle of the right foot link 140R. Here, “stable posture angle” means contacting with the walking surface G at at least three points that are not aligned on a straight line.

  The configuration of the legged robot of this embodiment will be described with reference to the block diagram of FIG. However, in this embodiment, the distance sensor 50 in FIG. Further, the target distance calculation unit 104 calculates the distance (target distance) between the sole part and the foot link in the virtual space instead of the distance (target distance) between the walking surface and the foot link in the virtual space. In addition, the distance calculation unit 106 obtains a distance deviation between the measured distance between the foot link and the sole measured by the sole distance sensor 150 and the target distance.

  Processing for calculating the correction amount of the foot link according to the present embodiment will be described. In order to simplify the description, in FIG. 7, attention is paid to the correction amount of the posture angle of the right foot link 140R around the straight line intersecting the paper surface. Therefore, attention is paid only to the measurement distances of the sole distance sensors 150Ra and 150Rc attached to the right foot link 140R. As shown in FIG. 7A, the measurement distance of the sole distance sensor 150Ra in a state where the sole part 160 is not in contact with the walking surface G is ha, and the measurement distance of the sole distance sensor 150Rc is hc. Moreover, the protrusion d exists in the walking surface G in the planned landing position of the right foot link 140R. It is assumed that the projection d does not exist on the virtual walking surface when creating the gait data.

FIG. 7B shows a state where the sole 160 is in contact with the walking surface G. Since the sole 160 in contact with the walking surface G is pressed against the walking surface G by the elastic member 152, the sole 160 stands still at a stable posture angle with respect to the walking surface G. As shown in FIG. 7B, the sole 160 is stably stationary in a state where the sole 160 is in contact with the walking surface G at the protruding portion d and other flat portions. That is, the sole 160 is stably stopped at a posture angle inclined with respect to the flat surface of the walking surface G. At this time, the measurement distance of the sole distance sensor 150Ra is Ha, and the measurement distance of the sole distance sensor 150Rc is Hc. As shown in FIG. 7B, Ha <Hc.
On the other hand, the protrusion d does not exist on the virtual walking surface. Accordingly, the target distance between the right foot link 140R and the sole 160 on the gait data at the attachment point of the sole distance sensor 150Ra is Ha _ref, and the right foot on the gait data at the attachment point of the sole distance sensor 150Rc. If the target distance between the flat link 140R and the sole 160 is Hc _ref , Ha _ref = Hc _ref . The correction amount calculation unit 110 (see FIG. 2) calculates the correction amount of the target posture angle so that the distance deviation between the target distance and the measurement distance at the attachment points of the sole part distance sensors 150Ra and 150Rc becomes small. Each joint is controlled based on the target position / posture angle to which the correction amount is added. As a result, the posture angle of the right foot link 140R is controlled so that the lower surface thereof is parallel to the sole 160. This is because a correction amount is calculated such that the measurement distance Ha of the sole distance sensor 150Ra and the measurement distance Hc of the sole distance sensor 150Rc are equal.
When the right foot link 140R is lowered while the lower surface of the right foot link 140R is parallel to the sole 160, the right foot link 140R is landed at a posture angle in contact with the sole 160 as shown in FIG. To do. Since the sole 160 is stationary at a stable posture angle on the actual walking surface G where the protrusion d already exists, the right foot link 140R is also landed at a stable posture angle.

The technique of the second embodiment provides a technique for stably landing the foot link without directly measuring the state of the walking surface such as unevenness. In order to directly measure the unevenness of the walking surface, a two-dimensional distance sensor having a resolution equivalent to the pitch of the unevenness is required. Such a distance sensor requires an image processing or a sensor in which a number of distance sensors for measuring the distance between two points are arranged in a grid. Both sensors are expensive. The technology of the second embodiment stably lands the foot link according to the unevenness of the walking surface by using at least three distance sensors that measure the distance between two points without using an expensive distance sensor. A legged robot is provided.
This technology focuses on the fact that it is only necessary to know in advance what kind of posture angle the bottom surface of the foot member stabilizes on the walking surface by adapting the foot link to the inclination or unevenness of the walking surface. . Therefore, it is assumed that a dummy plate having the same surface as the lower surface of the foot link is placed at the planned landing position. The dummy plate stops at a stable position and posture angle on the walking surface at the planned landing position by the action of gravity. The position and posture angle of the dummy plate at that time are determined as the target position and posture angle of the foot member. If each joint of the leg is driven so as to move the foot link to its target position / posture angle, the foot link is stably landed on the stationary dummy plate. In other words, the target position for stably landing the foot link by measuring the position and posture angle of the dummy plate placed on the existing unevenness without directly measuring the inclination and unevenness of the walking surface The attitude angle can be determined.

The legged robot of the second embodiment can also be expressed as follows. That is, the leg is connected to the torso, and is a legged robot that walks on the walking surface by driving the joint of the leg, and faces the lower surface of the foot link (foot) at the tip of the leg. Between the sole that is elastically supported by the foot and at least three points that are set on the lower surface of the foot and are not aligned on a straight line, and the sole that is in contact with the walking surface A legged robot comprising a sole distance sensor that measures the distance of the leg, and a drive control unit that drives a joint of the leg so that the measured distances are equal.
The bottom of the foot plays the role of the dummy plate described above. When the foot link approaches the landing position, the sole portion comes into contact with the walking surface first. Since the sole is elastically supported by the foot link, the foot rests at a position and posture angle suitable for the inclination and unevenness of the walking surface regardless of the posture angle of the foot link. The distance between the bottom of the foot link and the sole at at least three points is detected by the sole sensor. At least three points are set so as not to line up on a straight line. The leg joints are driven so that at least three measured distances are equal. Then, the lower surface of the foot link and the sole can be made parallel. The joint of the leg is driven so that the distance measured in that state becomes zero. By doing so, the foot link can be stably landed. In other words, the leg joints are driven so as to converge the distances to zero while maintaining the measured distances at substantially equal distances.
Note that the contact of the sole with the walking surface may be determined, for example, that the sole touches the walking surface when the measurement distance of the sole distance sensor becomes a predetermined value or less.
The sole distance sensor in the second embodiment corresponds to an example of a “distance sensor” in the claims.

  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.

For example, FIG. 8 shows a modified example of the foot link of the second embodiment. The right foot link 240R has a right toe portion 272R connected by a right joint 270R. By driving the right joint 270R, the relative angle of the right toe portion 272R with respect to the right foot link 240R can be changed. The right joint 270R can rotate around three axes. The bottom surface of the right foot link 240R includes a sole portion 260 that is elastically supported by the elastic member 152. The bottom surface of the right foot link 240R also includes plantar distance sensors 250Ra, 250Rb, 250Rc, and 250Rb that measure the distance from the plantar part. In FIG. 8, the illustration of the sole sensor 250Rb positioned on the back side of the paper surface of the sole distance sensor 250Ra and the sole sensor 250Rd positioned on the back side of the paper surface of the sole distance sensor 250Rc is omitted.
In addition, the bottom surface of the right toe portion 272R includes a toe foot sole portion 274 that is elastically supported by the elastic member 252. The lower surface of the right toe portion 272R also has toe foot sole distance sensors 276Ra, 276Rb, 276Rc, and 276Rb that measure the distance from the toe foot sole portion 274. In FIG. 8, the toe foot sole sensor 276Rb located on the back side of the toe foot sole distance sensor 276Ra and the toe foot sole sensor 276Rd located on the back side of the toe foot sole distance sensor 276Rc are omitted. Yes. Since the left foot link has the same structure as the right foot link 240R in FIG.
The foot link 240R can contact the walking surface by both the foot link 240R and the toe portion 272R. The foot link 240R and the toe portion 272R are connected by a right joint 270R that can rotate around three axes. Therefore, the foot link 240R and the toe portion 272R are stably landed on the walking surface by applying the processing shown in the second embodiment to the foot link 240R and the toe portion 272R. be able to.

Further, the sole distance sensor 150Ra shown in the second embodiment uses a distance sensor using a laser, an optical linear distance sensor using a reflector, a distance sensor using an ultrasonic wave, or an eddy current. A distance sensor can be used. In the second embodiment, the sole distance sensor 150Ra and the like are attached to the lower surface of the right foot link 140R, but may be attached to the sole 160 side.
It is also preferable to attach a cover that covers the periphery of the right foot link 140R and the sole portion 160. By attaching the cover, it is possible to prevent dust and the like from adhering to the light emitting unit and the light receiving unit, particularly when an optical distance sensor is used.

  Further, the PID control units 108, 122, and 130 shown in FIG. 2 are examples of control laws, and control laws other than PID control may be applied. For example, a control law consisting of one element or a control law consisting of two elements may be applied among PID (proportional, integral and derivative) elements. Further, a control law including a phase advance circuit, a phase delay circuit, or a secondary delay element may be used.

  In the first embodiment, four distance sensors are provided on one foot, one on each of the four corners of the foot link 40. There may be more distance sensors. For example, a number of distance sensors may be arranged in a grid pattern on the lower surface of the foot link 40. By doing so, the unevenness of the walking surface at the planned landing position of the foot link 40 can be detected. In addition, a flexible member such as rubber is elastically supported on the lower surface of the foot link, and distance sensors for measuring the distance between the flexible member and the lower surface of the foot link are arranged in a grid pattern on the lower surface of the foot link. May be. The unevenness of the walking surface can also be measured by displacing the flexible member along the unevenness of the walking surface and measuring the distance between the flexible member and the lower surface of the foot link at multiple points.

  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 exemplified in this specification or the drawings can achieve a plurality of objects at the same time, and has technical usefulness by achieving one of the objects.

It is a schematic diagram of the legged robot of an Example. It is a block diagram of the legged robot of an example. It is a figure explaining the relationship between a walking surface and a foot link. It is a figure which shows the calculation method of distance deviation. It is a figure which shows the example of the distance deviation when a target distance and measurement distance differ. It is a figure which shows the other example of a distance deviation when a target distance and measurement distance differ. FIG. 7A is a schematic diagram of a foot link in the second embodiment. FIG. 7B is a diagram illustrating a state in which the soles are in contact with the walking surface. FIG. 7C is a diagram showing a state when the foot link has landed. It is a schematic diagram of the other shape of the foot link which has a sole part.

Explanation of symbols

1: leg type robot 10: body part 11R, 11L: leg part 40a, 40A, 40b, 40B, 40R, 40L: foot link 50: distance sensor 51: motor 52: acceleration sensor 54: attitude angle sensor 100: controller 102 : Storage device 103: gait data 104: target distance calculation unit 106: distance deviation calculation units 108, 122, 130: PID control unit 110: correction amount calculation units 112, 118, 120, 128: adder 114: joint angle conversion Unit 116: motor driver 140R, 240R: foot link 150Ra, 150Rc: sole distance sensor 152, 252: elastic member 160, 260: sole

Claims (3)

Legs are connected to the torso, and are legged robots that walk on the walking surface by driving the joints of the legs,
A storage device for storing gait data describing a temporal change in a target position / posture angle of a foot portion of a leg tip with respect to a virtual walking surface in a simulation model;
A target distance calculation unit for obtaining a target distance between each of the at least three points that are set on the foot part and are not aligned on a straight line and a virtual walking surface from the gait data;
A distance sensor that measures the distance between each of the at least three points of the actual foot and the actual walking surface during walking;
A distance deviation calculating means for obtaining a deviation between a target distance of each point and a measured distance corresponding to each point;
Correction amount calculation means for obtaining a correction amount for correcting the target position and posture angle of the foot in the direction of reducing the deviation;
Adding means for adding the obtained correction amount to the target position / posture angle of the foot,
A drive control unit that drives the joint of the leg so that the foot follows the target position and posture angle to which the correction amount is added;
A legged robot characterized by comprising: It has posture angle detection means for detecting the posture angle of the body part,
Gait data includes data describing changes over time in the desired posture angle of the torso,
The legged robot according to claim 1, wherein the drive control unit drives the joint of the leg so that the posture angle detected by the posture angle detecting means follows the target posture angle. Opposite to the lower surface of the foot portion, further comprising a sole portion elastically supported by the foot portion,
The target distance calculation unit obtains a distance between each of the at least three points and a virtual sole contacting the virtual walking surface as a target distance,
3. The legged robot according to claim 1, wherein the distance sensor measures a distance between each of the at least three points and a sole contacting the actual walking surface.

Images