JP5268107B2 - Hybrid control device and method for multi-legged walking device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To simultaneously execute each of an optional leg part control and an autobalance control, even in an unknown environment such as irregular ground. <P>SOLUTION: This multi-leg walking type moving device walking by a plurality of legs includes a hybrid control device for simultaneously executing the optional leg part control and the autobalance control for automatically maintaining balance. The hybrid control device statically performs non-interference of the leg part control and the autobalance control according to a grounding state of each foot part. <P>COPYRIGHT: (C)2010,JPO&amp;INPIT

Description

  The present invention relates to a hybrid control device and method for enabling arbitrary leg control and auto balance control to be executed simultaneously in a multi-leg walking type mobile device that walks with a plurality of legs.

  Currently, the walking control of most multi-legged walking devices (walking robots) is performed on a trajectory plan basis. That is, on the assumption that the dynamic characteristics of the robot itself and the environment are known, a target trajectory for realizing a desired walking is calculated in advance based on a stabilization criterion such as a zero moment point (ZMP) (orbital trajectory). In the actual walking, the trajectory control is applied at the joint servo level so as to reproduce the trajectory plan. This is a method for separating the motion planning and control in walking of the robot, and has achieved some success, but there are also the following problems.

1) The dynamic characteristics of the robot and the environment must be known. To accurately perform trajectory planning, it is necessary to accurately construct not only the robot kinematic model but also the dynamic model including the interaction with the environment. There is. This is a difficult but not impossible task. Furthermore, even if a dynamic model is derived with great care, a so-called modeling error is unavoidable, and disturbance of walking control due to the modeling error always occurs at the time of execution. Therefore, some compensation means is required for stable walking control in an actual environment.

2) Trajectory control must be accurately realized ZMP described above is an amount that depends not only on the position and orientation of the robot but also on acceleration / angular acceleration. In other words, it means that the trajectory plan is not realized unless the acceleration / angular acceleration is accurately reproduced in the trajectory control. This is very difficult in practice and still requires some compensation means.

3) A large amount of sensor information is required for compensation. As described above, compensation means for modeling errors and trajectory control errors are required. For these compensations, joints usually provided in robots are required. In addition to the displacement sensor, additional sensors such as an inclination angle sensor, an acceleration sensor, a gyro sensor, a camera, and a force sensor are required.

  Because of the above problems, a lot of information and calculations are required to make the current walking robot walk stably, and its structure is complicated. Furthermore, robust walking was difficult on unknown rough terrain.

  Therefore, in order to realize robust walking even in unknown environments, walking control that is not based on a trajectory plan has been proposed. In other words, in the conventional trajectory planning-based method, the robot maintains the balance by controlling the movement of the whole body, but by actively manipulating the force applied to the ground with the foot (that is, stroking) This is a technique for maintaining balance.

  For example, Non-Patent Documents 1 to 4 propose a method that uses the amount of angular momentum of the entire robot as the controlled amount, not the trajectory control at the joint servo level. These methods have the advantage that the trajectory control of 2) above is not required because the trajectory planning of the robot is not performed and the angular momentum is directly feedback controlled.

  However, in order to construct the angular momentum feedback, it is necessary to accurately know in advance the dynamic characteristics that relate the speed / angular velocity of the joint of the robot and the total angular momentum. Furthermore, in these methods, it is implicitly assumed that the position and orientation of the robot as seen from the absolute coordinate system are known. In an unknown environment, an inclination angle sensor for measuring the absolute position and orientation of the robot, An acceleration sensor, a gyro sensor, a camera, etc. are required. Furthermore, when the robot receives an external force from other than the legs (such as pushing a wall with an arm), the influence of the force on the angular momentum must be explicitly known by using a force sensor.

  On the other hand, Patent Document 1 proposes a technique in which an ankle joint torque of a robot is an operation amount and an integral value of the ankle joint torque is a controlled amount. This is intended only to maintain the balance of a device having only a single leg, not a multi-legged walking type mobile device, but realizes robust balance maintenance control (auto balance control) even in an unknown environment.

  However, it is difficult to apply the control device as it is to a multi-legged walking type mobile device having a plurality of legs. In view of this, a method has been proposed in which the above control device is extended to a biped walking type mobile device (biped walking robot) having two legs (see Non-Patent Document 5 and Non-Patent Document 6).

  In the above control method, the generalized force applied to the center of gravity of the entire robot is used as the operation amount, and the floor reaction force center (CoP: Center of Pressure) of each leg in a state where the biped walking type moving device is in contact with both legs. CoP that is uniquely defined to integrate the values is used as the controlled amount. This realizes robust autobalance control in the biped walking type mobile device, but still has the following problems.

1) Difficulty in deriving CoP by floor reaction force sensor In the above control method, it is assumed that both legs are grounded on the same plane when deriving CoP that integrates both legs. In this case, it can be derived by measuring only the vertical component of the floor reaction force. However, when both legs are grounded on uneven ground that is not on the same plane (see Fig. 4), it is necessary to measure all components of the floor reaction force (6 directions x 2 legs = 12-axis force / torque in a three-dimensional space). There is. This means that an expensive and fragile 6-axis force sensor must be installed on each leg.

2) Difficulty in hybridization In general, in order for a multi-legged walking type mobile device to walk, two functions of “gait generation function” and “balance maintenance function” are important. Here, the gait generation function is what route the user takes from the current position to the destination, and how to move the left or right leg or foot to walk along that route. It is a function to plan what is good. The balance maintenance function is a function for preventing the vehicle from falling while stopped or walking.
Of these two functions, the balance maintaining function can be realized by applying a control method using the ankle joint torque as an operation amount, such as the above-described control method, as autobalance control. However, since the gait generation function cannot be realized only by auto balance control, it is necessary to introduce appropriate leg control that realizes the gait generation function and to make the hybrid capable of executing auto balance control and leg control at the same time. There is a case.

  However, in the above control method, the CoP that integrates both legs is the controlled variable, and the auto balance control is realized through the target trajectory of this CoP. Therefore, when it is desired to simultaneously realize the leg control other than the auto balance control. Will use one of the following two methods.

  The first is a method adopted in Non-Patent Documents 5 and 6. That is, since the movement of the center of gravity can be operated indirectly by changing the target trajectory of CoP, another leg control is introduced using this. However, this method cannot handle auto balance control and other leg control independently, and performs other leg control through auto balance control.

  The second is a method of extracting and applying only the leg motion that does not affect the movement of the center of gravity of the entire robot from the leg control input. According to this method, only an operation that does not affect the movement of the center of gravity can be realized quickly by an arbitrary leg control.

  As described above, if the two methods are combined, it is possible to achieve hybridization even by the above control method. However, as pointed out in Non-Patent Document 5, the leg control through the auto balance control has a slow response. In other words, it was not enough from the viewpoint of realizing walking.

Japanese Patent No. 3637387

Furujo, Yamada, “Dynamic Control of Biped Robot Considering Angular Momentum (Walking with Kicking in Supporting Both Legs)”, Transactions of the Society of Instrument and Control Engineers, vol. 22, no. 4, pp. 451-456, 1986 Sano, Furuso, “3D dynamic walking of a biped robot by angular momentum control”, Transactions of the Society of Instrument and Control Engineers, vol. 26, no. 4, pp. 459-466, 1990 Mitobe, Yajima, Nasu, “Control of walking robot by manipulating zero moment point”, Journal of the Robotics Society of Japan, vol. 18, no. 3, pp. 359-365,2000 Mitobe, Iksan, Shibata, Yamano, Nasu, “ZMP operation based on floor pressure and frictional force of walking robot and application to angular momentum control”, Journal of the Robotics Society of Japan, vol. 20, no. 5, pp. 515-520, 2002 Ito, Asano, Kawasaki, “Center-of-gravity movement during bipedal support with both floor reaction force control”, Journal of the Robotics Society of Japan, vol. 22, no. 4, pp. 535-542, 2004 Furuta, Amano, Ito, "Stepping motion control by CoP trajectory tracking", Proceedings of the 8th SICE System Integration Division Annual Conference, pp. 817-818, 2007

  Therefore, the problem to be solved by the present invention is to perform hybrid control of a multi-legged walking type mobile device capable of simultaneously executing arbitrary leg control and auto balance control even in an unknown environment such as rough terrain. It is to provide an apparatus and technique.

In order to solve the above problems, a multi-legged walking movement device according to the present invention is
In a multi-legged walking device that walks with multiple legs,
The leg includes a hybrid control device that simultaneously executes arbitrary leg control and auto balance control for automatically maintaining balance,
The hybrid control device statically performs non-interference between the leg control and the autobalance control according to the ground contact state of the foot.

Preferably,
The leg is independently provided with a joint actuator (LCA) on which leg control operates and a joint actuator (ACA) on which autobalance control operates.

Preferably,
In a single leg, the hybrid control device has a free leg phase in which the feet are completely grounded, a stance phase in which the feet are completely grounded, and a free leg in accordance with the ground contact state of each foot. There are three phases, a ground phase as a process of transition from the leg phase to the stance phase, and the control applied to the ACA is switched between auto balance control and appropriate control according to the phase transition.

Preferably,
The hybrid control device decouples the leg control applied to the LCA when there are two or more legs that are not in the free leg phase among all legs, and appropriately decouples otherwise. Switch the presence or absence.

Preferably,
The hybrid control device switches the decoupling of the leg control in the LCA by appropriately adjusting the weight coefficient in the control law.

Preferably,
The hybrid controller switches between auto balance control and appropriate control in ACA by appropriately adjusting the weighting factor, offset value, and integration start time of the integral term in the control law.

Preferably,
The sole of the leg is provided with a sensor for detecting the ground contact state, and the non-interference between the leg control and the auto balance control according to the values of the sensor for detecting the ground contact state and the angle sensor installed in the LCA and the ACA. To do.

In addition, the control method of the multi-legged walking device according to the present invention is as follows.
In the control method of a multi-legged walking device that walks with multiple legs,
The leg is a hybrid control method that simultaneously executes arbitrary leg control and auto balance control for automatically maintaining balance,
Non-interference between leg control and auto balance control is performed statically according to the ground contact state of the foot.

Preferably,
The leg is independently provided with a joint actuator (LCA) that performs leg control and a joint actuator (ACA) that performs autobalance control,
In a single leg, depending on the ground contact state of each foot, the swing leg phase in which the foot is completely grounded, the stance phase in which the foot is completely grounded, and the stance phase from the swing leg phase There are three phases of the ground phase as a process of transition to, and the control applied to ACA is switched between auto balance control and appropriate control according to the phase transition.

Preferably,
Of all the legs, when there are two or more legs that are not in the swing phase, the leg control applied to the LCA is made non-interfering. In other cases, the presence / absence of the non-interacting is appropriately switched.

Preferably,
The switching of the decoupling of the leg control in the LCA is performed by appropriately adjusting the weight coefficient in the control law.

Preferably,
Switching between auto balance control and appropriate control in ACA is performed by appropriately adjusting the weighting coefficient, offset value, and integration start time of the integral term in the control law.

  The present invention includes a hybrid control device that simultaneously executes arbitrary leg control and auto balance control for automatically performing balance maintenance, and the hybrid control device performs leg control according to the ground contact state of the foot. Statically decouples from auto balance control.

  As a result, in the multi-leg walking type mobile device, it is possible to apply leg control that realizes robust gait generation even in an unknown environment, and simultaneously execute auto balance control that realizes a robust balance maintaining function. According to the present invention, robust walking on an unknown rough terrain can be realized without requiring a dynamic model, trajectory planning, trajectory control, and a lot of sensor information.

The model figure which shows the bipedal walking robot on a sagittal plane. Explanatory drawing which shows the structure of a control mechanism. The model figure which shows the bipedal walking robot in the single leg support period. The model figure which shows the bipedal walking robot in a both leg support period. In the swing phase, (A) is a model diagram showing a foot, and (B) is a flowchart of auto balance control. In the ground phase, (A) is a model diagram showing a foot, and (B) is a flowchart of auto balance control. In the stance phase, (A) is a model diagram showing a foot, and (B) is a flowchart of auto balance control. The model figure explaining a foot part mechanism.

  DESCRIPTION OF EMBODIMENTS Hereinafter, embodiments of a hybrid control device and method of a multi-leg walking type moving apparatus according to the present invention will be described based on the drawings.

[Basic concept]
The reference coordinate system fixed to the absolute position of the inertial system is Σ _R , the origin is at the representative position of the body of the multi-legged walking device, and the body coordinate system fixed to the body of the multi-legged walking device is Σ _B. Unless otherwise stated, the following vectors and matrices, and expressed as viewed from the body coordinate system sigma _B rather than from the reference coordinate system sigma _R.

  In this embodiment, in order to simplify the explanation, a bipedal walking type movement device (bipedal walking robot) is adopted, and the movement is limited to a sagittal plane (side view) on a two-dimensional plane (FIG. 1). However, the method of the present embodiment can be extended to the movement of the multi-leg walking type moving apparatus in the three-dimensional space, and the generality is not lost.

  In the state where the foot i (i = a, b in the case of a bipedal walking robot) of the multi-legged walking device is in contact with the ground, a “virtual sole plane” as a substantial sole is defined for the foot i. Here, the “virtual plantar plane” refers to a plane including (passing through) all the grounding points that contact the ground on leveling or uneven ground.

For each plantar coordinate system Σ _iF fixed to the plantar plane of each foot, the plantar plane is the x _iF z _iF plane, the front toward the foot tip is the x _iF axis positive direction, and the plantar plane is above the normal line The y _iF axis positive direction and the z _iF axis are defined as a right-handed direction.

At this time, rotation around the x _iF axis is roll rotation, rotation around the z _iF axis is pitch rotation, and rotation around the y _iF axis is yaw rotation. In the three-dimensional space, roll rotation and pitch rotation among these become the “auto balance control axis” of the foot i. That is, the autobalance control axis refers to a degree of freedom for autobalance control on the sole coordinate system. And if it restrict | limits on a sagittal surface, only a pitch rotation will become the auto-balance control axis | shaft of the foot i.

The origin of the plantar coordinate system Σ _iF is set at an arbitrary position on the virtual plantar plane (however, it is preferable to place it at the center of the plantar as much as possible). This origin is used as the floor reaction force representative point, and the position and orientation viewed from the body coordinate system Σ _B is

と表現する。ここで、ｍは空間自由度を表し、三次元空間ではｍ＝６、二次元平面ではｍ＝３である。

It expresses. Here, m represents the degree of freedom in space, m = 6 in the three-dimensional space, and m = 3 in the two-dimensional plane.

From here, a biped robot (i = a, b) whose motion is limited to the sagittal plane (m = 3) will be described. Joint count n _i of each leg is n _i ≧ 6 if at least three degrees of freedom (n _i ≧ 3) must be (walking robot in a three-dimensional space. That is, each leg minimum m = 6 degrees of freedom is necessary.). At this time, variables are defined as follows.

但し、ｔは現在時刻である。ｎ_ＡＣは各足のオートバランス制御軸数とし、前記の通り、二次元平面上ではｎ_ＡＣ＝１であり、三次元空間上ではｎ_ＡＣ＝２である。

Where t is the current time. n _AC is the number of autobalance control axes of each foot, and as described above, n _AC = 1 on the two-dimensional plane and n _AC = 2 on the three-dimensional space.

The floor reaction torque τ _GR described above indicates the torque exerted by the robot's feet as a reaction against the ground. Therefore, the floor reaction torque τ _GR can be calculated from the measured value of the floor reaction force sensor that detects the distribution of the floor reaction force acting on the sole.

From here, an embodiment will be described in which each leg has three degrees of freedom (n _a = n _b = 3). In addition, even if each leg has a higher degree of freedom, the following description holds true also in the three-dimensional space.

[Control mechanism]
The configuration of the control mechanism for each leg will be described with reference to FIG. Each leg includes three joint actuators LCA1, LCA2, and ACA, and the first link portion 10, the second link portion 11, and the foot portion 5 are connected to the joint actuators LCA1, LCA2, and ACA. The multi-legged walking type movement apparatus includes two or more legs, and the joint actuator LCA1 is connected to the trunk (body).

  Each leg performs arbitrary leg control and auto balance control by a predetermined joint actuator. Control in which arbitrary leg control and auto balance control are simultaneously executed in this way is called “hybrid control”.

  For example (in the control in the joint space described below), the joint actuator ACA of the foot joint (third joint) in the foot 5 performs autobalance control, and the joint actuator LCA1 of the other joints (first joint and second joint) LCA2 performs arbitrary leg control. In joint actuators LCA1, LCA2, and ACA, joint displacement is detected by joint displacement sensors 3, 6, and 7.

  The joint actuators LCA1, LCA2, and ACA are controlled by the hybrid control device 1 and joint actuator driving devices 3, 6, and 7. By operating the controller 20, the hybrid controller 1 can drive the joint actuators LCA1 and LCA2, for example, with an arbitrary torque, and execute an arbitrary leg control.

  Furthermore, the hybrid control device 1 controls, for example, the joint actuator ACA based on an autobalance control program for maintaining balance.

The foot 5 includes a floor reaction force sensor 4 on the sole. The floor reaction force sensor 4 is, for example, a sheet-like pressure sensor, and the normal direction of the sheet surface is a detection axis. The floor reaction force sensor 4 outputs the measured floor reaction force distribution to the hybrid control device 1 and calculates the floor reaction torque τ _GR based on the measured value.

[Auto balance control during single leg support period]
Based on FIG. 3, the auto balance control in the case where the leg a is a standing leg (grounding leg) and the leg b is a free leg (non-grounding leg) will be described. As described above, since the pitch axis of the leg a is the _autobalance control axis, the effective _autobalance control axis displacement is only _θaAC .

The purpose of the _autobalance control in the single leg support period is to set the floor reaction torque τ _aGR = 0. In order to realize this control purpose, for example, when the control method disclosed in Patent Document 1 is applied, the _autobalance control torque command τ _{aACs in the} single leg support period is _expressed by the following equation (1).

但し、次のように変数を定義する。

However, variables are defined as follows.

In the above equation (1), the first and second terms are PD control terms for controlling the angle measured by the joint displacement sensor to a predetermined target angle θ _aACd , and the third term is a floor reaction torque. This is an integral term of floor reaction torque that is controlled so that τ _aGR = 0.

[Hybrid control during single leg support (joint space)]
Next, hybrid control that simultaneously executes arbitrary leg control and autobalance control will be described. In the hybrid control, it is necessary to appropriately distribute the auto balance control torque command τ _aACs defined by the expression (1) as τ _a to the actuator of the leg a. Furthermore, it is an object here to allow arbitrary leg control to act on τ _a simultaneously when the single leg _autobalance control τ _aACs is acting. A torque command τ _aLC by arbitrary leg control is determined as follows.

The most natural distribution in the joint space is to perform auto balance control by a specific joint actuator and arbitrary leg control by another joint actuator. That is, of the joint actuators driven around the pitch axis parallel to the pitch axis of the stance leg (leg a), n _AC = 1 actuator is an auto balance control actuator (ACA), and other n _a −n _AC = The two actuators are referred to as leg control actuators (LCA).
For example, if the third joint (ankle joint) of the leg a is selected as ACA, the joint displacement in the equation (1) is the displacement of the third joint, and θ _aAC = q _a3 .

The hybrid control torque command τ _{a in the} single leg support period is expressed by the following equation (2).

In the equation (2), an arbitrary leg control torque command τ _aLC is applied to the torque commands τ _a1 , τ _a2 acting on the joint actuators LCA1, LCA2 (first and second joints) (second term on the right side), The _autobalance control torque command τ _aACs is applied to the torque command τ _a3 acting on the joint actuator ACA (third joint) (first term on the right side).

  Therefore, in the single leg support period, any leg control and auto balance control are non-interfering in each joint actuator. However, from equation (2), in the torque command to the joint actuator ACA, since auto balance control is prioritized and leg control is ignored, it is preferable to perform the leg control on the premise thereof.

[Hybrid control during single leg support (work space)]
Next, a method different from the hybrid control in the joint space will be described. In this method, the _autobalance control torque command τ _aACs is equivalently generated in the entire leg a as viewed from the body.

Auto balance control axis, the pitch axis leg a of the generalized coordinates p _aF of the floor reaction force representative points on the foot a virtual plantar plane as viewed from the body coordinate system sigma _B. That is, the auto balance control shaft displacement in the equation (1) is θ _aAC = θ _aF (virtual sole plane angle viewed from the body).

In this case, it represents an auto balance control shaft displacement _θ aAC ₌ θ _aF and leg a joint displacement (displacement of all joints) q _a as seen from the body coordinate system sigma _B, the relationship between velocity in equation (3).

但し、次のように変数を定義する。

However, variables are defined as follows.

At this time, the generalized force command to the leg a actuator which is statically equivalent to the _autobalance control torque command τ _aACs according to the principle of virtual work is expressed by equation (4).

但し、

However,

  By the equation (4), the autobalance control in the single leg support period is realized by the driving force of the entire stance leg a. Further, hybrid control for performing leg control so as not to statically interfere with the autobalance control by the entire leg a in the formula (4) is represented by the formula (5).

（５）式において、

In equation (5),

であり、ある行列による写影に直交する（干渉しない）方向の写影行列を定義する。
但し、

And defines a mapping matrix in a direction orthogonal (non-interfering) to the mapping by a certain matrix.
However,

As described above, in the equation (5), by multiplying the orthogonal mapping matrix, only the component that does not interfere with the _autobalance control torque command τ _aACs is extracted from the arbitrary leg control torque command τ _aLC .

  Therefore, in the single leg support period, any leg control and auto balance control are non-interfering. However, from equation (5), auto-balance control is prioritized in the torque command to the joint actuator ACA, and components that interfere with auto-balance control in the leg control are ignored. It is preferable to carry out.

[Auto balance control during both leg support period]
Next, based on FIG. 4, the auto balance control when the legs a and b are both standing legs will be described. At this time, the stance a, pitch axis b is auto balance control shaft, effective auto balance control axis displacement is _AC theta.

The purpose of the autobalance control in the both-leg support period is to set the floor reaction torque τ _GR = 0. In order to realize this control purpose, for example, when the control method disclosed in Patent Document 1 is expanded and applied, the _autobalance control torque command τ _{ACd in the} both-leg support period is _expressed by the following equation (6).

但し、次のように変数を定義する。

However, variables are defined as follows.

  However, in equation (6), the autobalance control is independently applied to each of the legs a and b by making the gain a diagonal matrix. This extension is different from the method of Patent Document 1.

[Hybrid control during both leg support period]
Next, hybrid control will be described. In the hybrid control, it is necessary to appropriately distribute the auto balance control torque command τ _ACd defined by the equation (6) to the actuators of the legs a and b as τ _a and τ _b . Furthermore, it is an object here to allow arbitrary leg control to simultaneously act on τ _a and τ _b when the both-leg _autobalance control τ _ACd is acting.

  As with the single-leg support period, there are two types of distribution methods, joint space and work space. However, when auto balance control is distributed in the work space, both legs' auto balance control tries to drive the body directly. , The auto balance control of both legs cannot be made non-interfering (influencing each other), and the balance cannot be maintained. Therefore, the distribution in the both-leg support period is performed in the joint space.

  For example, if the third joint (ankle joint) of each of the legs a and b is selected as ACA, the displacement in the equation (6) becomes the following equation (7).

Further, the generalized coordinate vector representing the ACA displacement q _F, the generalized coordinate vector representing the LCA displacement When q _L, can be defined as the following equation (8).

Further, when the generalized force vector for driving ACA is τ _F and the generalized force vector for driving LCA is τ _L , it can be defined as the following equation (9).

An arbitrary leg control τ _LC acting on the LCA is defined as in the following equation (10). This takes out only the control input from τ _aLC and τ _bLC to the LCA.

Based on the above preparation, a hybrid control law in the joint space is derived. First, kinematic constraints are defined. In the both-leg support period, since both legs are grounded, not all the joints of the legs can move freely, but the degree of freedom is restricted. In other words, among the total degrees of freedom (n _a + n _b ) of both legs, m degrees of freedom are constrained by the foot being fixed to the ground, so the degrees of freedom left on the legs during the support period of both legs is (n _a + _n b -m) = 3. This constraint condition can be expressed by the following equation (11) as a joint velocity constraint equation.

但し、次のように変数を定義する。

However, variables are defined as follows.

Consider moving the indirectly q _F by moving under the above constraints, the q _L. At this time, the following equation (12) can be derived from the equation (11).

ここで、

here,

  Also, by considering the degree of freedom,

Therefore, the equation (12) becomes the following equation (14).

但し、

However,

とした。仮想仕事の原理より、次の（１６）式を導出できる。

It was. From the principle of virtual work, the following equation (16) can be derived.

Expression (16) is an expression representing τ _L which is statically equivalent to τ _F under the constraint condition (11) expression in the both-leg support period.

  Based on the above geometric considerations, the hybrid control law in the both-leg support period is derived.

ここで、

here,

但し、

However,

Under constraint condition (11), τ _L affects τ _F through geometric constraint, but by multiplying the orthogonal mapping matrix as in the above equation (18), any leg control τ _LC Only components that do not _statically interfere with the _autobalance control τ _ACd are extracted.

  Therefore, by applying the equations (17) and (18), it was possible to make a hybrid so that an arbitrary leg control and auto balance control are non-interfering. ACA can perform a hybrid operation in which auto balance control is performed, and LCA can realize arbitrary leg control within a range that does not statically affect auto balance control. However, in the command to ACA, priority is given to auto balance control, and components of the leg control that interfere with the auto balance control are ignored. Therefore, it is preferable to perform the leg control on the premise thereof.

[Hybrid control in walking motion]
In the above description, the hybrid control in which the arbitrary leg control and the auto balance control are simultaneously executed in the single leg support period and the both leg support period has been described. Here, since the biped walking operation is a transition from the single-leg support period to the both-leg support period and from the both-leg support period to the single-leg support period, hybrid control in biped walking will be described based on the above.
Here, for the sake of simplicity, the discussion will be made on the assumption that there are two legs (bipedal walking), but the following explanation is similarly applied to walking with three or more legs.

  In biped walking, it is preferable that the hybrid control between the single leg support period and the both leg support period is executed smoothly. Therefore, for each of the leg control and the autobalance control, the single-leg support period, the both-leg support period, and their transition behavior are defined below.

  As described above, in consideration of the both-leg support period, the hybrid control is performed in the joint space.

[Leg control in walking motion]
First, for the leg control during biped walking, the following equation (19) is defined with reference to the hybrid control law (18) in the both-leg support period.

但し、次のように変数を定義する。

However, variables are defined as follows.

At this time, in order to perform bipedal walking smoothly, the weighting factor w _LCw is switched as follows.

Here, in the both-leg support period, when both feet are grounded so that the above-mentioned equation (19) is equivalent to the above-mentioned equation (18) so that the leg control in both legs a and b does not interfere with the autobalance control. The change of w _{LCw must} already be completed and w _LCw = 1.

Also, in the single leg support period, from the above equation (2), no matter what leg control is performed, it does not interfere with auto balance control statically, so that the single leg can be executed as it is. The change of w _LCw is started from the start of the support period, and w _LCw = 0 is set for a predetermined time (w _LCw → 0).
Here, it is desirable that the change of w _LCw in these transitions is smooth so as not to give a discontinuous torque command.
The no-leg support period refers to a state where both feet are not in contact with the ground (such as a jump).

[Auto balance control in walking motion]
Next, the autobalance control during biped walking is defined as the following equations (20) to (22) with reference to the autobalance control law (6) in the both-leg support period.

但し、変数を次のように定義する。

However, variables are defined as follows.

  Equations (21) and (22) represent the third term on the right side of equation (20) with respect to legs a and b. In the formulas (21) and (22), the second term on the right side is an integral term of the floor reaction force torque as in the third term on the right side of the formula (6). As will be described later, in Equations (21) and (22), the first term on the right side is an integral term for maintaining the foot 5 at a predetermined angle, and the third term on the right side is the following phase. This is an offset term that is appropriately determined to ensure the continuity of the torque command.

  Then, as described below, automatic balance control in biped walking is realized by switching weighting factors and offset values. Note that, as in the above formulas (2) and (17), in the auto balance control, it is only necessary to focus on a single leg, unlike the switching of the leg control. The following description focuses on the leg a, but at this time, which phase the leg b is in (out of the following three phases) is not related to the discussion on the autobalance control. Further, the following discussion is similarly established when focusing on the leg b.

  In the bipedal motion, the auto balance control is switched in three phases: the stance phase, the swing phase, and the ground phase. “Standing phase” refers to a state in which the foot 5 is completely in contact with the ground (FIG. 7A). The “free leg phase” refers to a state in which the foot 5 is not completely in contact with the ground (floating) (FIG. 5A). The “grounding phase” refers to a state in which a part of the foot 5 is in contact with the ground after transition from the free leg phase (FIG. 6A), and a part of the foot 5 is transitioned from the stance phase. Is not included in the ground contact with the ground (assuming that the stance phase continues).

[Standing phase]
When the leg a is a standing leg, the weighting coefficient of the equation (21) is set to the following equation (23) so that the ACA performs the auto balance control.

According to the weighting coefficient of the equation (23), the equation (20) performs auto balance control on the ACA.
The offset value c _aACw holds the final value in the transition to the current stance phase. That is, since this offset value _caACw is a constant in the stance phase, it is canceled over time by the integral term of the torque command. The integration start time _ta0 of the integral term is the start time of the current stance phase.

[Swing phase]
When the leg a is a free leg, the ACA does not need to perform autobalance control. Therefore, for example, the weighting coefficient of the equation (21) is set to the following equation (24) so that angle control is applied instead.

According to the weighting factor of the equation (24), in the equation (20), the ACA performs the PID control related to the joint displacement on the ACA of the free leg without performing the auto balance control.
The offset value c _aACw holds the final value in the transition to the current swing phase. That is, since this offset value _caACw is a constant in the swing phase, it is canceled over time by the integral term of the torque command. The start time of the current swing phase is taken as the integration start time _ta0 of the integral term.

In this swing leg phase, when the weighting coefficient of the formula (19) in the leg control is w _LCw ≠ 0, for example, as described above, the start time of the current swing leg phase is _simply set in the leg control. _{Assuming the} start of the transition to the leg support period, w _LCw → 0 can be smoothly set from there with a predetermined time width.

[Standing leg → swing leg phase]
When the leg a transitions from the stance phase to the swing phase (FIG. 5 (B)), it is considered that the stance phase continues if even a part of the foot 5 is in contact with the ground (the above “stance phase”). The transition to the swing phase is made at the moment when the foot part 5 completely leaves the ground. Therefore, this transition has no time width, and all the following changes are made at the moment of the transition.

  In this case, ACA only needs to be switched from auto balance control to PID control (step S61 in FIG. 5B), so the weighting coefficient in equation (21) is switched as in equation (25) below (FIG. 5). (B) Step S60).

However, the control torque command τ _ACw is discontinuous only when the equation (25) is used, when the control law is switched from the stance phase to the swing phase. That is, the value of τ _{aACwIF in} the equation (21) becomes discontinuous at the time of switching (transition instant).

Therefore, continuity is ensured as follows (step S60 in FIG. 5B). That is, in the next swing phase from the stance phase, the value of τ _{aACwIF in} the formula (21) at the moment of transition is set as a new offset value _caACw, and this new offset value is held during the next swing phase. . Further, in the next swing leg phase, the integration start time t _a0 of the integral term in equation (21) is set to the start time of the swing leg phase, so the initial value of the integral term at the start of the swing leg phase is zero. . As a result, _continuity between the initial value of τ _aACwIF in the next swing phase and the final value of τ _aACwIF in the previous stance phase is ensured, and ACA can smoothly shift from auto balance control to PID control.

  Then, in the swing leg phase, the ground reaction state of the foot 5 is detected by the floor reaction force sensor 4, and the transition to the ground phase is made if even a part of the sole is grounded. Is continuously performed (step S62 in FIG. 5B).

[Swing phase → Ground phase]
When the leg a transitions from the free leg phase to the ground phase (FIG. 6B), the leg a transitions to the ground phase at the moment when a part of the foot 5 is grounded. However, as described below, the change in this transition is not completed instantaneously but has a time width.

That is, the weighting coefficient and the offset value of the equation (21) are changed with the time width as the following equation (26) so that the _autobalance control torque command τ _ACw changes smoothly (as shown in the following equation (26)). Step S70).

That is, by _setting the value of τ _aACwIF in equation (21) to 0, the _autobalance control rule (20) is switched to PD control (first and second terms on the right side) (step S71 in FIG. 6B). . With this PD control, the ACA follows the ground surface like a spring damper means. Then, by appropriately setting the PD control gain as necessary, the ground phase can be easily changed to the stance phase. The integration start time _ta0 of the integral term is assumed to hold the start time of the previous swing phase as it is.

  In the grounding phase, the ground reaction state of the foot 5 is detected by the floor reaction force sensor 4, and if the entire sole is grounded, the transition is made to the stance phase. It is determined whether it is grounded (step S72 in FIG. 6B). Then, if a part of the sole is in contact with the ground, the above PD control is continued, and if not in contact with the ground at all, a transition to the swing leg phase is made (step S73 in FIG. 6B).

[Grounding phase → Standing phase]
When the leg a transitions from the ground phase to the stance phase (FIG. 7B), the leg 5 transitions from the state where only a part of the foot 5 is grounded to the stance phase when the foot 5 is completely grounded. Therefore, this transition has no time width, and all the following changes are made at the moment of the transition.

  In this case, since ACA is switched to auto balance control (step S51 in FIG. 7B), the weighting coefficient in equation (21) is switched as in the following equation (27) (step S50 in FIG. 7B). ).

However, with the formula (27) alone, the _autobalance control torque command τ _ACw becomes discontinuous when switching from the ground phase to the stance phase, so that continuity is ensured as follows (FIG. 7B). Step S50). That is, the value of τ _{aACwIF in} the equation (21) at the moment of transition is set as a new offset value _caACw, and this new offset value is held during the next stance phase. Further, in the next stance phase, since the integration start time _ta0 of the integral term of equation (21) is set to the start time of the stance phase, the initial value of the integral term at the start of the stance phase is zero. As a result, _continuity between the initial value of τ _aACwIF in the next stance phase and the final value of τ _aACwIF in the previous ground phase is ensured, and ACA can smoothly shift to auto balance control.

  In the stance phase, the ground reaction state of the foot 5 is detected by the floor reaction force sensor 4, and if the part of the sole is also grounded, the above auto balance control is continued, and the sole is completely If it does not contact, it will change to a swing phase (step S52 of FIG. 7 (B)).

If the transition from the ground phase to the stance phase is a transition from the single leg support period to the both leg support period in the leg control, the weight of the formula (19) in the leg control at the start of the both leg support period Since it is necessary that the coefficient w _LCw = 1, the change in the weight coefficient in the leg control needs to be completed in the transition from the immediately preceding free leg phase to the ground phase. If the change w _LCw → 1 of the leg control weight coefficient is not completed within the time width of the transition from the free leg phase to the ground phase, the auto balance is completed until w _LCw = 1 is completed. It is necessary to wait for the transition from the ground phase to the stance phase in the control.

[Ground phase → swing leg phase]
When leg a transitions from the ground phase to the free leg phase (FIG. 5 (B)), from the state where only a part of the foot 5 is grounded, to the free leg phase at the moment when the foot 5 completely leaves the ground. Transition. Therefore, this transition has no time width, and all the following changes are made at the moment of the transition.

  In this case, since ACA only needs to be switched to PID control (step S61 in FIG. 5B), the weighting coefficient in equation (21) is switched as in the following equation (28) (in FIG. 5B). Step S60).

However, with the formula (28) alone, the control torque command τ _ACw becomes discontinuous when switching from the ground phase to the free leg phase, so that continuity is ensured as follows (step of FIG. 5B). S60). That is, the value of τ _{aACwIF in} the equation (21) at the moment of transition is set as a new offset value _caACw, and this offset value is held. Further, in the next swing leg phase, the integration start time t _a0 of the integral term in equation (21) is set to the start time of the swing leg phase, so the initial value of the integral term at the start of the swing leg phase is zero. . As a result, the _continuity between the initial value of τ _aACwIF in the next swing phase and the final value of τ _aACwIF in the previous ground phase is ensured, and ACA can smoothly shift to PID control.

  Then, in the swing leg phase, the ground reaction state of the foot 5 is detected by the floor reaction force sensor 4, and the transition to the ground phase is made if even a part of the sole is grounded. Is continuously performed (step S62 in FIG. 5B).

[Foot mechanism]
Next, the mechanism of the foot 5 will be described with reference to FIG. In the present embodiment, a three-point support structure is provided in which three grounding pieces 50 to be erected are disposed at the apexes of the triangle on the sole of the foot 5.
The grounding piece 50 has a shape constricted toward the ground (for example, a conical shape or a hemispherical shape). Then, the lower end portion of each grounding piece 50 is grounded by being grounded to the ground.

Each grounding piece 50 includes a uniaxial force sensor 50b that detects a floor reaction force acting in the normal direction of the virtual sole plane 5a formed by the three grounding points 50a. In this configuration, the floor reaction torque τ _GR around the autobalance control shaft will be described.

As described above, the floor reaction torque τ _GR represents the torque that the foot 5 exerts as a reaction against the ground. This is a value almost equivalent to ZMP. In many biped robots, in order to measure this in a three-dimensional space, a six-axis force / torque sensor is arranged at the sole of each leg. However, in general, the 6-axis force / torque sensor is expensive, fragile and difficult to handle.

However, in the present embodiment, by placing the autobalance control axis on the virtual sole plane, the measurement of the floor reaction torque τ _GR is simplified, and a 6-axis force / torque sensor is not required. The inventor has proposed this foot mechanism as PCT international applications PCT / JP2008 / 055737 and PCT / JP2008 / 057247. The force sensor, the virtual plantar plane, and the autobalance control axis are as shown in FIG. However, variables are defined as follows.

  At this time, the floor reaction torque received by the foot 5 is expressed by the following equation (30).

  In the present embodiment, the force sensor requires only the force in the normal direction of the virtual sole plane, so the measurement value of the force sensor can be written as the following equation (31).

  When the equation (31) is substituted into the equation (30), the measured floor reaction torque becomes the following equation (32).

Of the floor reaction torque τ _GR around the autobalance control axis, the floor reaction torque of the foot i

について、オートバランス制御軸（ロール）まわりのトルクが第１成分、オートバランス制御軸（ピッチ）まわりのトルクが第２成分とすると、本実施形態のオートバランス制御に用いられる床反トルクは次の（３３）式となる。

When the torque around the autobalance control axis (roll) is the first component and the torque around the autobalance control axis (pitch) is the second component, the floor reaction torque used in the autobalance control of this embodiment is (33)

負号は、床から受けるトルクに対するロボットからの反作用であることを表す。

The negative sign represents a reaction from the robot to the torque received from the floor.

According to the foot mechanism of the present embodiment, the floor reaction torque τ _GR required in the present invention can be measured with only three uniaxial force sensors for each foot in a three-dimensional space. Therefore, if the control of the present invention is performed using this floor reaction torque, it is possible to walk on two legs without installing a 6-axis force / torque sensor.

DESCRIPTION OF SYMBOLS 1 Hybrid control apparatus ACA, LCA Joint actuator 3, 6, 7 Joint displacement sensor and joint actuator drive device 4 Floor reaction force sensor 5 Foot part 50 Grounding piece 5a Virtual sole plane

Claims (12)

In a multi-legged walking device that walks with multiple legs,
The leg includes a hybrid control device that simultaneously executes arbitrary leg control and auto balance control for automatically performing balance maintenance;
The hybrid control device statically performs non-interference between the leg control and the autobalance control in accordance with the ground contact state of the foot.   The multi-leg walking according to claim 1, wherein the leg is independently provided with a joint actuator (LCA) on which the leg control acts and a joint actuator (ACA) on which the auto balance control acts. Mobile device.   In the hybrid control device, in the single leg, depending on the grounding state of each foot part, the free leg phase in which the foot part is completely detached, and the foot part is completely grounded There are three phases: a stance phase, and a ground phase as a process of transition from the swing phase to the stance phase, and control applied to the ACA according to the phase transition is appropriate for the auto balance control and the appropriate phase. The multi-legged walking type moving device according to claim 2, wherein switching is performed between the control and the control.   The hybrid control device performs the decoupling of the leg control applied to the LCA when there are two or more legs that are not in the swing leg phase among all the legs, and otherwise. The multi-legged walking type moving device according to claim 3, wherein the presence or absence of the non-interference is appropriately switched.   5. The multi-legged walking type according to claim 4, wherein the hybrid control device performs switching of the non-interference of the leg control in the LCA by appropriately adjusting a weighting factor in a control law. Mobile device.   The hybrid control device performs switching between the auto balance control and the appropriate control in the ACA by appropriately adjusting a weighting factor, an offset value, and an integration start time of an integral term in a control law. The multi-legged walking type moving apparatus according to claim 3.   The sole of the leg includes a sensor for detecting the grounding state, and the leg control and the auto are controlled according to the values of the sensor for detecting the grounding state and the angle sensors installed in the LCA and the ACA. The multi-legged walking type moving device according to any one of claims 1 to 6, wherein the non-interference with balance control is performed. In the control method of a multi-legged walking device that walks with multiple legs,
The leg is a hybrid control method for executing arbitrary leg control and auto balance control for automatically performing balance maintenance,
A hybrid control method for a multi-legged walking type mobile device, wherein the non-interference between the leg control and the auto balance control is performed statically in accordance with the ground contact state of the foot. The leg independently includes a joint actuator (LCA) that performs the leg control and a joint actuator (ACA) that performs the autobalance control,
In the single leg, depending on the ground contact state of each foot, the free leg phase in which the foot is completely grounded, the stance phase in which the foot is completely grounded, and the free leg There are three phases, a ground phase as a process of transition from the leg phase to the stance phase, and the control applied to the ACA is switched between the auto balance control and the appropriate control according to the phase transition. The hybrid control method according to claim 8.   Of all the legs, when there are two or more legs that are not in the swing leg phase, the non-interference of the leg control applied to the LCA is performed, and in other cases, the non-interference is appropriately performed. The hybrid control method according to claim 9, wherein presence / absence of interference is switched.   The hybrid control method according to claim 10, wherein the decoupling of the leg control in the LCA is performed by appropriately adjusting a weighting factor in a control law.   The switch between the auto balance control and the appropriate control in the ACA is performed by appropriately adjusting a weighting factor, an offset value, and an integration start time of an integral term in a control law. Hybrid control method.

Images