JP2007007796A - Walking robot - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To set an offset value for canceling a bias value contained in output of an acceleration sensor fixed to the trunk of a robot including a plurality of links. <P>SOLUTION: A robot body 100 is caused to take a predetermined static attitude. In that state, the output of the acceleration sensor fixed to the trunk and the output of an attitude angle sensor for detecting the attitude angle to the vertical direction of the acceleration sensor are read. The acceleration detecting direction of the acceleration sensor is calculated from the attitude angle. The component of the gravity acceleration vector in that direction is calculated. The component of the calculated gravity acceleration vector is subtracted from the value of the taken acceleration sensor. The value of subtraction is taken as an offset value to the bias value. In controlling the robot, when the output value of the acceleration sensor is taken, the offset value is subtracted from the taken value to be used for control. It is possible to use an accurate acceleration value within the roboto control device. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a control technology for a walking robot having a trunk and a leg link group. More specifically, the present invention relates to a robot control technique that changes the posture using acceleration output from an acceleration sensor fixed to the trunk.

A walking robot (hereinafter simply referred to as a robot) having a trunk and a leg link group has been developed. This robot walks by changing the relative posture of the trunk and leg links.
A walking robot has been developed in which an acceleration sensor is fixed to the trunk, the acceleration generated in the trunk is detected, and the posture of the trunk and the leg link group is changed using the acceleration output from the acceleration sensor. In order to control the position and posture of the trunk to the intended one, the acceleration acting on the trunk must be accurately acquired by the acceleration sensor.

  Patent Document 1 discloses a robot in which an acceleration sensor is fixed to a trunk. In this technique, based on the output of the acceleration sensor fixed to the trunk, it is determined whether the trunk is in a lying state or a longitudinal state.

Japanese Patent Laid-Open No. 11-156765 (paragraph 0039)

  As long as it is sufficient to determine whether the trunk is in a horizontal state or a vertical state, even if there is a slight error in the output value of the acceleration sensor, the control is not affected. However, in order to control the position and posture of the trunk so as to be intended, it is necessary to accurately acquire the direction and magnitude of acceleration acting on the trunk.

The acceleration sensor outputs an electrical quantity such as voltage, current, or capacitance. Therefore, the acceleration sensor requires a relationship for converting the amount of electricity into acceleration. For example, since the output voltage is 1 mV, an acceleration of 10 mm / sec ² is required, and since the output voltage is 2 mV, a conversion relationship of 20 mm / sec ² is required. This conversion relationship is calculated in advance for each acceleration sensor.
It has been found that when an acceleration sensor whose conversion relationship is known in advance is mounted on the robot, an event that deviates from the conversion relationship measured in advance frequently occurs. For example, when an acceleration sensor that has been found to have the above relationship is mounted on a robot, 1.2 mV is output when the acceleration is 10 mm / sec ^{2 and} the acceleration is 20 mm / sec ^2. Events such as 2.2mV are frequent.
As a result of investigating an event that deviates from the conversion relationship that has been clarified in advance, the effect of deviation of the bias value existing in the electric quantity is greater than the deviation of the gain (proportional constant) established between the electric quantity and acceleration. I understand. In the case illustrated above, it has been found that when mounted on a robot, the output voltage of the acceleration sensor is increased by 0.2 mV, and if converted as it is, an acceleration larger by 2 mm / sec ² than the actual acceleration is detected.
As a cause of the deviation, for example, it is conceivable that the drive voltage applied to the acceleration sensor deviates from the voltage when the conversion relationship is measured. Another possible cause is that the AD converter that converts the sensor output into a digital signal is different from the AD converter when the conversion relationship is measured. In any case, a phenomenon in which the bias value existing in the electric quantity output from the acceleration sensor deviates from the value when the conversion relationship is measured often occurs.

Therefore, after the acceleration sensor is mounted on the robot, it is necessary to measure again the conversion relationship for converting the amount of electricity output from the acceleration sensor into acceleration.
A normal acceleration sensor detects an acceleration component generated in a specific direction. Hereinafter, in this specification, the “specific direction” when the acceleration sensor detects the acceleration component is referred to as an acceleration detection direction. In order to measure the conversion relationship after the acceleration sensor is mounted on the robot, it is necessary to accurately specify the acceleration detection direction of the acceleration sensor mounted on the robot. It is also necessary to accurately specify the true magnitude of acceleration that should be originally obtained from the acceleration sensor. That is, in order to measure the conversion relationship, it is necessary to accurately specify both the acceleration detection direction and the true magnitude of the acceleration actually generated in that direction. If both cannot be specified, a conversion relationship cannot be established.
The robot uses data for commanding the posture angle of the trunk, and changes the posture angle of the trunk so as to match the command data. Therefore, the posture angle of the trunk should be equal to the command value. If the posture angle of the trunk is known, the acceleration detection direction of the acceleration sensor fixed to the trunk should be specified.
Conventionally, the acceleration detection direction of the acceleration sensor is calculated from data that commands the posture angle of the trunk. However, it is difficult to increase the rigidity of the leg link group (the rigidity including the angle deviation of each leg joint), and it is inevitable that the posture angle of the trunk deviates from the command value. As long as the acceleration detection direction of the acceleration sensor is calculated from the command value of the posture angle of the trunk, the acceleration detection direction of the acceleration sensor cannot be accurately specified. Furthermore, the true magnitude of the acceleration that should be originally obtained from the acceleration sensor cannot be specified. Therefore, the conversion relation established between the output value of the acceleration sensor and the acceleration cannot be accurately specified.
The present invention proposes a technique that can be corrected to an accurate conversion relationship by measuring again the conversion relationship for converting the amount of electricity output from the acceleration sensor into acceleration after the acceleration sensor is mounted on the robot.

In the conversion relationship correction technique of the present invention, the robot is stopped, and the acceleration caused by the movement of the robot is set to zero. In other words, an environment in which only gravitational acceleration acts is realized, and an environment in which the actual acceleration can be set to a known amount is prepared.
In the present invention, a posture angle sensor is used to directly detect a posture angle formed by the acceleration detection direction of the acceleration sensor with respect to the vertical direction.
As a result, the acceleration detection direction can be specified. Furthermore, a gravitational acceleration of 9.8 m / sec ² acts in the vertical direction. Therefore, the magnitude of the true acceleration acting in the acceleration detection direction of the acceleration sensor becomes the acceleration selection direction component of the gravitational acceleration. The true magnitude of the acceleration to be obtained from the acceleration sensor is also obtained as the acceleration selection direction component of the gravitational acceleration. That is, by directly detecting the posture angle that the acceleration detection direction of the acceleration sensor makes with respect to the vertical direction, both the acceleration detection direction and the true magnitude of the acceleration actually generated in that direction can be accurately specified. Can do.
Even if the acceleration detection direction is known, the true magnitude of the acceleration generated in the acceleration detection direction is known, and the output value of the acceleration sensor at that time is already known, the amount of electricity output by the acceleration sensor can It is not possible to specify the conversion relationship for converting to. This is because it cannot be determined whether the bias value existing in the electric quantity has changed or whether the gain established between the electric quantity and the acceleration has changed.
In the present invention, it is not actually necessary to consider the fluctuation of the gain, and the knowledge that the conversion relationship can be corrected to a sufficiently accurate practical value by taking the fluctuation of the bias value into consideration is utilized.

The present invention can be embodied in a walking robot that includes a trunk and a leg link group and detects the acceleration generated in the trunk and changes the posture of the trunk and the leg link group.
The walking robot of the present invention is fixed to the trunk and is directly or indirectly fixed to the acceleration sensor for detecting an acceleration component generated in a specific direction (hereinafter referred to as acceleration detection direction). At the same time, a posture angle sensor that detects the posture angle that the acceleration detection direction makes with respect to the vertical direction, and the amount of electricity that the acceleration sensor outputs and the posture that the posture angle sensor outputs while the robot is stationary in a predetermined stationary posture Means for obtaining an angle; means for converting the amount of electricity obtained by the obtaining means into a known relationship and converting the acceleration; and means for calculating a gravitational acceleration component in an acceleration detection direction from an attitude angle obtained by the obtaining means And an acceleration obtained by subtracting the gravitational acceleration component calculated by the calculation means from the acceleration converted by the conversion means and converting the electric quantity output by the acceleration sensor by the conversion means. And a means for storing the offset value is subtracted from.

In the present invention, the robot is stopped in a predetermined stationary posture, and the acceleration caused by the movement of the robot is zero, and a measurement environment in which only the gravitational acceleration acts is realized.
In the present invention, a posture angle formed by the acceleration detection direction of the acceleration sensor with respect to the vertical direction is detected by using a posture angle sensor fixed directly or indirectly to the acceleration sensor fixed to the trunk of the robot. To do. Note that “indirectly fixed” means that the acceleration sensor and the attitude angle sensor are fixed to the same member, and the positional relationship is known (in advance) before being incorporated into the robot. .
As described above, the magnitude of the acceleration that should be detected by the acceleration sensor can be set to a known amount.
The acceleration sensor of the present invention converts the amount of electricity output from the acceleration sensor into acceleration using a conversion relationship measured in advance. At this time, there is often a difference between the true magnitude of the acceleration detected by the acceleration sensor and the converted acceleration.
In the present invention, it is assumed that the deviation is not because the gain established between the electric quantity output by the acceleration sensor and the acceleration fluctuates, but the bias value applied to the electric quantity fluctuates. In that case, there is no need to re-create the pre-measured conversion relationship itself, and the idea that it is sufficient to increase / decrease the amount due to the fluctuation of the bias value from the acceleration converted using the pre-measured conversion relationship. adopt.
In the present invention, the true acceleration is obtained by subtracting the actual acceleration (gravity acceleration component) from the acceleration converted by a known conversion relationship in which the amount of electricity output from the acceleration sensor is measured in advance, thereby correcting the increase / decrease. The offset value to be corrected is obtained, stored, and used for the subsequent conversion result.
According to the present invention, the amount of electricity output from the acceleration sensor is converted into acceleration according to a conversion relationship that has been measured in advance, and then the acceleration that is accurate in both direction and size is detected by subtracting the offset value. Is possible.

The robot includes a plurality of acceleration sensors. The acceleration sensors include a first acceleration sensor in which an acceleration detection direction is defined in the vertical direction of the trunk and an acceleration detection direction in the longitudinal direction of the trunk. It is preferable that a second acceleration sensor for which is defined is included. With these acceleration sensors, it becomes possible to adjust the posture of the trunk when the walking robot is observed from the side to the intended one.
In this case, it is preferable to obtain each offset value for each acceleration sensor and store each offset value for each acceleration sensor.
Based on the above, it is possible to accurately detect both the acceleration generated in the vertical direction of the robot and the acceleration generated in the front-rear direction.

Similarly, even if the robot includes a first acceleration sensor whose acceleration detection direction is defined in the vertical direction of the trunk and a second acceleration sensor whose acceleration detection direction is defined in the horizontal direction of the trunk. Good. With these acceleration sensors, it is possible to adjust the trunk posture when the walking robot is observed from the front.
Also in this case, it is preferable to obtain each offset value for each acceleration sensor and store each offset value for each acceleration sensor.
Based on the above, it is possible to accurately detect both the acceleration generated in the vertical direction and the acceleration generated in the horizontal direction of the robot.

Of course, the robot has a first acceleration sensor whose acceleration detection direction is defined in the vertical direction of the trunk, a second acceleration sensor whose acceleration detection direction is defined in the longitudinal direction of the trunk, and the horizontal direction of the trunk. May include a third acceleration sensor in which an acceleration detection direction is determined. In this case, even if the walking robot is observed from any direction, it is possible to adjust the trunk posture to the intended one.
Also in this case, it is preferable to obtain each offset value for each acceleration sensor and store each offset value for each acceleration sensor.
According to the above, it is possible to accurately detect any acceleration generated in the vertical direction of the robot, acceleration generated in the front / rear direction, and acceleration generated in the left / right direction.

What the acceleration sensor really wants to detect is the motion of the robot, and the acceleration caused by the motion. For that purpose, during the operation of the robot, means for acquiring the electrical quantity output by the acceleration sensor and the attitude angle output by the attitude angle sensor, and the gravitational acceleration in the acceleration detection direction at the acquisition timing from the attitude angle acquired by the acquisition means Means for calculating a component, and means for subtracting the gravitational acceleration component in the acceleration detection direction at the acquisition timing and the offset value from the acceleration obtained by converting the amount of electricity acquired by the acquisition unit according to the known relationship. Just do it.
In the above case, accurate acceleration is detected to subtract the offset value, and the influence of gravitational acceleration can be eliminated to subtract the gravitational acceleration component in the acceleration detection direction. Only the acceleration caused by the movement of the robot can be accurately detected.

When the above technique is used, it is possible to employ a control logic that does not take gravity acceleration into account when controlling the robot using the output of the acceleration sensor. The control logic can be simplified.
Particularly in the robot compliance control, when the robot is about to lose balance, control is performed to correct the position of the trunk so that the robot recovers the balance. At this time, if an acceleration excluding the gravitational acceleration component can be detected, a control logic that considers only the acceleration caused by the operation of the robot may be constructed. Compliance control logic can be simplified.

The acceleration sensor and the attitude angle sensor are preferably configured as an integrated unit. In this case, it is preferable that the unit includes an acceleration sensor and an angular velocity sensor, and includes means for calculating a posture angle that the acceleration detection direction forms with respect to the vertical direction based on the output values thereof.
Since the acceleration sensor and the posture angle sensor are configured in one unit, the posture angle sensor can output an accurate value of the posture angle that the acceleration detection direction forms with respect to the vertical direction. Furthermore, by calculating the output values of the acceleration sensor and angular velocity sensor built in the sensor unit to obtain the posture angle inside the unit, calibration of the posture angle sensor and acceleration sensor are fixed before fixing to the trunk. Calibration can be performed accurately at the same time.

  According to the present invention, in the state where the acceleration sensor is fixed to the trunk of the robot, the offset value necessary for canceling the error due to the bias value, which is added when the output value of the acceleration sensor is converted into the acceleration, is accurately calculated. Can be set.

The main features of the examples are listed.
(First Mode) When the robot is held in a predetermined stationary posture, it is preferable to turn off the original acceleration feedback control. This is because the robot can be held in a predetermined stationary posture regardless of the output value of the acceleration sensor. When acceleration feedback control is performed, the robot may vibrate slightly in some cases. By turning off the feedback circuit, it is possible to prevent an unnecessary acceleration component from being added to the acceleration sensor.
(Third embodiment) The predetermined stationary posture is preferably a stationary posture (so-called upright posture) in which the vertical direction of the trunk is perpendicular to the walking surface. By setting the offset value in a posture in which the trunk is perpendicular to the walking surface, the value obtained by subtracting the offset value from the acceleration value acquired from the acceleration sensor is the most accurate posture at the time of offset value setting. That is, it becomes the time of an upright posture. When giving a posture that causes the trunk to be perpendicular to the walking surface as gait data during acceleration feedback control, the actual posture of the robot can be controlled to be very close to the gait data. It becomes.
(3rd form) It is also preferable to fix an inclination-angle sensor to an acceleration sensor as a detection means to detect the attitude | position angle of an acceleration sensor.

Embodiments will be described in detail with reference to the drawings.
First, FIG. 7 illustrates the posture change of the robot 100 due to the presence of a bias value when the robot control apparatus takes in the output of the acceleration sensor.
FIG. 7A is a side view when the robot 100 takes an upright posture. Reference numeral 80 denotes a trunk. Reference numeral 110 denotes a leg link group. The acceleration sensor 30 is fixed at the reference point P on the trunk 80. The acceleration detection direction of the acceleration sensor is the X-axis direction shown in FIG.

  Now, the robot 100 is compliance-controlled in the X-axis direction. That is, the reference point P of the trunk 80 of the robot 100 is connected to the virtual wall 200 by the virtual spring 202 and the virtual damper 204. The virtual wall 200, the virtual spring 202, and the virtual damper 204 are only virtual existences and are constructed in the control logic of the robot 100. When an external force is applied to the trunk 80, the trunk 80 vibrates in the X-axis direction by the action of the virtual spring 202 and the virtual damper 204 (however, the trunk 80 does not vibrate if the damping force of the damper 204 is large). Then, due to the damping force of the virtual damper 204, the trunk 80 stops at a position where the restoring force of the virtual spring 204 balances with the external force after a predetermined time. As described above, the trunk 80 is controlled as if it is operating while being connected by the virtual wall 200, the virtual spring 202, and the virtual damper 204.

  FIG. 8A shows the posture of the robot 100 in a state where no external force is applied. The natural length of the virtual spring 202 is L. Therefore, the trunk 80 is controlled to be in a stationary state on the assumption that no force is received from the virtual spring 202.

Here, if the robot 100 is stationary, the output from the acceleration sensor 30 should be zero. However, if the offset value deviates from the known offset value, the acceleration in the X-axis direction is not zero due to the presence of the deviated bias value when the control device of the robot 100 takes in the output from the acceleration sensor 30. It will be converted to dGX. The control device of the robot 100 is controlled assuming that the acceleration dGX is acting in the X-axis direction of the trunk link 80. Now, let M be the mass of the trunk 80. In the control device, it is determined that an external force of M × dGX is acting on the trunk 80 due to the acceleration dGX. This external force (M × dGX) is indicated by reference numeral 210 in FIG. Therefore, the compliance control logic in the control device controls the position of the trunk 80 in the X-axis direction so that the external force and the restoring force of the virtual spring 202 are balanced. Specifically, when the spring constant of the virtual spring 202 is K, the length dL is determined so that the magnitude of the restoring force (K × dL) is balanced with the magnitude of the external force (M × dGX). Then, the trunk 80 is moved in the X-axis direction by the length dL. The posture of the robot 100 at this time is shown in FIG. The trunk 80 is controlled to a position moved by dL in the X-axis direction. In order to move the trunk 80 by dL in the X-axis direction, the angles of the joints are controlled so that the posture of the leg link group 110b shown in FIG.
At this time, in the compliance control logic, an external force (M × dGX) (indicated by an arrow 210 in FIG. 7) and a restoring force (K × dL) of a virtual spring 202 (indicated by an arrow 220 in FIG. 7B) are generated. It is judged that it is balanced. However, the acceleration dGX is not an acceleration actually acting on the trunk 80 but is generated due to the presence of a bias value that is shifted when the output of the acceleration sensor 30 is taken. As a result, the posture is shifted although no external force is applied to the trunk 80. In particular, if the posture of the robot 100 deviates when it is stationary, it can be clearly recognized. The existence of a bias value that is deviated greatly affects the posture control of the robot.
If the offset value for canceling the shifted bias value is accurately obtained, the robot posture shift as in the above example does not occur. It is important to obtain an offset value that cancels the bias value.

In the posture shown in FIG. 7B, it seems that the robot may fall forward. FIG. 7B is an exaggerated illustration of the change in the posture of the robot in order to explain the deviation of the bias value, and there is little possibility that the robot will actually fall. Further, the virtual wall 200, the virtual spring 202, and the virtual damper 204 are not actual but are computational in the compliance control logic, but are drawn with solid lines in FIG. 7 for easy understanding.
Further, in order for the trunk 80 to move forward in the X-axis direction as shown in FIG. 7B, the posture of the leg link group changes as indicated by reference numeral 110b in FIG. 7B. At that time, the trunk 80 is shifted vertically downward from the geometric constraint condition of the leg link group 110, but in FIG.

Next, the control of the robot according to the embodiment of the present invention will be described with reference to FIGS.
FIG. 1 is a block diagram of a robot control apparatus 10 and a robot body 100 according to the present embodiment. A combination of the robot controller 10 and the robot body 100 is a “robot”. The robot control device 10 includes an acceleration calculation unit 22 to which gait data 20 (temporal data of target positions of reference points provided at various locations on the robot body 100) for instructing the robot is input. The acceleration calculation unit 22 calculates target accelerations at reference points at various locations of the robot from the gait data 20. In the subtracting unit 23, the acceleration at the reference point of the robot main body 100 obtained from the acceleration sensor 30 installed in the robot main body 100 is subtracted from the target acceleration of each reference point of the robot main body 100 calculated by the acceleration calculating unit 22. . The acceleration at the reference point of the robot body 100 subtracted by the subtracting unit 23 is an acceleration obtained by subtracting the offset value from the acceleration acquired from the acceleration sensor 30 and further removing the gravity component. The removal of the offset value and the gravity component will be described later.
The robot control unit 10 also includes a feedback control unit 24 to which the acceleration calculated by the subtraction unit 23 is input. In the feedback control unit 24, the compliance control described when the posture change of the robot 100 is illustrated in FIG. 7 is executed. The feedback control unit 24 outputs a target position for each sampling of each reference point of the robot obtained by the compliance control logic.
Furthermore, a target value conversion unit 26 is provided that converts the output of the feedback control unit 24 into a target joint angle that instructs an actuator 28 (hereinafter referred to as a joint actuator) provided in each joint of the robot. The converted target joint angle instructed to the joint actuator is sent to the actuator 28 of each joint, and the robot body 100 operates.
The robot control device 10 also includes a conversion unit 31 that converts the amount of electricity output from the acceleration sensor 30 fixed to the trunk of the robot body 100 into acceleration. The conversion unit 31 multiplies the amount of electricity output from the acceleration sensor 30 by a gain (proportional constant) and adds a bias value. The gain and bias value at this time are values measured when the acceleration sensor 30 is calibrated before being fixed to the robot.
The robot control device 10 further includes an offset setting unit 34 that sets an acceleration offset value from the acceleration output from the conversion unit 31 and the value of the attitude angle sensor 32, and an offset storage unit 36 that stores the set offset value. Prepare. The stored offset value is subtracted from the trunk acceleration output from the conversion unit 31 in the acceleration feedback control in the subtraction unit 37. By this subtraction processing, an acceleration (generated in the robot 100) is obtained by removing the acceleration corresponding to the shifted bias from the acceleration converted by the conversion unit 31 including the bias value shifted from the bias value at the time of calibration output from the acceleration sensor 30. True acceleration). In order to facilitate acceleration control feedback control, a gravitational acceleration component removing unit 38 is provided that removes the gravitational acceleration component from the true acceleration value generated in the robot obtained by subtracting the offset value (acceleration for the shifted bias).
The switches 40 and 42 are switches for switching the control circuit in accordance with the setting of the offset value and the actual control of the robot.

Next, a description will be given of main components of the robot control apparatus 10.
The gait data 20 describes data of target positions over time of reference points set in various places of the robot. The acceleration calculation unit 22 calculates the target acceleration of the reference point of the robot from the target position data with time.
On the other hand, the acceleration generated in the robot is detected by the acceleration sensor 30. The amount of electricity output from the acceleration sensor 30 is converted into acceleration by the conversion unit 31, and the offset value is subtracted by the subtraction unit 37 (at this time, the value becomes the true acceleration value). From the value obtained by subtracting the offset value by the subtracting unit 37, the gravitational acceleration component removing unit 38 further removes the gravitational acceleration component (this is the true acceleration value generated by the robot operation at this point).
The subtracting unit 23 subtracts the true acceleration generated by the operation of the robot from the target acceleration calculated by the acceleration calculating unit 22 and sends a deviation from the target acceleration to the feedback control unit 24. In addition to the above-described compliance control logic, a control circuit such as a PID control logic is built in the feedback control unit 24. The feedback control unit 24 outputs a target position for each sampling of the reference point of the robot. This output is input to the target value conversion unit 26, converted into a target joint angle for each sampling of each joint actuator, and each joint actuator 28 of the robot body 100 is driven.

The acceleration sensor 30 provided in the robot body 100 is fixed to the trunk. As an example of the acceleration sensor 30, three triaxial acceleration sensors or three independent single axis acceleration sensors for each of the three axes are used so that accelerations in the trunk vertical direction, front-rear direction, and body side direction can be detected. The attitude angle sensor 32 detects the attitude angle of the acceleration sensor 30 with respect to the vertical direction. The attitude angle sensor 32 is fixed directly or indirectly to the acceleration sensor 30. Here, indirectly fixed means that the acceleration sensor and the attitude angle sensor are fixed to the same member, and the positional relationship is known before being incorporated into the robot.
As the inclination angle sensor, an inclination angle sensor that directly detects a posture angle with respect to the vertical direction can be used. In addition, the sensor unit includes an acceleration sensor and an angular velocity sensor, and based on the output values of these two types of sensors, the attitude angle formed by the acceleration detection direction with respect to the vertical direction is calculated and output inside the sensor unit. It may be a thing.

The output value of the acceleration sensor 30 is an electric quantity such as voltage or current. The conversion unit 31 converts this amount of electricity into acceleration. Since the acceleration sensor 30 itself is calibrated before being incorporated into the robot, the conversion value is known. For example, when the output of the acceleration sensor 30 is a voltage value and the output of 1.0 [V] corresponds to 9.8 [m / sec ² ], the conversion constant for the output value from the acceleration sensor 30 is 9.8 [m / sec ² / V. ]. The conversion unit 31 multiplies the amount of electricity output from the acceleration sensor 30 by the known conversion constant described above to convert it into a unit of acceleration and outputs it.

  When setting the offset value of the acceleration sensor 30, the switch 40 is set to the connected state and the switch 42 is set to the disconnected state. The offset setting unit 34 sets an offset value from the output of the acceleration sensor 30 and the output of the attitude angle sensor 32. The set offset value is stored in the offset storage unit 36. The reason why the feedback circuit is disconnected with the switch 42 in a disconnected state is to prevent the robot from vibrating due to the stability problem of the control system.

  After setting the offset value, the switch 40 is disconnected and the switch 42 is connected. This is because it is not necessary to execute the offset setting unit 34 after setting the offset value, and acceleration feedback is required during robot control.

  The gravitational acceleration component removing unit 38 is particularly effective when causing the robot to perform compliance control. In compliance control, when an external force is applied to the robot, control is performed so that the robot apparently moves as a mass / damper system or a spring / mass / damper system by the external force. Therefore, by removing the gravitational acceleration component from the output value of the acceleration sensor 30, a control system in which only the external force applied to the robot is considered in the feedback control unit 24 while ignoring the gravitational acceleration. Therefore, the feedback control logic can be simplified.

  When the robot controller 10 acquires the output of the acceleration sensor 30, an error (bias value) is added to the output value of the sensor. This error is an error that did not appear during sensor calibration. The cause is, for example, due to the difference between the A / D converter provided in the robot controller and the sensor calibration.

Next, each module in the offset setting unit 34 will be described with reference to FIG.
The acceleration detection direction calculation unit 56 calculates the acceleration detection direction of the acceleration sensor 30 based on the attitude angle data 50 input from the attitude angle sensor 32. The attitude angle of the acceleration sensor 30 with respect to the vertical direction is represented by an angle formed between the reference line fixed in a predetermined direction of the acceleration sensor and the vertical direction. If the reference line fixed in the predetermined direction coincides with the acceleration detection direction, the posture angle itself corresponds to the acceleration detection direction of the acceleration sensor 30. If the reference line fixed in the predetermined direction does not coincide with the acceleration detection direction, the acceleration detection direction of the acceleration sensor 30 can be calculated from the positional relationship between the reference line and the acceleration detection direction and the posture angle. Next, the gravitational acceleration vector component calculation unit 58 calculates a component in the acceleration detection direction of the acceleration sensor 30 from the gravitational acceleration. Then, a gravitational acceleration component in the acceleration detection direction with respect to the vertical direction obtained by the acceleration detection direction calculation unit 56 is calculated.
The offset calculation unit 60 receives acceleration data 52 converted from the amount of electricity output from the acceleration sensor 30 by the conversion unit 31. In the offset calculation unit 60, an offset value is set by subtracting the acceleration detection direction component of the gravitational acceleration calculated from the acquired acceleration data 52. The set offset value is stored in the offset storage unit 36.

In the above apparatus, since the posture angle sensor 32 directly detects the posture angle of the acceleration sensor 30 with respect to the vertical direction, the trunk may have any posture.
With reference to the posture angle of the acceleration sensor 30, the acceleration detection direction of the acceleration sensor 30 and the direction of gravitational acceleration with respect to the vertical direction can be obtained. Thereby, it is possible to subtract the acceleration detection direction component of the gravitational acceleration from the value obtained from the output of the acceleration sensor 30. In addition, since the robot holds a predetermined stationary posture, no acceleration acts on the robot other than gravitational acceleration. Accordingly, a value obtained by subtracting the acceleration detection direction component of the gravitational acceleration from a value obtained by converting the output of the acceleration sensor 30 into an acceleration is an error due to a bias value (hereinafter, an error in units of acceleration is referred to as an acquisition error). This acquisition error is set as an offset value, and at the time of robot control, the output of the acceleration sensor is subtracted from the acceleration value obtained by subtracting the offset value from the acceleration value obtained by converting the electric quantity acquired from the acceleration sensor 30 by the conversion unit 31. It is possible to cancel the acquisition error added when importing into the.

  Next, the flow of the offset value setting process will be described with reference to the flowchart of FIG. First, in step S100, a command value (gait data) of the target position is output so as to maintain a predetermined stationary posture with respect to the robot body 100. This command value is represented by a line indicated by reference numeral 20 in FIG. The predetermined stationary posture may be an arbitrary posture, but during the robot control, the trunk is often controlled to be held in a posture perpendicular to the walking surface. For example, a so-called upright posture. Therefore, a stationary posture in which the vertical direction of the trunk is a posture perpendicular to the walking surface is preferable. By setting the offset value in a posture in which the trunk is perpendicular to the walking surface, the value obtained by subtracting the offset value from the acceleration value acquired from the acceleration sensor is the most accurate posture at the time of offset value setting. That is, it becomes the time of an upright posture. When giving a posture that causes the trunk to be perpendicular to the walking surface as gait data during acceleration feedback control, the actual posture of the robot can be controlled to be very close to the gait data. It becomes.

Next, in step S102, the output of the attitude angle sensor 32 and the output value of the acceleration sensor 30 are acquired. At this time, when a high-frequency component such as noise is included in the output from the sensor, it is also preferable to pass through a low-pass filter.
Next, in step S104, the electric quantity acquired from the acceleration sensor 30 is multiplied by a conversion constant for converting the unit of acceleration. Thus, the value output from the acceleration sensor 30 can be handled in the dimension of acceleration.
Next, in step S106, the acceleration detection direction of the acceleration sensor 30 is calculated. This process can be calculated from the positional relationship between the reference line of the acceleration sensor and the acceleration detection direction in the attitude angle data (attitude angle with respect to the vertical direction of the acceleration sensor) acquired in step S102.
Next, in step S108, a component in the acceleration detection direction of the gravitational acceleration is calculated. Since the gravitational acceleration is a vector pointing vertically downward, and the acceleration detection direction is also calculated in step S106, the component of the gravitational acceleration in the acceleration detection direction can be calculated from these.
Next, in step S110, the value calculated in step S108 is subtracted from the acceleration value converted in step S106. The result of subtraction is the offset value. Finally, the offset value calculated in step S112 is stored in the offset storage unit 36. Thus, the offset value of the acceleration sensor 30 is set.

  There is a possibility that a bias (error) is added to the posture angle output from the posture angle sensor 32 when the posture angle is taken into the robot controller. However, if the error factor for acceleration and the error for posture angle added at the time of loading into the control device are the same, it is assumed that the captured posture angle is an accurate value and the error in acceleration is specified. Has a physically beneficial meaning. This is because the dynamic dimension differs between acceleration and attitude angle. The value obtained by integrating the acceleration twice is the position. Therefore, when the acceleration generated in the robot is detected and the detected acceleration is taken into the control device to estimate the movement of the robot, the error added to the acceleration value at the time of taking in the control device is accumulated by the two integrations to obtain the position. It is reflected in. On the other hand, since the attitude angle is not integrated any more, the error added when it is taken into the control device remains as it is. The error accumulated by the two integrations is larger than the error added in the posture angle dimension (that is, the position dimension). In other words, the error added in the position dimension has a smaller effect on the robot motion than the error added in the acceleration dimension. If the value (acceleration) including the error having a large influence is corrected based on the value including the error (attitude angle) including the error having a small influence, the error of the value (acceleration) including the error having a large influence can be reduced.

When there are a plurality of acceleration sensors 30, the process of the flowchart of FIG. 3 is repeated for each acceleration sensor. The acceleration sensor 30 preferably includes a first acceleration sensor that faces in the vertical direction of the trunk and a second acceleration sensor that faces in the longitudinal direction of the trunk. In particular, the second sensor is suitable for detecting the longitudinal acceleration of the robot when the robot walks. This is because it is more advantageous to directly detect the longitudinal acceleration of the robot in order for the robot to walk without losing the longitudinal balance.
Further, with respect to one acceleration sensor 30, the robot holds a plurality of stationary postures and repeats the process of FIG. It is also preferable to average the offset values obtained for a plurality of stationary postures to obtain a final offset value.

Next, a specific example of offset value setting will be described with reference to FIG. FIG. 4 shows only the trunk 80 of the robot main body 100, and other link parts are omitted. The acceleration sensor 30 is attached to the trunk 80 as shown by the dotted line. In order to simplify the description, a method for setting an offset value in a two-dimensional plane will be described. Further, hereinafter, a value after being converted from an electric quantity, which is an output of the acceleration sensor 30, into an acceleration will be described as an output value from the acceleration sensor 30.
The acceleration sensor 30 is attached to the trunk 80 as shown by the dotted line. Here, it is assumed that the center of the acceleration sensor 30 is the origin, and the coordinate system shown in FIG. The X axis faces the trunk front, and the Z axis faces the trunk downward. Here, the gravitational acceleration is represented by the symbol g. Therefore, the direction of the arrow indicated by g is vertically downward. At this time, the attitude angle of the acceleration sensor 30 is an angle indicated by reference numeral 82. This posture angle is acquired from a posture angle sensor. Note that the posture angle of the acceleration sensor 30 is determined only by the angle 82 since it is currently limited to a two-dimensional plane. The acceleration detection direction of the acceleration sensor 30 is the X-axis direction. In this case, the X-axis component of the gravitational acceleration, that is, the acceleration detection direction component, is a value obtained by multiplying the magnitude of the vector g by the sine of the angle 82. An arrow indicated by reference numeral 84 in FIG. 4 is an acceleration detection direction component of gravitational acceleration. At this time, it is assumed that the output value of the acceleration sensor 30 is an arrow indicated by reference numeral 86. Since the robot holds a static posture, no acceleration other than gravitational acceleration is generated in the trunk 80. Accordingly, a value obtained by subtracting the acceleration detection direction component 84 of the gravitational acceleration from the output value 86 of the acceleration sensor 30, that is, the size of the arrow indicated by 88 in FIG. 4 is set as an offset value in units of acceleration with respect to the bias value. Although the above description is limited to a two-dimensional plane, an offset value can be set similarly in a three-dimensional space.

While the offset value is calculated, the robot is in an arbitrary stationary posture. It is also preferable that the robot is stationary in an upright posture as this stationary posture. Robots often have their trunks oriented vertically, such as during stationary periods before and after movement, or during straight forward movements. Therefore, by calculating the offset value in such a posture, the value obtained by subtracting the offset value in the posture often taken during the operation can be made more accurate.
In the block diagram of FIG. 1, the acceleration sensor 30 and the attitude angle sensor 32 are depicted as independent. An example of the attitude angle sensor is an inclination angle sensor. In addition, as a unit combining the acceleration sensor 30 and the attitude angle sensor 32, an acceleration sensor and an angular velocity sensor are housed in one housing, and an attitude angle with respect to the vertical direction is calculated and output from the outputs of the acceleration sensor and the angular velocity sensor. May be.

  Next, the control of the robot after setting the offset value will be described with reference to FIGS. This control performs acceleration feedback control on the robot using a value obtained by removing the offset value and the gravitational acceleration component from the value of the acceleration sensor attached to the robot.

FIG. 5 is a flowchart showing a flow of gravity acceleration component removal in the acceleration feedback loop performed during robot control. The flow of processing will be described with reference to FIG. 1 together with FIG. The following processing is processed by the gravity component removal unit 38 of FIG.
In step S102b, output values of the acceleration sensor 30 and the attitude angle sensor 32 provided in the robot main body 100 are acquired for each sampling period during the robot operation. In step S102b, a process of multiplying the amount of electricity output from the acceleration sensor by a known conversion constant is also performed. Therefore, after step S102b, the output from the acceleration sensor is treated as a physical quantity having a dimension of acceleration. In step S106b, the acceleration detection direction of the acceleration sensor is calculated. In step S108b, a component in the acceleration detection direction of the gravitational acceleration is calculated. In step S110b, the acceleration detection direction component of the gravitational acceleration calculated in step S108b is subtracted from the output of the acceleration sensor 30. In the processing so far, the processing performed at each step for each sampling is the same as the processing shown in the flowchart of FIG. Specifically, the process of step S102b is the same as the process of steps S102 and S104 of FIG. 3, step S106b is the same as step S106, step S108b is the same as step S108, and step S110b is the same as the process of step S110.

Here, the process of subtracting the acceleration detection direction component of the gravitational acceleration from the output value of the acceleration sensor 30 in step S110b is the process of removing the gravitational acceleration component.
Finally, in step S114, the offset value stored in the offset storage unit 36 is subtracted from the acceleration sensor output value after the gravitational acceleration component is removed. The result thus obtained becomes the output of the gravitational acceleration component removing unit 38. Then, as shown in FIG. 1, a feedback loop is formed together with the acceleration target value output from the acceleration calculation unit 22.

Next, the process of removing the gravitational acceleration component will be illustrated using FIG. Similar to FIG. 4, the description will be limited to a two-dimensional plane.
Now, it is assumed that an acceleration indicated by an arrow 98 is generated in the trunk 80 at a predetermined sampling time (predetermined timing). At this time, the output value of the acceleration sensor 30 with the X axis direction as the acceleration detection direction is an arrow indicated by 96. The sensor output value 96 includes a gravitational acceleration g and an acceleration detection direction component of the acceleration vector 98 applied to the trunk 80. At the same time, sensor error (offset value) is also included. Therefore, by subtracting the acceleration detection direction component of the gravitational acceleration g and the offset value from the output value 96 of the acceleration sensor 30, only the acceleration detection direction component of the acceleration applied to the trunk 80 can be extracted. This will be described below using the arrows in FIG.
The posture angle of the trunk 80 is an angle indicated by reference numeral 92. Therefore, the acceleration detection direction component of the gravitational acceleration g is 94. Further, the acceleration detection direction component of the acceleration 98 applied to the trunk 80 is an arrow indicated by 98a. The arrow 98a is shown not translated on the X axis but moved parallel to the X axis for easy understanding. An arrow 98b is drawn on the X axis as a vector having the same size as this 98a. Moreover, the magnitude | size of the arrow 88 is a magnitude | size of the offset value of the acceleration sensor 30 illustrated in FIG. That is, the output 96 of the acceleration sensor 30 is obtained by adding the acceleration detection direction component 94 of the gravitational acceleration g, the offset value 88, and the acceleration detection direction component 98b of the acceleration vector 98 applied to the trunk 80. Therefore, by subtracting the acceleration detection direction component 94 of the gravitational acceleration g and the offset value 88 from the output value 96 of the acceleration sensor 30, the acceleration detection direction component 98b of the acceleration vector 98 applied to the trunk 80 can be obtained.

Thus, by subtracting the offset value for the error caused by the bias value of the electric quantity output from the acceleration sensor 30 and the acceleration detection direction component of the gravitational acceleration, an accurate value of the acceleration applied to the trunk 80 in addition to the gravitational acceleration is obtained. It can be obtained.
By controlling the robot using this acceleration value, the robot can be controlled by a simple acceleration feedback control system with an accurate acceleration value with the bias value canceled.
In the above embodiment, the case where there is one acceleration sensor is illustrated, but the same processing can be applied even when there are a plurality of acceleration sensors.

  In the above embodiment, an attitude angle sensor fixed directly or indirectly to the acceleration sensor is used to specify the acceleration detection direction of the acceleration sensor. Here, “indirectly fixed” means that the acceleration sensor and the attitude angle sensor are fixed to the same member, and the positional relationship is known (in advance) before being incorporated into the robot. . In order to specify the posture of the trunk, the posture of the trunk can also be obtained by so-called forward conversion from the value of the encoder provided in each joint of the leg link group and the shape of each link of the leg. However, the encoder value also includes errors. Furthermore, if forward conversion is performed from the structure of a leg link group in which a plurality of joints are connected in series via a link, errors of the encoders accumulate. Further, the rigidity of the link itself is not considered in the forward conversion. For the above reasons, the accuracy in identifying the acceleration detection direction of the acceleration sensor from the posture angle sensor is more prominent than specifying the posture angle of the trunk and the acceleration detection direction of the acceleration sensor based on the encoder value. High accuracy. By using the attitude angle sensor, the offset value of the acceleration sensor can be obtained with high accuracy.

In the above embodiment, the front-rear direction of the trunk is taken as an example as the acceleration detection direction of the acceleration sensor (corresponding to the “specific direction” recited in the claims). However, the acceleration detection direction is not limited to the longitudinal direction of the trunk. Furthermore, the acceleration detection direction is not limited to one.
The robot includes a plurality of acceleration sensors. The acceleration sensors include a first acceleration sensor in which an acceleration detection direction is defined in the vertical direction of the trunk and an acceleration detection direction in the longitudinal direction of the trunk. May be included. An offset value is obtained and stored for each acceleration sensor. The robot is controlled using the stored offset value. This makes it possible to adjust the trunk posture when the walking robot is observed from the side to the intended posture.
Similarly, even if the robot includes a first acceleration sensor whose acceleration detection direction is defined in the vertical direction of the trunk and a second acceleration sensor whose acceleration detection direction is defined in the horizontal direction of the trunk. Good. The offset values of these acceleration sensors are obtained and stored. By controlling the robot using the stored offset value, it becomes possible to adjust the trunk posture when the walking robot is observed from the front.
Of course, the robot has a first acceleration sensor whose acceleration detection direction is defined in the vertical direction of the trunk, a second acceleration sensor whose acceleration detection direction is defined in the longitudinal direction of the trunk, and the horizontal direction of the trunk. Even if the third acceleration sensor in which the acceleration detection direction is determined is included, the same effect can be obtained.

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.
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 block diagram of a robot controller. It is a block diagram in an offset setting part. It is a flowchart figure of an offset setting process. It is the figure which illustrated the mode of offset setting processing. It is a flowchart figure of a gravity acceleration component removal process. It is a figure which illustrates the mode of a gravity acceleration component removal process. It is a figure which illustrates the change of the posture of a robot when there is an offset value.

Explanation of symbols

10: Robot control device 22: Acceleration calculation unit 23, 37: Subtraction unit 24: Feedback control unit 26: Target value conversion unit 28: Joint actuator 30: Acceleration sensor 31: Conversion unit 32: Attitude angle sensor 34: Offset setting unit 36 : Offset storage unit 38: Gravity acceleration component removal unit 40, 42: Switch 56: Acceleration detection direction calculation unit 58: Gravity acceleration vector component calculation unit 60: Offset calculation unit 100: Robot g, G: Gravity acceleration vector

Claims (4)

A walking robot that includes a trunk and a leg link group, and detects the acceleration generated in the trunk, thereby changing the posture of the trunk and leg link group.
An acceleration sensor that is fixed to the trunk and detects an acceleration component generated in a specific direction (hereinafter referred to as an acceleration detection direction);
An attitude angle sensor that is directly or indirectly fixed to an acceleration sensor and detects an attitude angle that the acceleration detection direction forms with respect to a vertical direction;
Means for obtaining an electrical quantity output by the acceleration sensor and a posture angle output by the posture angle sensor while the robot is stationary in a predetermined stationary posture;
Means for converting the amount of electricity acquired by the acquisition means into a known relationship and converting to acceleration;
Means for calculating a gravitational acceleration component in the acceleration detection direction from the posture angle acquired by the acquisition means;
Means for storing a value obtained by subtracting the gravitational acceleration component calculated by the calculation means from the acceleration converted by the conversion means, as an offset value for subtracting the electric quantity output by the acceleration sensor from the acceleration converted by the conversion means;
A walking robot characterized by comprising:   A plurality of the acceleration sensors are provided, and in these acceleration sensors, a first acceleration sensor in which an acceleration detection direction is defined in the vertical direction of the trunk and an acceleration detection direction in the longitudinal direction of the trunk are defined. The walking robot according to claim 1, wherein second walking sensors are included, and offset values corresponding to these sensors are stored. Means for acquiring an electrical quantity output by the acceleration sensor and a posture angle output by the posture angle sensor during operation of the robot;
Means for calculating the gravitational acceleration component in the acceleration detection direction at the acquisition timing from the posture angle acquired by the acquisition means;
Means for subtracting the gravitational acceleration component in the acceleration detection direction at the acquisition timing and the offset value from the acceleration obtained by converting the amount of electricity acquired by the acquisition unit according to the known relationship;
The walking robot according to claim 1, further comprising: The acceleration sensor and the attitude angle sensor are configured as an integrated unit,
The posture angle sensor is an acceleration sensor built in the unit, an angular velocity sensor built in the unit, and a means for calculating a posture angle formed by the acceleration detection direction with respect to the vertical direction based on output values thereof. The walking robot according to claim 1, further comprising:

Images