The invention relates to a signal processing method designed for a tactile sensor, said tactile sensor comprising a tactile surface and elementary sensors arranged along the tactile surface and capable of three dimensional force sensing, wherein sensing is carried out by reading the signals of the elementary sensors. The invention also relates to a tactile sensor suitable for carrying out the signal processing method.
There are many known methods in the prior art for processing read-out signals of elementary sensors during a tactile event sensing carried out by tactile sensors. Tactile sensors are most often used in robotics, but they may of course be applied in any other technical field, where tactile sensing is required. In the case of objects held by robot hands, various events may take place, because the objects may be subjected to many kinds of sudden forces. There is a demand for designing systems that allow an appropriately quick response to these impacts coming from outside.
There are known elementary sensors designed for three dimensional force sensing. For tactile sensors, such elementary sensors may be preferably used, which are arranged preferably along a tactile area defined by the tactile sensor. Objects in contact with each other are in contact in most cases along an extensive surface, and therefore a large number of elementary sensors are required. The force distributions or in other words the pressure distribution maps or force maps determined by the signals of a large number of elementary sensors must be evaluated as quickly as possible, and the events must be recognised and classified possibly in real time.
It is a disadvantage of prior art methods that the slow processing of elementary sensor signals makes it difficult to take a rapid decision, and to apply intervention and control on the basis of the sensing provided by tactile sensors. Due to the fact that generally more than one elementary sensor is applied for one tactile sensor in order to achieve a higher sensing reliability, and the signals may be different subject to the local tactile events taking place on the tactile surface, the problems described above are aggravated in practice.
A further disadvantage of prior art solutions is that sliding and torsion events—considered to be extremely problematic from the aspect of sensing—cannot be handled properly by known signal processing methods.
It is an object of the invention to provide a signal processing method for a tactile sensor, which is exempt from the disadvantages of prior art approaches. It is a further object to provide a method which enables faster sensing, and thereby faster intervention and control than known solutions. It is a further object to provide a signal processing method that enables the appropriate sensing of sliding and torsion events. It is also an object of our invention to design a tactile sensor which enables the meeting of the criteria above.
The method and tactile sensor according to the invention are defined in the independent claims. Preferred embodiments and implementations of the invention are defined in the dependent claims.
The invention will hereinafter be described on the basis of preferred embodiments depicted by drawings, where
The method according to the invention recognises and classifies typical events taking place on the tactile surface by processing the three dimensional forces acting on the surface. Preferably, the algorithm senses twelve basic events and their combinations. The twelve events are on the one hand the positive and negative forces/displacements taking place along the three normal spatial axes x, y and z, and the positive and negative torques/swivels around the axes x, y and z.
According to the invention, for each elementary sensor, force components Tx, Ty and Sn measured along the three normal spatial axes x, y and z are determined, and a force map MTx, MTy and MSn is created from each of the force components along the same axis.
The force maps MTx, MTy and MSn are preferably generated in a way that a function of N values is assigned to the force components Tx, Ty and Sn, preferably the force maps are coded as a greyscale image. Each elementary sensor corresponds to one pixel in the greyscale images. For example, coding can be carried out in a way that one extreme value of the greyscale is assigned to the positive force component, the other extreme value of the greyscale is assigned to the negative force component, and the medium value of the greyscale is assigned to the force component of essentially 0 magnitude.
Consequently, the magnitude of the force component acting in a given point is shown by the intensity of the greyscale image. The normal rest status is shown in grey, the positive force is represented by a totally black colour and the negative force is in white. Hence, the three greyscale images so generated codes for each component the forces acting on the surface in a given moment of time.
According to the invention, the force maps MTx, MTy and MSn are evaluated and this can be implemented preferably by image recognition with a cellular neural network. After evaluating each force map separately, the results of the three evaluations are jointly evaluated. According to the invention, these evaluations preferably mean that the actually appearing value combination is determined out of predetermined value combinations.
The invention is suitable for both static and dynamic sensing. In the case of static sensing, the magnitude of force components is coded, while in the case of dynamic sensing their changes can be coded. In the latter case, the unchanged force component will be grey, the one changing in positive direction will be black and the one changing in negative direction will be white.
As shown in
These images are preferably processed by a CNN-UM (Cellular Non-linear/Neural Network Universal Machine). In a way known per se, the CNN-UM based on non-linear local interactions is excellent for detecting events in space and time. The characteristic patterns appearing in the images are recognised by individual, special templates. The event that has actually taken place can be concluded from the logical combination of recognised patterns.
The 2×2 squares below the legends Tx Ty Sn stand for coded force components of forces arising on the surface of the 2×2 sensor unit. In case (a) shown at the top of the figure, under the impact of a lateral tensile force, component Tx of the resulting forces will grow in a positive direction, coded in black, while the vertical and normal components do not change and therefore they are coded in grey (hatched). Of course, a different coding principle may also be applied and a number of shades other than three may be used for coding. Case (b) shown at the bottom of the figure indicates a Tx component of negative direction.
In
In the case depicted in
In the case shown in
In an example tactile sensor embodiment, a piezo-resistive sensor block consisting of 8×8 elementary sensors and designed for three dimensional force sensing is applied. Si based tactile sensors are used as the components of the sensor block. Each elementary sensor consists of a bridge suspended at four points above an etched hole. Piezo-resistors are inserted at the four suspension points. Under the impact of forces acting on the bridge, the piezo-resistors are compressed or expanded, and hence their resistance changes. This design enables measuring the forces acting on the suspended surface. Each piezo-resistor is a varying member of a voltage divider bridge. By tracking the voltage in the central node of the voltage divider, the voltage generated in the four bridges can be measured. On the basis of the voltages measured, the three dimensional components of the force acting on the sensor surface can be calculated in a way known per se.
To make sure that the voltages of all bridges in the sensor block can be measured, 8×8×4 that is 256 voltage rates are to be processed in parallel. In order to reduce the number of parallel lines, a decoder circuit as shown in
The thirty-two voltage values so obtained are the input values of a multiplexer 21. The output of the multiplexer 21 is passed via an ND converter 23 to a microcontroller 24. The addressing/control signals of the multiplexer 21 are also generated by the microcontroller 24.
The four piezo-resistors located in one elementary sensor of the sensor block behave as the varying branches of four semi-bridge circuits. For operating the sensor block, as an example, a supply voltage of 5 V is required. In an unloaded condition, the four output voltages of one elementary sensor will be around 2.5 V, because the 5 V is shared on a 50% basis by the constant and varying branches of the semi-bridge circuits. However, the scatter of these voltages shows a deviation of some 100 mV in practice, and therefore it is required to calibrate the sensor block. When the sensor is subjected to a force, the value of measured voltages changes, but only in a magnitude of some 10 mV (in a range of approx. 0 to 50 mV). Because during the measurements it is exactly this few mV difference which is significant, an approach is chosen which only amplifies and digitises this change. After multiplexing, the voltages are supplied from the sensor block to a non-inverting input of a differential amplifier 22. A voltage measured in an unloaded condition on the given sensor and stored in the register 26 is supplied from the output of a D/A converter 25 to the inversion input of the differential amplifier. Hence, the differential amplifier 22 only amplifies the difference vis-à-vis the rest position.
The measured values are passed on from the microcontroller 24 via an interface circuit 27 (e.g. RS 232) for processing.
By means of the invention it becomes possible to hold without slipping various fragile objects of unknown strength, for example a slippery thin-walled drinking glass. Such a task arises for example when a robot arm with two fingers holds a drinking glass, into which water is filled, consequently the weight changes constantly. In this case, when the weight of the object increases, the clamping force must be raised automatically and proportionally. In the case of a too low a clamping force, the slipping of the glass must be sensed by the system and on the basis of the tactile information it must increase the force exerted by the robot's fingers in order to stop the slipping.
The system using the method covered by the invention is able to automatically control this force. If the robot clamps with a constant force a drinking glass filled up half way, the force Sn normal to the tactile sensor surface does not change. The vertical component Ty of the lateral shear forces increases proportionally with the increasing weight of the glass, because the component is directly proportional with the weight of the drinking glass. The slipping does not occur as long as Ty is lower than the static adherence-frictional force between the glass and the robot's fingers. If the object starts to slip, the value of Ty immediately starts to decrease and then settles at a constant value which corresponds to the slipping-frictional force.
The increasing and then sudden decreasing of Ty as described above can be appropriately detected by the method and system according to the invention and hence it is excellent for solving a task considered to be problematic in robotics.
The invention, is of course not limited to the preferred embodiments and implementations shown as examples in figures, but further modifications are possible within the scope of the following claims.
Number | Date | Country | Kind |
---|---|---|---|
P 07 00080 | Jan 2007 | HU | national |
Filing Document | Filing Date | Country | Kind | 371c Date |
---|---|---|---|---|
PCT/HU07/00120 | 12/11/2007 | WO | 00 | 7/23/2009 |