| Tactile Sensors for Robotic Applications |
|
|
|
Tactile data contains information about magnitudes, distributions and locations of forces. It also provides information about the contact area and the pressure distribution over it. With resistive tactile sensors, changes in electrical resistance are detected by an element made from electrically conductive foam. The soft, deformable surface detects continuous pressure with excellent sensitivity and resolution. The electrical resistance measured between two electrodes on the same side of the conductive foam is derived from electrical conductivity through a number of simultaneously conducting paths.
The electrical resistance measured between two electrodes on the same side of the conductive foam (one tactile element). The tactile sensors have been developed with the following specifications: One finger consists of two 16x4 cells, two 16x2 cells, and one 6x2 cells, making up the total 408 cells for the two fingers. The width of the fingers is 20 mm, their length is 55 mm excluding an aluminum core and they have a thickness of 12 mm. Figure below shows the variations in the contact resistance when forces are applied to the tactile sensor surface. The response was monotonic, although not perfectly linear, when the forces were small (between 0 and 4 Newton). The measured values show that such a tactile sensor has a high sensitivity within this range and a lower sensitivity to increasing forces outside of this range. The advantage of a smooth and continuous curve lies in the ease of mathematical fitting and the applicability of other computational processes.
The tactile sensors are mounted on both surfaces of a 2-finger gripper on all four sides and their tips where they are used to sense force and moment data during operation of a 12-DOF mobile robot "ATHENE".
In the figure below, tactile force images are shown using the NURBS technique. The data are displayed using three dimensional plots. The height of each pair of images corresponds with the tactile element data output. The first image at the upper left, (a), is derived robotic prehension of a ball. The second image, (b), represents a toroidal ring of approximately 1.4 cm diameter. The third image, (c), is that of a cylinder. The final three images (d-f) show the tactile surfaces when a metal cuboid was pressed against it at different angles. In all cases, variations in the pixel height along an object’s edges resulted from variations in pressure distribution during indentation of the object against the sensor surface.
Minimum Grasped Force A tactile sensing system, capable of measuring near static acceleration, has been developed. A predictive model which uses basic methods adapted for use in real grasp optimization applications has been proposed. Prevention of premature release and the application of minimum prehension force have been addressed without the need to determine the coefficient of friction between object and robot gripper. Predictive models have been used to develop a set of rules to predict pre-slip based on fluctuations in tactile signal data. The techniques proposed in the research utilise considerably less computing power and faster for slip-detection comparing with 3D-MEMS accelerometer.
With the composition model of Maxwell and Kelvin-Voigt as shown, it can describe both elastic behaviour which represents a crosslinked polymer and the steady state creep typical of an uncured polymer. The models are simple and suitable for experimental representation of almost any polymer foam over an extended period of time.
To simplify the analysis as much as possible, but to retain the essential features to be investigated, the vibration considered at a contact point is a finite cubic block attached to a rigid wall by a simple spring and dashpot. The normal stiffness, linearized about the mean rider position. In reality, the coefficients of friction are different for each contact of the rider. With regard to constant friction, the argument is that in order for friction to change, the real contact area, and thus the mean normal separation, of the surface must change. Efforts to verify this demonstrated a reduction in friction due to normal vibration. With the measured frictional shear force being a function of real contact area, an apparent reduction in friction in the presence of normal vibrations can be expected. The idea was that normal vibrations could influence the mean surface separation and hence the real area of contact. The two models in previous figure can be applied to explain the operation of robot gripper fingers covered by such tactile sensor arrays. The effective coefficient of friction will reduce as the magnitude of the exerted force increases. In other words, the compressive areas have lower coefficients of friction than the tensile areas. In order to implement the proposed models in practical tactile applications, the rules set are verified using a bubble search and subsequent comparisons. A bubble search orders the history of data at different contact points and times. It is therefore possible to resolve and localize multiple separations and stick points on a contact surface caused by micro-movements. The frequency of vibration and the location of the vibrating elements together with x-bar and y-bar values become the relevant conditions because this tactile sensor can sense in all three dimensions of deformation. However, the contact deformation changes several times within a grasping cycle making compliant control difficult. Nevertheless, the algorithm presented here is capable of generating satisfactory compliant motion control in spite of changing contact deformation. The use of this algorithm in several experiments under various grasping conditions has shown that the tactile information can provide a rapid detection of precursor conditions to slippage (pre-slip). Contact Identification A novel and fast framework for contact recognition based on the eigenvalue trajectory of a 3-dimensionally deformed surface onto Quadric parameters has been presented and mathematically proved. The technique called “eigenvalue trajectory analysis”, is introduced and adopted for specifying the margin of classification and classification thresholds. It has been shown that the matrix from the parameter of Quadric surfaces by interpolation of tactile data may be formulated by eigenvalue decomposition and can reflect under all contact geometries. The smallest component of an eigenvalue can be used to estimate and identify object shapes without using any other references, whereas classification is used as the principal indication of surface identity. The shape reflectance parameter pertaining to (unique to) each surface may be recovered and identified. It has been shown that the reliability of the surface classification method and the accuracy of transformation are dependent of object shapes. The test objects used were: an oval object with two major axes of 14 mm and 11.7 mm; a cylindrical object of 6.0 mm diameter and 20 mm length; a cube with the dimensions 10x 15.9x10 mm; a sphere with a 9.5 mm diameter; and dumbbell-shaped object 13 mm in length and 10 mm in diameter.
According to above figure, the thresholds of object 1, object 2, object 3, object 4 and object 5 correspond with the oval, cylindrical, cube (box), ball shapes and dumbbell-shape object, respectively. Each object has a different eigenvalue in the eigenvalue trajectory with no particular increasing or decreasing order in terms of their levels. The principal idea used to classify the contact is in the matched threshold of the eigenvalue trajectory. The noise level is maintained below 8% of ADC data. It will not be statistically meaningful for, nor affect, classification. To improve classification the surface properties must be modified. One option is to parameterize the Quadric surface in higher space and obtain the higher surface discrimination whilst preserving the eigenvalue trajectory behavior. Future research concerns formulating a set of such challenge problems.
|
| Next > |
|---|