The self calibration orientation system that is used for actuation means
Technical field
The present invention relates to the control to the orientation/positioning equipment with sensor device, the plan that the limits real behavior that can carry out by the control device (manipulator) controlled by this orientation/positioning equipment changes the input space of this sensor device.Therefore, actuation means orientation that imagination will be detected by this sensor device (promptly being covered by the input space of this sensor device)/the navigate to assigned target position of the input space coordinate system of this sensor device.
Background technology
Background of the present invention is orientation/positioning equipment field, with reference to Fig. 1 and Fig. 2 this orientation/positioning equipment is described below.Fig. 1 represents and can carry out directed stereo camera with a plurality of degree of freedom.Suppose will shown in cameras oriented be to make the visual object of certain position of the detected input space that appears at this video camera move to the true origin place of this input space.Fig. 2 represents to be used for calculating and carrying out action (that is, video camera moves) to realize the known treatment of described purpose.In first step, the output of this sensor (is video camera at this) is assessed, with the target location in the multidimensional input space of determining this sensor, this target location for example can be the sound source of being discerned.In the step of back, use predefined look-up table or analytic function that this sensor coordinates (angle etc.) is mapped to motor coordinate (voltage, electric current etc.), with required actuator (motor) order of the new orientation that generates the video camera of determining the front.
Provide the typical case of the sensor device of explicit directed information for example to be image sensor array, radar image sensor array or chemoreceptor sensor array.The typical sensors device that the implicit expression directed information is provided is the stereo microphone that is used for localization of sound source.Before can using described directed information, provide the sensor device of implicit expression directed information need extract this directed information.
The robot that comprises the sensor that is used to provide directed information is called as directed robot.The shortcoming of this directed robot is: in order can correctly to work, need carry out frequent calibration to them.After system's geometric shape changes each time (for example, when the physical relation between the actuator of sensor and robot changes), (for example change each time, change camera lens, change motor) afterwards and change or not controlled environment in use under the situation of robot, all must carry out this calibration.
In specific environment, carrying out manual calibration by the user is impossible or undesirable (being to deliver under the situation of the detector on other planet in this robot for example).In addition, in system's geometric shape frequent variations, frequently change sensor or motor, perhaps under the situation of uncontrolled environment, the very time-consuming and effort of frequent calibration.
Therefore, the directed robot of self calibration has been proposed.
The directed robot of this self calibration uses calibration program (scenarios), mathematical analysis or the parameter/hardware adjustment under people's supervision usually.
More than all collimation techniques all have following shortcoming:
The use of calibration program requires the actuator of robot to turn back to reference point usually.Therefore, it is impossible calibrating continually in new and/or uncontrolled environment.In addition, the described calibration of being undertaken by frequent use reference point is not enough to solve the nonlinear problem of motor response.
The calibration of being undertaken by mathematical analysis has following shortcoming: need the hypothesis of a plurality of implicit expression, this often causes great simplification.Usually only consider the subclass of variable.Therefore, the precision of described calibration steps is normally not enough.In addition, because the complexity height, and make the calibration of being undertaken by mathematical analysis not be suitable for on-line calibration usually.
The shortcoming of parameter/hardware adjustment of carrying out under people's supervision is, needs human expert's intervention.Therefore, this calibration program is not suitable for autonomous robot.In addition, this collimation technique is very consuming time usually, and highly depends on expert's experience.
Summary of the invention
Therefore, an object of the present invention is to provide a kind of method that is used to control orientation system, and a kind of self calibration orientation system, this system can not have the self calibration that carries out system under the artificial situation of intervening accurately.
Another object of the present invention provides a kind of method that is used to control orientation system, and a kind of self calibration orientation system, this system is suitable for solving the nonlinear problem in motor/sensor response, and be suitable for change and/or uncontrolled environment in use, wherein can not or not wish to safeguard.
Can realize above purpose by the feature of independent claims.Dependent claims has further been expanded central idea of the present invention.
According to a first aspect of the invention, a kind of method that is used to control orientation/positioning system has been proposed,
Described orientation system comprises at least one sensor device, and be used to control the orientation of actuation means (for example robot) and/or the actuator devices of location action, actuation means can be detected by described sensor device, and can carry out changing the input space of described sensor device with plan alternately with real world according to system.
Said method comprising the steps of:
-(S1) output information before the action of described sensor device is assessed, to detect the position of described actuation means in the input space of described sensor device;
-(S2) determine the target action back position (targeted post-action position) of described actuation means in the input space of described sensor device;
-(S3) by using predefined mapping function, action front position in the input space coordinate of described sensor device and target action back any of position are departed from, be mapped to actuator control coordinate, define the order that is used for described actuator devices;
-(S4) come by described actuator devices according to defined order directed/locate described actuation means, to carry out orientation/location action; And
-(S5) detect position after the actual act of described actuation means in the input space of described sensor device;
-(S6) according to any difference between the position, position and target action back after the actual act in the input space of described sensor device, adjust the mapping function that uses among (adapting) step S3, to carry out the increment adaptive learning (incremental adaptive learning) of mapping function
Use wherein respectively that repeating step S1 to S6 is at least once circularly through the mapping function adjusted.
In step S6, can come correction signal according to any difference between position and the target location after the actual act of actuation means in the input space of described sensor device.Can consider this correction signal and this actuation means action front position in the input space of described sensor device, adjust this mapping function.
Before adjusting mapping function, can multiply each other with a factor adjusting vector (adaptation vector), with the regulation (adaptation rate) of control mapping function, this factor is more than or equal to 0, and is less than or equal to 1.
Described method can also may further comprise the steps: according to the rationality (plausibility) of the difference between position and the target location after the actual act of described actuation means in the input space of described sensor device, for example, calculate the value of the confidence (confidence value) by the correlation of selected pattern.Can modulate described adjustment vector with this value of the confidence then, so that described regulation depends between for example selected sensor patterns before action and the degree of confidence of the correlativity after the action.
Can realize this mapping function by look-up table.This look-up table is associated sensor coordinates with corresponding actuator coordinate.
In step S1 to S6, for each time circulation, not only use the position of pattern, and used the interior neighborhood of limited range around this position, wherein can between two follow-up circulations, change the scope of this neighborhood.
Can in sensor coordinates or actuator coordinate, carry out adjustment to mapping function.
Another aspect of the present invention relates to a kind of method that is used to control orientation/positioning system, described orientation system comprises at least one sensor device, and be used to control the orientation of actuation means (for example robot) and/or the actuator devices of location action, this actuation means can be detected by described sensor device, and can carry out alternately with real world, change the input space of described sensor device with plan according to system
Described method preferably includes following steps:
-(S1) use predefined mapping function, described sensor device is arranged in the output information of the input space coordinate of described sensor device (1), be mapped to actuator control coordinate;
-(S2) determine position, target action back, this determines to carry out in actuator control coordinate;
-(S3) define the order that is used for described actuator devices (41,42,43) according to described definite result; And
-(S4) according to defined order, come described actuation means is carried out orientation/location, to carry out the orientation/location action of described sensor device by described actuator devices.
The output of a plurality of sensor devices can be mapped to respectively in the actuator coordinate, make up subsequently, wherein can come described among the execution in step S2 to determine according to the result of this combination.
In carrying out step S2, really regularly, can consider the constraint of described actuator devices.
Another aspect of the present invention relates to a kind of computer software product, realizes said method when this computer software product is used for moving on computer equipment.
Another aspect of the present invention relates to a kind of automatic calibration orientation/positioning system.This optimum system choosing ground comprises at least one sensor device, calculation element, and be used to control the orientation of actuation means (for example robot device) and/or the actuator devices of location action, this actuation means can be detected by described sensor device, and can carry out alternately with real world, change the input space of described sensor device with plan according to system
Described calculation element comprises:
-be used for output information before the action of described sensor device is assessed, to detect the device of the position of actuation means in the input space of described sensor device (1);
-be used for determining the device of described actuation means in the position, target action back of the input space of described sensor device;
-be used for sending the device of ordering to actuator devices, this order is by using predefined mapping function, with any the departing between action front position in the input space of described sensor device and the position, target action back, be mapped to actuator control coordinate and generate;
-be used for detecting the device of described actuation means position after the actual act of the input space of described sensor device; And
-be used for according to described actuation means calibrating described mapping function, with the device of the increment adaptive learning of carrying out mapping function in any difference between the position, position and target action back after the actual act of the input space of described sensor device.
Description of drawings
In the following detailed description, the present invention will be described with reference to the accompanying drawings, in institute's drawings attached, uses similar label to represent similar part, in the accompanying drawings:
Fig. 1 represents the orienting device according to prior art;
Fig. 2 represents the orientation action computing according to prior art;
Fig. 3 a, 3b have schematically shown according to alternative of the present invention and have handled;
Fig. 4 represents the embodiment of correction signal with the adaptive learning that carries out mapping function;
Fig. 5 represents to be used to quicken to adjust the option of processing;
The study that Fig. 6 is illustrated in Fig. 3 a or 3b is introduced the value of the confidence in handling; And
Fig. 7 represents the block scheme of self calibration orientation/positioning system of the present invention.
Embodiment
Now treatment in accordance with the present invention is described with reference to Fig. 3 a.(referring to Fig. 2) compares with known treatment, signal of sensor is assessed, to determine the target location of this actuation means in the input space of sensor device 1, this target location obtains by precalculated " once (one-strike) " action.This means, at first constitute an action, promptly mutual with the plan of real world, carry out whole action then.Only after this action is finished, just the action of original plan and the difference between the actual result are assessed, and planned and carry out extra corrective action.
According to the present invention,, used the mapping of adaptive learning for the sensor coordinates with determine action is mapped in the actuator coordinate.In first circulation of orientation, the mapping function that uses rough estimation or set arbitrarily.In at least one other circulation, will use improved " study " mapping function.
This method has the following advantages: under the situation of change in the sensor/actuators characteristic (wearing and tearing, the variation, replacing, the temperature fluctuation that are provided with ...), automatically realize correct sensor/actuators mapping by less study/adjustment circulation, and need not manual intervention, also need not by the correlativity between analytic definition actuator and the sensor coordinates.
Handle in order to carry out adaptive learning, when having finished circulation and having carried out action, whether assess this action successful, promptly, whether this actuation means is in fact near the target location, and generate correction signal according to any deviation, and it is fed to map unit, to carry out the incremental learning step.Can repeat this circulation, till the deviation between action of being planned and the actual result is less than predetermined threshold.Alternatively, can repeat this circulation continuously.
Note, generate this correction signal this action is finished after, this action means the required result of acquisition in process.
In a word, according to this example, in sensor coordinates, determine the action that to carry out; To determine that subsequently the result who produces is mapped to actuator (motor) coordinate.
Fig. 3 b represents that an alternative is provided with, and wherein carries out in the actuator coordinate system and determines at first sensor coordinates to be mapped to the actuator coordinate before the processing.This set has the following advantages: for example, can more easily consider the restriction (maximum magnitude, speed, acceleration etc.) to the actuator coordinate in the actuator coordinate in the current process of just determining.
In addition, this alternative according to Fig. 3 b is provided with, different, final diverse signal of sensor can be mapped to respectively in the actuator coordinate separately, can easily these actuator coordinates be made up, with the action of determining according to the actuator coordinate that is made up to carry out.
First embodiment of the adjustment of correction signal and mapping function is described with reference to Fig. 4.As mentioned above, the purpose of mapping is that the coordinate conversion of the action will be in sensor coordinates determined is the coordinate that is applicable to this actuator devices of control.Therefore, Fig. 4 a shows sensor array and motor array respectively.In this example, imagination will be carried out following actions, by this action, will move the front position
Transfer to the center of sensor array (the sensor input space).Use default mapping function, suppose that corresponding actuator (motor) order is
Yet,, will not move the front position by using this motor order
Accurately move to the target location in the sensor array, but moved to position, action back
According to present embodiment, existing increment adaptive learning (and the increment adaptive learning of this mapping function will be used for next circulation) as linear extrapolation execution mapping function, i.e. the adaptively correcting motor order of coordinate (degree of freedom) x is:
Wherein
And
α
xBe optional parameter (be equal to or greater than 0, and be equal to or less than 1), be used to control the regulation of coordinate x.
Note, this extrapolation type study processing can also be used for Nonlinear Mapping.
In addition, study as shown in Figure 4 can be handled the definite strategy that is applied to Fig. 3 a and Fig. 3 b.
At last, this study can also be handled the situation that has single and a plurality of stimulation emissive sources (stimulus-emitting sources) in the input space that is applied to sensor.
(this study is handled and is adjusted the rate parameter alpha in order to quicken the study processing
xSlow down potentially), can in each time circulation, use more than one position.As shown in Figure 5, for the convergence of quickening to adjust, can use the qualification neighborhood of a position.Therefore, can change in time by the scope that makes this neighborhood, can realize following mapping function, this mapping function keeps plasticity under the situation of the unexpected misalignment of system, and keeps local stable simultaneously, to guarantee any nonlinear accurate expression.
Fig. 6 has schematically shown and can obtain confidence measure according to correction signal, for example carries out reasonableness test.Can use this confidence measure (having the value between 0 to 1) that regulation is modulated subsequently.The introducing of confidence measure makes the possibility minimum (can produce wrong set-up procedure under the situation of estimating correction signal improperly) that produces wrong set-up procedure.Usually come correction signal by correlation method, in this case, can use (after normalization) this correlativity as confidence measure.
Fig. 7 has schematically shown according to self calibration orientation/positioning system of the present invention.
This system comprises sensor device 1, is used to gather the information of real world, in the input space with the one dimension at least that this actuation means is positioned at this sensor device 1.Alternatively, this sensor device 1 can be radar image sensor array, chemoreceptor sensor array, perhaps is used to locate the stereophony microphone of the sound that is sent by object 2, perhaps any other orientation sensor.
Sensor device 1 is installed on the frame, and this frame comprises the actuator devices 41 and 42 that for example is applicable to and waves this sensor device 1.In this example, described actuator devices 41 and 42 for example can be a linear electric machine.
Therefore, actuator devices 41 and 42 is suitable for itself changing by mobile sensor apparatus 1 input space of sensor device 1.Note, in the structure of this instructions, will be suitable for handling and be called " action " by any plan that controlling of real world changes the input space of sensor device 1.Therefore, " action is preceding " and " action back " is meant that respectively this action by this actuator devices control takes place before and state afterwards.
Sensor device 1 links to each other with calculation element 3 with 42 with actuator devices 41.
In Fig. 7, described calculation element 3 is computing machines, and this computing machine is applicable to the target action back positional information of this actuation means of definition in the input space of sensor device 1, and calculates the order that is used to make actuator devices 41 and 42 mobile sensor apparatuses 1.Alternatively or additionally, this calculation element is used for mobile robot 43 the order of control device (manipulator) 8 applicable to calculating, and moving of control device 8 also is the action that is suitable for changing the input space of sensor device 1 under the control of actuator devices 41 and 42.
When calculating is used for the order of each actuator devices 41,42 or robot 43, this calculation element uses following information: the action front position information that is provided by described sensor device 1, promptly, before this action generation, the information that the position of this actuation means in the input space of sensor device 1 is relevant; And target action back positional information, that is, and represent this action generation after, the information of the target location of actuation means 2.For example, this target location can be the center of the input space of sensor device 1.Alternatively, especially under the situation of the control device arm of actuator devices control example such as robot etc., this target location can be set at the position (still being arranged in the input space of sensor device 1) of the object that will control by this robot.
Calculation element 3 comprises determines device 5 and comparison means 6.
Determine that device 5 is suitable for determining: finish described actuator devices 41,42 and/or described robot 43 by moving described sensor device 1 or by after moving the action that 8 pairs of described objects 2 of described control device and/or described control device 8 carry out, by positional information after the action of sensor device 1 output whether with the targets action back position consistency of calculation element 3 previous definition.
Determine that action back object-sensing station information and destination object-sensing station information do not match if determine device 5, then calculation element 3 further by using the comparative result by comparison means 6 outputs, is proofreaied and correct the calculation command that is respectively applied for actuator devices 41,42 and control device 43.Order after proofreading and correct can be used for subsequent action then.
In order to realize the learning strategy of Fig. 3 a, calculation element 3 is suitable for calculating the difference between action back object-sensing station information and the target action back object-sensing station information, and by using following formula that the calculation command that is used for actuator devices 41,42 or control device 43 is carried out from normal moveout correction:
Wherein
Be the corrective command that is used for actuator devices (41,42,43) at the dimension x of moment t+1,
Be the not corrective command that is used for actuator devices (41,42,43) at the dimension x of moment t-1,
Be after described actuator devices (41,42,43) not having be carried out control according to corrective command, object-sensing station information after the action of the dimension x of moment t,
Be target action back object-sensing station information at the dimension x of moment t-1, and α
xIt is the constant that is used to control the regulation of dimension x.
Owing to considered the current precision that realizes by the order that before was used to control actuator devices 41,42 and/or control device 43 clearly, so the use of described difference makes it possible in point-device mode the calculation command that is used to control actuator devices 41,42 and/or robot 43 be proofreaied and correct.
Order after can will proofreading and correct by calculation element 3, that is, the mapping function of adaptive learning stores in the described memory storage 7 of described command mapping.
Use the order that is stored in the command mapping to calculate any newer command that is respectively applied for actuator 41,42 and control device 43 by calculation element 3.
Export before the action by output and this sensor device after the action of automatic relatively sensor device, and use described comparative result that the calculation command that is used for actuator devices is proofreaied and correct, method of the present invention makes this positioning system to calibrate automatically, and adapts to variation and not controlled environment.
Therefore, the invention provides a kind of method and a kind of self calibration orientation system that is used to control orientation system, this method makes this system high precision to calibrate automatically, and need not manual intervention.In addition, method and the self calibration orientation system of the present invention that is used for controlling orientation system of the present invention is suitable for the nonlinear problem of processing motor/sensor, and is suitable for using in variation of can not or not wishing to safeguard and/or not controlled environment.Therefore, system and method for the present invention can be operated under the situation of manual intervention for a long time not having.
With comparing of prior art from orientation/self calibration system, system of the present invention more simply, more flexible, and need less maintenance.Can change the geometric shape of system of the present invention, and not need brake gear and sensor device are carried out new calibration.