GB2146801A

GB2146801A - Control of robots

Info

Publication number: GB2146801A
Application number: GB08423313A
Authority: GB
Inventors: Stephen John Williams
Original assignee: Thorn EMI PLC
Current assignee: Thorn EMI PLC
Priority date: 1983-09-16
Filing date: 1984-09-14
Publication date: 1985-04-24
Also published as: GB8423313D0; GB8324866D0; GB2146801B

Abstract

A robot has a movable arm formed of two pivotally interconnected links 1, 2, whose motion is controlled by a control system which generates input signals defining the demanded position of links 1 and 2. These input signals are applied at lines 3 and 4 respectively to algebraic summing junctions 6 and 8 for combination with signals from a local feedback loop 5 or 7 and from a cross-coupled feedback loop 9 or 10. Each of the feedback loops 5, 7, 9, 10 has an appropriate gain; in certain applications the value of each loop gain can be varied selectively. <IMAGE>

Description

SPECIFICATION Control of robots This invention relates to the control of robots having at least two relatively pivotable links.

The dynamics of modern multi-linked industrial robots are extremely complex. Generally there are large dynamic (and static) interactions between one link and another. Conventional industrial robot control systems are usually designed by considering one joint at a time, i.e. each degree of freedom will commonly have its own independent servo system with local feedback. The local feedback might comprise proportional and velocity feedback for an electric DC driven robot, or proportional, velocity and acceleration feedback in a hydraulically powered robot. This approach fails to take into account the way in which the movement of one link can be influenced by the movements of other links. If dynamic interations between links are not taken into account the robot will be slower and less accurate in its movement, for example by overshooting a commanded position.

In an attempt to improve the control of robots a recent proposal involves predicting the way in which the movement of one link affects another link and feeding forward the appropriate dynamic terms. This is sometimes known as "on-line dynamic control" and whilst this improves performance to some degree it is not as effective as the present invention which relies on the feedback technique and therefore does not require the dynamic terms to be modelled on-line.

In a publication entitled "R.A.I.R.O. Automatique/Systems Analysis and Control" (Vol.

13, 1979 p189-201), an article by W. Khalil et al entitled "Commande Dynamique De Robots" describes a control system which incorporates feed-forward correction and a closed loop with an adaptive compensator; consideration is given to the calculation of position errors. Another article of interest is entitled "Manipulator control using the configuration space method" by M. H. Raibert and B. K. P.

Horn ("The Industrial Robot", June 1978).

The present invention provides a robot comprising an arm having a plurality of links capable of relative pivotal movement in accordance with input signals defining the demanded positions of the links, a control system to effect control on the movement of the arm, the control system incorporating a crosscoupled feedback loop which has means to modify the input signal to a first link in accordance with information derived from the output of at least one other of said links, and means to use the modified input signal in order to drive the first said link, the control system having means to monitor, over a time period, the difference between actual position of a link and the intended position of the link, and means to effect a compensating movement of the link when the difference exceeds a predetermined amount.

The monitoring means of the control system has means to detect, at a given moment, the difference between the actual position of a link and the intended position of the link, means to repeat the difference-detection operation at regular intervals, means to sum the detected differences, and means to effect a compensating movement of the link when the summation of detected differences exceeds a predetermined value; preferably the control system has means to monitor the position of each of a number of links in the arm.

A robot according to the invention has the cross-coupled feedback loop which takes account of the important dynamic interactions between the two links.

The robot may have at least one additional cross-coupled feedback loop having means to modify the input signal to a link other than the first link in accordance with information derived from the output of any one or more of the links, and means to use the modified input signal in order to drive said other link; preferably each of the links has an associated cross-coupled feedback loop.

The control system may incorporate at least one local feedback loop having means to modify the input signal to a link in accordance with information derived from the output of that link.

It will be appreciated that the invention is applicable to a robot having more than two links, and in the general case of n links there will be n2 cross-coupled feedback loops if the input signal to each link is to be varied in dependence upon the output positions of all of the other links. In practice this might not be necessary because in certain dynamic configurations the influence of one particular link upon another can be minimal.

Each cross-coupled feedback loop has a particular gain or transfer function chosen to provide the appropriate interactive or "weighting" effect on the signal to which it is applied. The gain or transfer function of each cross-coupled feedback loop may be constant but better results are achieved if the gain or transfer functioncan be made to vary. For example, it is advantageous to vary the gain or transfer function in any particular crosscoupled feedback loop in dependence upon the angular positions of the two links which the loop interconnects. The gain or transfer function may also be made to vary selectively in dependence upon the load to which the robot is subjected. The variations in gain can be determined theoretically.

In the preferred embodiment to be described, feedback loops are taken from the outputs of two links and the signals derived therefrom are fed into a common processor which operates in a time multiplexed fashion to alter the torque applied to the first link in accordance with the outputs from both links and to alter the torque applied to the other link in accordance with the outputs of both links. The micro-processor is preferably sufficiently powerful to implement control algorithms achieving this result, and in a particular form of micro-processor the entire control algorithm can be executed in between one and two milliseconds for a two link system.

The present invention also provides a con troi system to effect control on the movement of an arm in a robot as defined above.

The present invention also provides a method of controlling the movement of a plurality of robot links movable in dependence upon input signals defining the demanded angular positions of the links, the method comprising modifying the input signal to a first link in dependence upon information derived through a cross-coupled feedback loop from at least one other said links, using the modified input signal to drive the first said link, monitoring, over a time period, the difference between the actual position of a link and the intended position of the link, and effecting a compensating movement of the link when the difference exceeds a predetermined amount.As before, a cross-coupled feedback loop may be provided to modify the input signal to said other link in dependence upon the position of said one link, and each link may additionally be provided with a conventional local feedback loop.

According to one aspect of the invention a robot has a pair of links capable of relative pivotal movement in accordance with input signals defining the demanded positions of the links, and a cross-coupled feedback loop operative to modify the input signal to one link in accordance with information derived from the output of the other link, the modified input signal then being used to drive said one link.

According to another aspect of the invention a method of controlling the movement of two robot links moved in dependence upon input signals defining the demanded angular positions of the links, comprises modifying the input signal to one link in dependence upon information derived through a cross-coupled feedback loop from the output of the other link, and using the modified input signal to drive said one link.

The present invention is particularly suited for use in electrically-powered robots; it is also applicable, inter alia. to hydraulically-powered robots.

The invention will now be further described, by way of example, with reference to the accompanying drawings in which: Figure 1 is a schematic block diagram showing the control applied to two links of a robot according to the invention; Figure 2 shows in schematic form the hardware used in the preferred robot; Figure 3 is a flow diagram of the operational steps used by the robot; and Figure 4 shows schematically the robot.

Referring to Fig. 1, the robot has two pivotally interconnected links 1 and 2, and respective input signals defining the de manded position of the links 1 and 2 are applied at 3 and 4 respectively. The output 1 3 of the link 1 is fed back through a local feedback loop 5, having a gain K11, to alge braic summing junction 6. Similarly, the out put 1 4 of the link 2 is fed through a local feedback loop 7, having a gain K22, to an algebraic summing junction 8. Each local feedback loop 5 or 7 typically carries information relating to the angular position of the corresponding link 1 or 2 and its angular velocity.

To provide the multivariable control which is the essence of the invention, a cross-coup led feedback loop 9, having a gain K21, is taken from the output 1 3 of link 1 to the summing junction 8. In a similar manner, a second cross-coupied feedback loop 1 0. hav ing a gain K12, is taken from the output 14 of link 2 to the summing junction 6. Accord ingly the input signal actually applied to link 1 is the input signal 3 modified by the signals from the feedback loops 5 and 1 0, whilst the input signal actually applied to link 2 is the input signal 4 modified by the signals applied by the feedback loops 7 and 9. This ensures that the output 1 3 of link 1 affects the input signal applied to link 2, and the output 14 of link 2 affects the input signal applied to link 1.This interactive control prevents a com manded movement on one link causing an undesired "kickback" on the other link.

Fig. 2 shows in schematic form a practical way of realising the control structure shown in Fig. 1. In Fig. 2, angle and velocity signals are taken from the output of link 1 on lines 1 5 and 1 6 respectively and applied to the input of an analogue to digital converter 1 7.

-Similarly, angle and velocity signals from link 2 on leads 1 8 and 1 9 are also applied as inputs to the analogue to digital converter 1 7.

Input signals, forming the demanded position signals for the links 1 and 2, are applied on lines 3 and 4 to further inputs of the converter 1 7. The digitised signals from the converter 1 7 are applied to a control unit 22 incor porating a Motorola 68000 micro-processor which acts as the central processing unit for the control system. The output of the control unit 22 is supplied to a digital to analogue converter 23 having two outputs 24 and 25.

The output 24 feeds an amplifier 26 driving a motor 27 which applies torque to the link 1.

Similarly, the lead 25 supplies an analogue signal to an amplifier 28 driving a motor 29 applying torque to the link 2. The control unit 22 is operated in a time multiplexed fashion and executes a complete control algorithm in between one and two milliseconds. It will be appreciated that the lines 1 5 and 1 6 in effect provide local feedback for the link 1 and cross-coupled feedback for the link 2, whereas the leads 1 8 and 1 9 provide local feedback for the link 2 and cross-coupled feedback for the loop 1.

Reverting to Fig. 1, the gains K12 and K21 are preferably variable in dependence upon the angles of the links and the load applied to the robot links or arms. It has been found that the gain/load variation is theoretically linear, and the gain can therefore be automatically varied by means which detect the magnitude of the load.

The velocity signals on lines 1 6 and 1 9 may be derived from tachogenerators, but it is cheaper to use software algorithms to generate the required signals, e.g. by differentiating position signals or using a so-called "observer". A multivariable observer is a software algorithm which is a model of the actual system. The observer produces an output which is compared with the actual system output to produce an error signal which is driven to zero by the observer.

Fig. 3 shows a flow diagram which represents the operational steps performed by control unit 22. Thus it stores the value of angle appropriate to each of the links 1 and 2 (namely , and 2 respectively): then it stores values of velocity appropriate to each of the links 1 and 2 (namely V, and V2 respectively).

Using this information, it then calculates new values for the angle and velocity motion for each of links 1 and 2 by using the following operations: A: , = (Krll X A (K012 X 02 2) - (Kv11 X #,)(KV12 X H2) B: t1 = (to22 x A02) + (to22 x A02) - (Kvj2 X 02) - (Kv2i X 6 1) where:: try = value of angle for link n; = = difference between desired value of angle for link n and actual value of angle for link n; Kn,y = the gain, for angle 8 signals, in respect of the feedback loop appropriate to x, y (see Fig. 1); Kvx,, the gain, for velocity signals, in respect of the feedback loop appropriate to x, y (see Fig. 1).

Each of these calculations is checked by suitable error-detection techniques, then unit 22 generates electric signals representing these new values for transmision to the respective links after further processing by the digital-to-analogue converter 23, amplifier 26, 28 and motors 27, 29.

When the present invention is used in a hydraulically-powered robot, control unit 22 also stores a value for the acceleration of each link, and uses these acceleration values in the calculations A and B.

Fig. 4 illustrates the major elements of the robot described above. Thus the robot 40 has a control system 41, which operates in the manner mentioned previously, and a movable arm 43 (formed of the links 1 and 2) on the end of which is a gripper 44 for holding any article to be handled.

Some slight offsets in actuator positioning may occur due to stiction; occurrence of such offsets are particularly likely when gravitational torques are present. Thus advantageously, the robotic control system incorporates integral action, i.e. the time-dependent analysis of the actions of the control system in order to calculate what corrections are required to the existing (and immediately future) commands. However, this integral action should only be used in the final convergence stage of the actuator positioning operation, because if it were used generally, when actuators may saturate, it would cause substantial overshooting of the intended position.

A small amount of integral control action, in the form of a lag compensator, could be incorporated within the transfer function.

One unexpected advantage of a robot embodying the present invention is that it provides a particularly robust structure when the robot is used in tasks requiring the exertion of a pressure on a surface, for example a routing operation or a grinding operation. Conventional robots are programmed for such an application by using a positional offset technique so that a restoring force is set up; in order to exert a torque at the end effector, torques need to be exerted at several of the degrees of freedom of the robot. However as single-input single-output controllers will not give rise to these torques, they cause linkages other than the end one to move the grinding, resulting in deformation of the entire linkage structure of the robot.However, in a robot embodying the present invention with a multivariable control, the sensing of a positional error at one servo would set up compensation torques at the other axes without deforming, so making the robot structure more rigid. In this way, the robot provides movement of the end link (or the actuator itself, attached to the end link), while ensuring substantially no relative movement between the remaining links.

When used in this mode of operation, the robot should not incorporate integral action.

A robot embodying the present invention can readily overcome any problems of overcompensation when the actuators are used in saturated, or near-saturation, conditions. If robots are to move quickly it is desirable that the actuators should be saturated for a considerable part of any motion to be performed, in order to optimise use of power available; the actuators should not be saturated on final convergence to a set point however, as full control is required in those conditions in order to achieve accuracy and to prevent overshoot.

The problem of overcompensation in conventional robots arises when the control signal to an actuator of one link exceeds the saturation value for the actuator so that the link, while moving at its maximum rate, is moving at a slower rate than determined by another link which is monitoring the control signal.

Thus incorrect information is used when calculating what motion the second link should make in response to the movement of the first link with the saturated actuator, thereby resulting in incorrect compensatory motion of the second link.

The robot of the present invention overcomes this problem by providing demandupdating in the cross-coupling feedback loops.

The feedback loop therefore has an upper threshold limit equivalent to the saturation value, this limit usually being incorporated into the transfer function. Hence, when an activator is operating in saturation conditions, the activator(s) for the other link(s) will compensate in accordance with the actuator saturation value rather than the value of its input.

The use of cross-coupled feedback loops effectively removes the influence of one link on the other(s) and this decouples the system to provide independent control of the links.

When the robot is operating in the integral control mode, the control system regularly determines the difference between the actual position of one or more links with the intended or theoretical position of the link(s).

Any differences are summed over a given time period, and when the total exceeds a specified amount, a compensating signal is generated in order to position correctly the link(s).

Thus in the integral action mode, the control system makes the following calculation:

where 0 = actual value of angle of a link at a time t; Ot = intended value of angle of that link at the time t; W = compensation threshold; to, t1 = beginning and end points of the time interval for the integral action mode.

As soon as C = 0, the control system produces an electric signal of suitable size and direction to move the respective link(s) to the intended position. This integral action mode can be operated on a single link, or on a number of links individually, or on a group of links together (e.g. in the latter case the position error in the end link of the group may be monitored).

Claims

1. A robot comprising an arm having a plurality of links capable of relative movement in accordance with input signals defining the demanded positions of the links, a control system to effect control on the movement of the arm, the control system incorporating a cross-coupled feedback loop which has means to modify the input signal to a first link in accordance with information derived from the output of at least one other of said links, and means to use the modified input signal in order to drive the first said link, the control system having means to monitor, over a time period, the difference between the actual position of a link and the intended position of the link, and means to effect a compensating movement of the link when the difference exceeds a predetermined amount.

2. A robot according to Claim 1, wherein the monitoring means of the control system has means to detect, at a given moment, the difference between the actual position of a link and the intended position of the link, means to repeat the difference-detection operation at regular intervals, means to sum the detected differences, and means to effect a compensating movement of the link when the summation of detected differences exceeds a predetermined value.

3. A robot according to Claim 1 or Claim 2, wherein the control system has means to monitor the position of each of a number of links in the arm.

4. A robot according to any one of the preceding claims, comprising at least one additional cross-coupled feedback loop having means to modify the input signal to a link, other than the first link, in accordance with information derived from the output of any one or more the links, and means to use the modified input signal in order to drive said other link.

5. A robot according to any one of the preceding Claims, wherein the control system incorporates at least one local feedback loop having means to modify the input signal to a link in accordance with information derived from the output of that link.

6. A robot according to any one of the preceding Claims, wherein at least one crosscoupled feedback loop has a transfer function of a value to provide an appropriate weighting effect on the signal to which it is applied, means to vary selectively the value of transfer function effected by the loop.

7. A robot according to any one of the preceding claims, wherein a feedback loop has a transfer function which has an upper threshold value appropriate to the saturation limit of means to effect movement of the respective link.

8. A robot substantially as hereinbefore described with reference to and as illustrated in the accompanying drawings.

9. A control system to effect control on the movement of an arm of a robot according to any one of Claims 1 to 8.

10. A method of controlling the movement of a plurality robot links moved in dependence upon input signals defining the demanded positions of the links, comprises modifying the input signal of a first link in accordance with information derived through a cross-coupled feedback loop from the output of at least one other of said links, using the modified input signal to drive the first said link, monitoring, over a time period, the difference between the actual position of a link and the intended position of the link, and effecting a compensating movement of the link when the difference exceeds a predetermined amount.

11. A method according to Claim 10, comprising modifying the input signal of at least one link, other than the first link, in accordance with information derived through a cross-coupled feedback loop from the output of one or more of the links, and using the modified input signal to drive said at least one link other than the first link.

12. A method of controlling the movement of robot links substantially as hereinbefore described with reference to, and as illustrated in, the accompanying drawings.