EP0995080A1

EP0995080A1 - Method for determining the rotating position of a self-contained mobile unit, and self-contained mobile unit

Info

Publication number: EP0995080A1
Application number: EP98942495A
Authority: EP
Inventors: Rudolf Bauer; Martin Soika; Wolfgang Rencken
Original assignee: Siemens AG
Current assignee: Siemens AG
Priority date: 1997-07-16
Filing date: 1998-07-06
Publication date: 2000-04-26
Also published as: WO1999004225A1; JP2001510890A; DE19730483C2; DE19730483A1

Abstract

The invention concerns a method for determining the rotating position of a self-contained mobile unit, such as a small robot, for domestic or industrial use. Said unit comprises a gyroscope and position sensor for determining first and second rotating position variations. Said variations are compared with one another and if the measurement results are reciprocally confirmed, the gyroscope can be calibrated while the unit is moving. In the event that the measurement results are not confirmed, it can be concluded that there is an error in the odometry measurement, and the gyroscope is momentarily used as a more dependable measuring instrument. The invention therefore discloses a novel method for calibrating a gyroscope while the mobile self-contained unit is moving, using odometry data. The drawbacks of gyroscopic or odometry measurements are obviated, the correct data being used in each case using a threshold value for the comparative result.

Description

description

Method for determining the rotational position of an autonomous mobile unit and autonomous mobile unit.

The invention relates to a method and an arrangement for determining the rotational position of a mobile system while driving, as can be used for example in autonomous mobile robots and other autonomous vehicles.

The general problem with autonomous mobile systems is that they have to get an idea of their surroundings while driving. Different sensors are used for this purpose, which record measurement data from the environment. A particularly favorable method for determining displacement measurement data is odometry, which is known from DE 33 15 422 AI. For example, encoders on the wheel axles are used to measure the wheel rotations. The position of the unit can then be continuously estimated, for example, using the odometry data obtained from these internal sensors. A disadvantage of this form of dead reckoning navigation is that the position estimation based on odometry data is fraught with errors. With these errors, a distinction can be made between systematic and unsystematic errors. Systematic errors are understood to mean, for example, unequal wheel diameters or an imprecisely measured wheel distance. However, these systematic errors can be completely eliminated by skillful calibration. Then only the unsystematic errors contribute to falsifying the estimated position. Such errors are caused, for example, by the wheels slipping on the floor or by running over unknown objects lying on the floor, such as cables, door sills or tile joints. From the different ways that the different wheels of the auto- nomen mobile unit can be measured, a change in rotational position can then be determined based on the odometry data.

Another possibility for determining a rotational change in position is the use of a gyroscope, which is arranged on the autonomous mobile unit _¬ actual means of this gyroscope, accelerometer or another can change the rotary position of the autonomous mobile unit be relatively determined. The absolute rotational position can then be determined by integrating the raw data once or twice. The disadvantage of these systems is that the sensor signals have to be integrated in order to obtain the desired measured variable. Every smallest constant error in the sensor data leads to a drift of the measured variable. Since this error increases over time, these systems can only be used alone for a short period of time, for example approx. A few minutes. An orientation break must then be carried out in order to eliminate the offset of the system, which is also caused by temperature fluctuations and other environmental influences, and to calibrate the system. From the prior art of "Fang PD, Hung J. C: Gyro-compassing on a Moving Land Vehicle, Proceedings of the 15th Southeastern Symposium on System Theory, Mar. 28-29, 1988, Huntsville, AL, USA" is a Method known, which uses a gyroscope on an autonomous mobile unit and calibrates this using a Cayman filter. There, however, in addition to quasi-static parameters such as the offset and gain factors, dynamic variables such as vehicle orientation are estimated for calibration of the gyroscope.

It is also known from the prior art DE 38 31 166 AI to briefly use a further orientation system for correcting an odometry system used to navigate an autonomous mobile unit.

It is known from the prior art DE 39 10 945 AI that two redundant ones are used for the navigation of autonomous mobile units To use dead reckoning systems, switching between the two systems in a controlled manner. DE 39 10 945 AI mentions the combination of odometer and gyro system as a redundant system. The switching between the two systems is controlled in accordance with DE 39 10 945 AI in that both the actual position of the autonomous mobile unit determined by the odometry system and the actual position of the autonomous mobile unit determined by the gyro system with the target position of the autonomous mobile unit is calculated from a driving course stored in a computer, is compared. Depending on the two comparison results, the system with the smaller deviation of the actual value from the target value is used for navigation of the autonomous mobile unit.

The object on which the invention is based is to specify a further method and a further arrangement for determining the rotational position of an autonomous mobile unit while driving.

This object is achieved for the method according to the features of claims 1-9 and for the autonomous mobile unit according to the features of claim 10.

Further developments of the invention result from the dependent claims.

A particular advantage of the method according to the invention is that the change in the rotational position of the autonomous mobile unit is determined by two different measuring means and that in each case that measuring means is used to determine the rotational position that provides the more reliable measurement results at the moment.

In the method according to the invention, the distance measurement can advantageously be carried out by means of odometry, since this method ren used in many autonomous units and inexpensive sensors are available.

According to the method according to the invention, the rotational position can be determined particularly advantageously as the rotational position that was measured by the gyroscope, or as the rotational position that was determined from the odometry data.

Particularly advantageously, the method according to the invention can be used to distinguish, by means of a preferably experimentally determined threshold value, which of the two measuring systems is currently delivering the more reliable results and can use this measuring system to determine the rotational position.

The difference in the amount is formed particularly advantageously to form the comparison result from the two rotational positions, since this can be computationally calculated with little effort and sign effects are eliminated.

According to the method according to the invention, a gyroscope is particularly advantageously calibrated while the autonomous mobile unit is traveling if the odometry data appear reliable at the moment, which can be determined on the basis of the threshold value.

According to the method according to the invention, a Cayman filter is advantageously used to estimate the offset of the gyroscope, since this filter is used in a variety of ways and known methods exist for implementing treasures for time-dependent variables.

According to the method according to the invention, an error payer is implemented particularly advantageously, which is paid up as a function of the threshold values and which is used to detect a sensor error after a predetermined error limit has been exceeded. An autonomous mobile unit, which has measuring means for determining displacement measurement data and a gyroscope, the rotational position of which is determined by the method according to the invention, is particularly advantageous since it is thus possible for the first time to calibrate a gyroscope while driving.

The invention is explained in more detail below with reference to figures.

Figure 1 shows an example of a unit in a reference coordinate system. Figure 2 illustrates an unsystematic error in the odometry measurement. Figure 3 shows an example of a Cayman filter. FIG. 4 shows a block diagram as an example of a method according to the invention. FIG. 5 shows a method for determining the sensor error on the basis of the threshold value and an error counter. Figure 1 shows the movement model of an autonomous vehicle

AE. The configuration of the mobile system in the form of position and rotational position in a fixed reference system is shown as an example. This position information is to be determined continuously, for example, relative to the coordinate system while driving. This configuration

z _k =, y _k> θ _k ] ^T E tt ² x [0, 2π] (i; for a periodically performed scanning step k consists of the (x, y) position and the orientation of the autonomous mobile unit in a global Coordinate system K: (x, y) The determination of the rotational position of the vehicle is particularly critical, since an orientation error - albeit a small one - causes an ever larger translation error while the vehicle is in motion. According to the invention, this orientation error is to be reduced. results from odometry measurement with sensors OMl and OM2 on wheels Rl and R2 and combined by a gyroscope Gy.

For configuration determination and motion modeling, for example, an autonomous system with two driven rear wheels R1 and R2 is assumed. Kinematically, this form of drive corresponds to the widespread tricycle kematics found in cars and many mobile robots. The movement model of such an autonomous vehicle is shown in a global coordinate system K: (x, y).

The following equations can be used to estimate small changes in position and rotational position,

u + u rriicc ++ ιι l lc - l

Xk + 1 cos (θ _k )

Uk + i - θk ⁺

2W _S where W ₃ denotes half the track width and U _r and U _j the movement changes on the two wheels. The index k denotes the discrete state. For the relative position determination, preferably no sensor information about the external environment is used for position estimation. Odometry has the following advantages:

l.kem the drift offset to be determined

2. simple incremental encoder

3. very robust against environmental influences such as temperature, air pressure and air humidity 4. allows the determination of the coordinates and the orientation

Your disadvantages are:

1.Slip of the wheel remains unmodified „„, „„ „,, - PCT / DE98 / 01866 O 99/04225

7 2. Spinning the wheels due to running over thresholds or cables or tile joints leads to an incorrect estimate of the rotational position

Using an example, the falsification of the displacement measurement results is explained as one of these disadvantages in FIG

As an example, FIG. 2 shows how a wheel travels over a small obstacle H. This is to show the effect of an unsystematic error on the position estimate. In FIG. 2, a wheel R with a radius r is about to pass an unknown obstacle H with a height h. For example, it is idealized that the wheel R moves exactly and without slipping over the contact point C. The wheel center M rotates over the contact point C until it is exactly above this in the position M '. During this movement, the wheel encoders preferably measure the wheel rotation φ, which is interpreted as the travel distance d _mßß . The real horizontal movement of the wheel in the direction Bew is only d _hor -

If only one wheel of the unit drives over this obstacle, a path change Δd of

Δd = 2- | d _fwΛ -d, (3; determine. The factor 2 results from the fact that the wheel has to drive up to the obstacle and then down again.

With a wheel radius r and an obstacle height h, the path difference can be r - h = r • cos (φ) if h <rd _hor = r - sin (φ) (4) ^d meas = ^{r '} Φ be determined.

If only one wheel of a robot with three-wheel kinematics drives over the obstacle, the change in path Δd of a wheel causes an angle error Aθ of when determining the position

Δd

Δθ =

2W _S with half a track width of W ₃ .

A calculation example with the dimensions of an experimental robot should illustrate the effects. r = 60 mm h ~ \ Q mm 2W _S = 300 mm

(6) => Δd = 3.94 mm => AΘ = 0.013 [rad \ = 0.15 ° => position error after 10 m straight travel: 130 mm If a wheel with radius r = 6 cm over a threshold of h = l cm, there is an angle error of Δθ = 0.75 °. This angular error causes e.g. on a subsequent straight ahead drive of dist = 10 m with

ΔT = dist • sin (Δθ) a translational error ΔT of approximately 13 cm. The angle error is actually the decisive error, since the resulting position error grows indefinitely when you continue driving.

To improve the position and rotational position determination of an autonomous mobile unit, a second measuring means, for example in the form of a gyroscope or an accelerometer, which does not have such errors, can therefore be used expediently. This allows the rotation of a robot to be determined relatively. The rotational position is determined by integrating the measured raw data once or twice. The disadvantage of these systems is that the Sensor signals must be integrated in order to obtain the desired measurement size. Every smallest constant error in the sensor data leads to a drift of the required measurement quantity. The error in determining the rotational position thus grows over time without limits. If these systems are only used in general, they can therefore only guarantee an accurate determination of the orientation for a few minutes.

For example, the gyroscope delivers a voltage _oul ; / which is proportional to the angular velocity ω.

If the voltage is sampled periodically every T _G , the rotational position can be with

Determine θ _{k + 1} = θ _{k +} κ (U _{out ^} - offset _k ) τ _G (9) by Euler-Forward integration, where K includes a gyroscope-dependent proportionality factor

K = 22.2 mV / deg / sec dθ).

A major problem with inexpensive gyroscopes, which appear particularly suitable for consumer applications such as household robots or mailing lists, is the offset, which mainly depends on the temperature, but also, for example, on the air humidity, the switching time and other factors. With commercially available gyroscopes, the offset can change with a temperature change of 3 K u approx. 100 mV. That means a continuous error of

(100 / 22.2) deg / sec «4 deg / sec, UD which is only caused by the offset shift. According to the results of experimental investigations, the offset should therefore preferably be recalibrated every minute when the gyroscope is used for orientation. Then:

Offset = U _out . (12) If the robot is to carry out a transport job, for example, a robot downtime per minute is unacceptable. The invention therefore presents a method which allows the offset to be calibrated / readjusted during the robot run.

The advantage of the gyroscope is its slip-dependent determination of the rotational position, its disadvantages are:

The speed of rotation is determined, to determine the position of rotation it must be integrated. It is therefore very much dependent on an exact determination of the zero point. 2. The zero point is exposed to a very strong drift, which is also dependent on changing environmental influences. The zero point must therefore be constantly recalibrated. Previous solution: Every minute the robot comes to a standstill for 2 seconds. 3. Only the rotational position can be determined

The invention takes advantage of the advantages of the odo-etπe and the measurement with gyroscopes, without having to accept their disadvantages

It guarantees the combined use of odometry and gyroscope to determine the rotational position of a mobile system without the need for orientation breaks to adjust the offset on the gyroscope. The aim at the time of sampling (k + 1) is to determine the change in the ΔΘ _{OdometπeGyr0sk0p} and thus the

Rotational position of the mobile unit using

θ + l = θk + Δθ odometer / gyroscope < ^{1 3} >. The following procedure is preferred Determination of the angle changes using odometry (Index Odo) and gyroscope (Index Gyro) at the time (k + 1)

U _r -Uι

Δθ odo

Odometry: 2W _S = f (u _r , u,)

ΔΘ _Gyro = K (U _out -OfFset _k ) T _c

Gyroscope: = f (U _0Ut , offset _k )

b) Comparison of the two angle changes according to the following scheme

99/04225

12th

The two sensors deliver different angular velocities. If it is assumed that both sensors are working properly, it is much more plausible that the correct angular velocity is provided by the gyroscope. The wheels can spin here or drive over a bump, for example.

It should preferably be noted that the gyroscope offset is not too old, i.e. definitely not older than about 1 minute. Otherwise, the deviation of the sensors could result from a temperature change.

The gyroscope is preferably not recalibrated if both sensors deliver different angular velocities.

c) Calibration of the gyroscope offset

If the difference between the angular velocity values of the gyroscope and the odometry is smaller than a threshold value determined experimentally, for example, it can be assumed that neither odometry measurement nor gyroscope measurement are faulty. In addition, it is very unlikely that the vehicle would have driven a wheel over an obstacle, as shown in FIG. 2, and thus the odometry was adversely affected. Therefore, odometry is used to determine the rotational position.

The measured value of the gyroscope and the odometry values are preferably used according to the invention additionally for calibration / re-calibration of the gyroscope offset while the mobile unit is traveling. FIG. 3 shows an example of its estimation using a filter. As FIG. 3 shows, a Kaiman filter K can be used, for example, to determine the gyroscope offset. Since the gyroscope offset to be estimated changes very strongly over time, an iterative or recursive estimator should be used for calibration. Since the measurement data is also very noisy (eg due to A / D conversion), the estimator should preferably be able to take system uncertainties and measurement noise into account.

The use of a Cayman filter K is favorable for such cases. The use of this filter is then explained further.

However, it should be expressly mentioned here that in principle other filters would also be suitable for the method according to the invention, possibly even fuzzy rules or neural networks. Calibration with a Cayman filter is already reported in the prior art. In contrast to this, where, in addition to quasi-static parameters such as offset, gain factors, dynamic variables such as vehicle orientation are also estimated, only the sensor parameters of the gyroscope such as the offset and gain factors, in particular the offset, should be optimized according to the invention, since this mainly affects the determination of the rotational position. Here, too, the determination of the vehicle orientation could be integrated into the filter. However, this would make the case distinction below difficult.

The parameters characterizing the gyroscope, in particular, are the states for the Kaiman filter

• the zero point → Θ _Q ^. • the slope of the characteristic for positive θ → G ~

• The slope of the characteristic curve is used for negative θ -> G ^~ .

With the state vector 99/04225

14

AGyro ^σ Offset x = G ⁺ (14) G- the system equations result in

1 0 0

^X k + 1 - 0 1 0 x _k - Ax _k - Ex _k - x _k (15) 0 0 1

with the system matrix A (E = unit matrix).

Depending on whether the angular velocity measurement value M supplied by the gyroscope is greater or smaller than the current zero point, the angular velocity is increased via the measurement equation of the Cayman filter

A Gyro ^σ offset

Z _k = θ? "] = [- 1 M 0]. G ^' = H, Xt (16) G ^" or.

A Gyro ^u Offset

Z _k -Θ? "] = [- 1 0 M] - G ⁺ = H, x,: i7; G ^" estimated. This is then preferably compared with the angular velocity θ _k determined from the odometry data and the difference is used for the estimation of the gyroscope parameters. A Kaiman filter for estimating the gyroscope parameters can be constructed from these equations and the associated covariance matrices.

With this method it is now possible to calibrate the gyroscope in such a way that its orientation estimate normally matches that of the odometry without interference. In the case of a detected malfunction of the odometry, the vehicle orientation can then be reliably determined with the help of the gyroscope. d) Test: sensor error

In the method described so far, correctly functioning sensors were always assumed. Of course, it can happen at any time that a sensor fails.

For this reason, in the method according to the invention, the odometry and the gyroscope are preferably monitored for the agreement of their data.

FIG. 5 shows an example of a flow chart for determining sensor errors using an error counter. If the sensor measurement results over F_max consecutive cycles do not match, then there is a plausibility error in either the gyroscope or in odometry. In this case, the robot should preferably cancel its mission and stop.

In summary, the following advantages of the method according to the invention result:

• Combines advantages of odometry and gyroscope

• Enables "online" calibration of the gyroscope while driving

• Makes sensible use of a cheap gyroscope (with an unstable offset) possible. • Allows detection of a sensor failure

The relationships and the relevant variables for the OV method according to the invention are illustrated in FIG. 4 in a block diagram

Claims

claims

1. Method for determining the rotational position of an autonomous mobile unit with the following features: a) by means of distance measurement at least at a first location and at a second location of the unit, depending on a first movement of the unit, a first travel distance traveled from the first location and one from the second The second distance traveled in the location is measured in the form of first and second distance measurement data and a first change in the rotational position of the unit is determined from the known position of the locations on the unit and a known angle function. b) by means of a gyroscope connected to the unit, the rotational position change in the form of rotational position measurement data is measured as a function of the first movement of the unit and is determined as the second rotational position change of the unit; c) the first and second changes in rotational position are compared with one another and a first comparison result is formed; d) the rotational position of the autonomous mobile unit is determined depending on the comparison result.

2. The method according to claim 1, in which distance measurement is carried out by means of odometry

3. The method of claim 1 or 2, wherein the rotational position is set as the first rotational position.

4. The method according to claim 1 or 2, wherein the rotational position is determined as the second rotational position.

5. The method according to any one of the preceding claims, in which the comparison result is compared with a threshold value and the rotational position is determined depending on whether the threshold value is greater, less than or equal to the comparison result.

6. The method as claimed in one of the preceding claims, in which the first change in rotational position or the first and second displacement measurement data, and the second change in rotational position or the rotational position measurement data are converted into numbers, and the result of the comparison is the amount of the difference between the first and second rotational position changes becomes.

7. The method according to any one of claims 2 to 6 without 4, in which the gyroscope has an offset and this is calibrated by means of the first change in rotational position while the unit is traveling.

8. The method according to claim 7, in which a Kalman filter is used for the calibration as an iterative estimator of the offset, to which the first and the second change in rotational position are supplied.

9. The method as claimed in one of claims 5 to 8, in which an error count is counted as a function of the exceeding or falling below the threshold value and, if this exceeds or falls below a specified error limit depending on the counting method, a measurement error is detected.

10. Autonomous mobile unit with determination of the rotational position, a) which means for measuring a first movement of the path

Unit at least at a first location and at a second location, which measure a first route covered by the first location and a second route covered by the second location in the form of first and second distance measurement data; b) which has a gyroscope for measuring a second change in rotational position; c) which evaluation means for calculating a first change in rotational position from the first and second displacement measurement data by means of a known angle function; d) which comparison means has for comparing the first and the second change in rotational position; and has the fixing means for fixing the rotational position of the autonomous mobile unit as a function of the comparison result.