Evaluation of arbitrarily preconfigurable collisions between any body position of a living being and any object of any shape

Abstract

The invention relates to an assessment of a collision between a body position of a living being and an object, comprising the method steps of: a) providing a 3D model of the object to a calculation unit; b) providing a shell-mesh grid model of the body position to the calculation unit, wherein for each field of the mesh grid a stress-deformation characteristic curve is predefined; c) aligning the 3D model and the shell-mesh grid model; d) stepwise displacing the 3D model and the shell-mesh grid model relative to each other, wherein for each step of displacement an indentation image is generated with indentation pixels, the pixel values ​​of the indentation pixels representing the indentation depth of the 3D model in the field of the mesh grid; e) determining a stress value for the indentation pixel according to the stress-deformation characteristic curve and the pixel value; and f) calculating the force acting on the body position by summing up the products of the stress values ​​and the area of ​​the field of the mesh grid.

Description

[Technical field]

[0001] The present invention relates to a method for assessing any given collision between any given body position of a living being and any given object, in particular with regard to injury risk. In particular, the living being may be a human and / or the object may be a robotic device, e.g. a collaborative robot. The present invention also relates to a corresponding collision assessment device for presetting a speed limit that prevents the robotic device from causing injury in case of a collision with the living being, e.g. a human. The method or device may also be represented and understood as a method or device for repeatedly identifying a safe robot speed. [Background technology]

[0002] When an object, such as a robot, impacts a living organism, such as a human, deformations occur in the skin and underlying tissue at the relevant body location of the organism. Here, the deformation or impact depth is related to the reaction force with which the tissue resists the impacting object. If the reaction force or deformation exceeds a certain threshold, serious injuries can occur. For industrial robotic devices that move around humans without additional protective measures, so-called collaborative robots (also called "cobots"), ISO10218 and ISO / TS15066 prescribe biomechanical limit values ​​that prevent injuries from occurring during collisions with robots.

[0003] Due to the hyperelastic and viscous behavior of human, animal, or plant tissues, the biomechanical relationship between reaction force and deformation path is highly nonlinear. The force-deformation characteristic curve profile describing the tissue's biomechanical reaction or response behavior ("biomechanical response") is close to exponential, with an approximately linear slope above a certain threshold. This is shown by an exemplary profile in Figure 1. The specific profile of the force-deformation characteristic curve associated with a given impact is highly dependent on the involved body positions, the shape of the impacted object, and its impact velocity.

[0004] For the assessment of injury risk of a robot, the important force-deformation characteristic curve plays a key role. Once the characteristic curve for the relevant body position and the contact position of the robot is known, the collision profile can be determined based on the characteristic curve and the robot model, and the maximum contact force at the assigned maximum deformation can be calculated based on the robot's physical parameters and the robot's speed. Correspondingly, it can be shown that a (simulated) collision will result in a significant injury risk if the calculated contact force exceeds the biomechanical limit values ​​applied to the robot under observation. Alternatively, it is also possible to iteratively calculate the safe velocity that the object (here the robotic device) must not exceed at each observed time, avoiding the exceeding of the limit values ​​at that time due to a collision with the organism and thus the organism's injury.

[0005] In collaborative robotics, model-based risk assessment is of particular interest, since the diverse movements of the robot can result in many contact situations that can be highly dangerous for humans. Due to the many different possible contact situations, to date there is only one measurement method available for safety testing of collaborative robots. However, this measurement method has high costs and technical limitations that prevent the widespread use of these robots in industry and other areas such as care.

[0006] The assessment of a given collision can be done in the prior art using the Finite Element Method (FEM), where the relevant force-deformation characteristic curves are identified by numerical calculations. However, the computational complexity of using FEM together with the corresponding tissue deformation and contact models is generally very high, making this method unsuitable for integration into the control of robotic devices. This is because, when assessing a collision in a real scenario, the required characteristic curves must be calculated in a short time, in any case less than a minute, and based on this the speed of the robot or other safety-related parameters must be adjusted to prevent injuries during the collision. Summary of the Invention [Problem to be solved by the invention]

[0007] The challenge is therefore to quickly and accurately assess any pre-definable collision between the body position of a living being and any object and, based on the assessment results, preferably directly make adjustments in the robot, such as robot speed, which reduces, for example, the risk of injury to an operator, preferably to an acceptable level. [Means for solving the problem]

[0008] This problem is solved by the subject matter of the independent patent claims. Advantageous embodiments emerge from the dependent patent claims, the description and the drawings.

[0009] One aspect relates to a method for assessing (any) predefined collision of (any) predefined body position of a living being with (any) predefined object, in particular with respect to injury risk (suitable for controlling a robotic device in an automated and / or real-time manner). The living being may be a human and / or an animal and / or a plant. The object is preferably a rigid object compared to the body position under observation, i.e. an object that does not deform upon collision. For example, the object may be (any) predefined part of a robotic device, i.e. a robot. The collision may comprise an impact and / or a clamping. Wherein the method comprises a series of method steps:

[0010] One method step is to provide a geometrical volume model of a given object that impinges on the body position under observation. This model can be in a file format common for 3D models and is provided to the computation unit. A further method step is to provide a geometrical shell-mesh grid model of the given body position to the computation unit. The geometrical shell-mesh grid model of the body position under observation can also be called a body part model. Here, the shell or envelope of the shell-mesh grid model is close to a three-dimensional mesh grid structure as in the FEM, and each field Fij of the associated mesh grid of the shell-mesh grid model is a square. Correspondingly, by the indices i, j, each field of the mesh grid is uniquely determined or determinable as a two-dimensional part manifold of the three-dimensional space. Preferably, all square fields have the same side length and, similarly as known in matrices, the indices can designate columns and rows of the mesh grid.

[0011] Here, for each field Fij of the mesh grid, a stress-strain characteristic curve is predefined. The stress-strain characteristic curve can also be predefined in a modified form, for example in the form of a respective force-strain characteristic curve as an area-adapted stress-strain characteristic curve. A later-described multiplication of the determined stress values ​​is thereby preferred, whereby for an indentation pixel Pij, a respective force value is determined according to the respectively assigned force-strain characteristic curve and the respective pixel value Pw. Since the commutative law applies for the multiplication, the use of the stress-strain characteristic curve and the subsequent multiplication with the respective area is equivalent to the use of an area-adapted stress-strain characteristic curve in the form of the respective force-strain characteristic curve. The term "stress value" in the detection method step described below therefore also includes a "force value" corresponding to the product of the stress value with the area from the calculation of the applied force method step. The described method steps can be performed in a predefined (in the claims) order, but this is not necessary.

[0012] The square basic shape of the field or mesh element Fij allows the 3D or volumetric model to bend only around one axis. Therefore, non-square mesh structures do not work with the method proposed in this application. Here, the side length of the mesh element has a significant impact on the accuracy of the model, i.e., the shorter the side length, the more accurately a given collision can be quantified and evaluated. For example, the side length can be less than 2 mm, less than 1 mm, or less than 0.5 mm. Here, the stress-deformation characteristic curves for each square mesh element Fij can be stored in a database. The stress-deformation characteristic curve reflects the strain behavior of a finite surface at the body position represented by the square mesh element or field Fij. This can be determined, for example, by subject trials or FEM analysis.

[0013] Using the computation unit, an alignment is performed in the virtual space between the provided 3D model and the provided shell-mesh grid model, where the relative positioning of the two models corresponds to the relative positioning of the object and the body position at a given collision in the real space. Advantageously, the alignment is performed such that the distance between the two models is zero after the alignment at least at one contact point of the two models (i.e. at one or more contact points of the two models), i.e. at the point where the object and the body position first come into contact at the collision.

[0014] After this method step, a stepwise displacement of the 3D model and the shell-mesh grid model relative to each other is performed by the calculation unit in the virtual space according to the impact direction determined by the given impact. Preferably, this is performed up to the expected maximum deformation path. For example, the impacting object is displaced in discrete steps, in particular in steps as small as possible, so that it geometrically sinks into the three-dimensional shell model of the body location. The displacement can be performed iteratively correspondingly, but also in just one displacement step. Now, after each displacement step, for each (iteration) step k of the stepwise displacement, an indentation image is generated with a respective indentation pixel Pij, where the respective pixel value w of the indentation pixel Pij represents the indentation depth of the 3D model in the field Fij of the mesh grid of the shell-mesh grid model assigned to the respective indentation pixel Pij. If the indentation image is filtered, weighted or otherwise processed, as described below, the initially or originally generated indentation image is also called the "original indentation image" and the further processed indentation image is called the "filtered indentation image". Like the surface of the body location, i.e. the shell, the indentation image is a two-dimensional image. Thereby, for each displacement step there is (exactly) one two-dimensional indentation image whose pixels indicate by indices i, j the positions at which the struck object deforms the body model and also by the assigned pixel value Pw how strongly the body model is deformed at these positions. The stepwise displacement of the object in the virtual space can be repeated (iterated) until a pre-set maximum displacement distance is reached. Here, the indentation depth and therefore the pixel value Pw can be determined in various ways, as further described below in exemplary embodiments.

[0015] A further method step is to determine the respective stress values ​​for the indentation pixels Pij, ideally with non-zero entries, for said or at least one indentation image, preferably for a plurality of indentation images, according to the stress-deformation characteristic curves and the respective pixel values ​​Pw assigned to the corresponding fields Fij. In practice, fields Fij and indentation pixels Pij with identical indices i, j can be assigned to each other. This allows a significantly simplified calculation step, since the method essentially reduces the considerations to two dimensions.

[0016] Finally, for at least one step k corresponding to an indentation image, but preferably for many or all of the indentation images corresponding to step k, the force F acting on a given body position is calculated by summing the products of the stress values ​​determined for the indentation pixels and the areas Aij of the assigned fields Fij of the mesh grid. k This is calculated using the formula F k =Σ i,j σ(Pij k ) × Aij, where σ(Pij k ) represent the stresses preset by the pixel values ​​Pw by the respective stress-deformation characteristic curves at the pixels Pij and thus at the fields Fij. Finally, a collision assessment can then be carried out based on the result of one of the aforementioned method steps, in particular on the acting forces calculated here in step k, and / or on the force-deformation characteristic curves derived from the calculated acting forces, and / or on similar results, for example on the energies derived from the forces or the force-deformation characteristic curves, in particular on the maximum permissible kinetic energy for the object or robot.

[0017] The force-deformation characteristic curve and the physical model of the robot can then be used to calculate the collisions during the observation. The calculations are then repeated with varying initial speed until the calculated contact force corresponds to the limit value applied to the collision. As iteration method, preferably a bisection method is used. The initial speed calculated by the iteration corresponds to a speed limit that the robot must not exceed. The speed limit can be selectively used according to these calculations to adjust the robot's movement speed (diagram 1) or to parameterize a monitoring device or a collision evaluation device for a model-based safety evaluation of the robotic device (diagram 2), which switches off the robot as soon as the speed limit is exceeded. [Table 1] [Table 2]

[0018] The method proposed here can calculate biomechanical force-deformation characteristic curves or similar quantification of collisions, similar to FEM. In contrast to FEM, the method proposed here can significantly reduce the calculation time for the quantification assessment of collisions, e.g. to calculate force-deformation characteristic curves for further processing steps or to calculate only individual limit values ​​for the resulting forces or energies, as will be explained further below. Due to its efficiency, it is particularly suitable for a model-based, i.e. quantitative, assessment of the risks of collisions and crushing during human-cobot collaboration.

[0019] The method proposed here therefore enables, for example, a crash assessment device for controlling a robot to quickly calculate the deformation of soft tissues at a body position that occurs when an object of any shape hits it. In contrast to the well-known approach of the finite element method, the method proposed here uses a geometric shell model that is realized by a mesh grid that is as regular as possible, consisting of square planar elements. By using square mesh elements for the body parts and thus restricting to one-dimensional curvatures, the model surface can be converted into a two-dimensional image with only a small amount of calculations. When the shell model is deformed by a three-dimensional object, the deformed positions are represented as pixels in the two-dimensional indentation image, each pixel value corresponding to a deformation value at the relevant position on the shell model, which also indicates the displacement of the relevant mesh element with respect to the initial state. Since each mesh element is associated with a specific, preferably individual, stress-deformation characteristic curve, it is possible to decompose according to the discrete deformation steps and calculate the contact forces from the deformation values ​​indicated by the pixels. These contact forces can already be used for the assessment of the crash. In combination with the position of the robot, the calculated forces correspond to points in the force-deformation characteristic curve that are sought for downstream calculations. Thus, by iteratively using this method, an entire characteristic curve can be calculated for each position step of the robot, which corresponds to a stepwise displacement of the object in the virtual space, and can then be used for further assessment of collisions with the physical robot model.

[0020] The method may comprise the steps of g) creating a force-deformation characteristic curve by mapping the calculated forces from method step f) (calculating the force F acting on a given body position) onto the discrete displacement paths used in method step d) (displacing the 3D model and the shell-mesh grid model relative to each other in a stepwise manner); h) evaluating the collision by the force-deformation characteristic curve and the physical model of the object, in particular the robotic device, upon a change in the initial velocity of the object, in particular the robotic device, by a dichotomy up to the initial velocity of the object relative to the collision force corresponding to a limit value applicable for this collision; and i) transmitting the calculated initial velocity to a robot control for adjusting or monitoring the robot velocity, preferably adjusting and / or monitoring the robot velocity in response to the transmitted initial velocity.

[0021] Correspondingly, in an advantageous embodiment, it is provided to determine for a number of indentation images, in particular for all indentation images, the respective stress values ​​for the indentation pixels Pij, to calculate the force F acting on a given body position for a number of steps k corresponding to the respective indentation images, in particular for all steps k, and to generate a force-deformation characteristic curve based on the determined forces for a given collision between a given object and a given body position. Then, preferably based on the generated force-deformation characteristic curve, an evaluation of the collision can also be performed, preferably with respect to the movement of the object underlying the collision, by the specification of an allowable, i.e. a predetermined maximum energy and / or a maximum allowable deformation and / or a maximum allowable force. The evaluation can provide as an evaluation result a limit value for a speed that the object or the robot must not exceed, in particular a robot speed. Correspondingly, as part of the method, a control or monitoring of the robot device can be performed based on the evaluation result and the limit value for the speed. The energy transferred from the object to the body position, which must not exceed the maximum allowable energy, is given by the integral of the force-deformation characteristic curve from zero to the corresponding deformation (distance) value. If the collision is evaluated with the specification of a maximum allowable deformation, the generation and use of the force-deformation characteristic curve is not necessary.

[0022] In a further advantageous embodiment, it is provided that before the determination of the respective stress value for the indentation pixel Pij, a filtering of one or more generated (original) indentation images is performed. This filtering can be performed in particular using at least image-based filters known from image processing and / or filtering based on machine learning methods. Correspondingly, the determination of the respective stress value and the calculation of the acting forces are then performed based on the filtered indentation image, i.e. on the pixel values ​​Pw of the one or more filtered indentation images.

[0023] Filtering has the advantage that it allows to take into account the deformation of mesh elements or fields adjacent to the field that is actually deformed by direct contact with the object. Even relatively simple filters, such as blurring filters, have been found to represent or mimic well the surroundings of the fields on the body surface that are deformed by direct contact. Thus, the filter can create a more realistic deformation image, as would occur in reality if an object were to act on the body part in a collision under observation. Due to the geometry of the shell-mesh grid model and the appropriate selection of the parameters of the filters, especially the image-based filters, it is now possible to well mimic the natural deformation patterns of the filtered indentation image. Here, the corresponding parameters for the selected filters can be empirically determined.

[0024] That is, the purpose of the filtering is to smooth the previously obtained indentation image so that the indentation image returned to the shell model after smoothing reproduces an image of a realistic and smooth deformation of the skin as typically occurs in humans at the contact area under observation. Thus, each pixel value in the filtered indentation image corresponds to a realistic displacement of the associated mesh element Fij, and thus, when returned to the shell model, a realistic 3D image of the skin deformation is obtained.

[0025] For example, the filtering can be performed in two steps. In this embodiment, in a first step, the individual pixels are filtered in separate filtering processes, and then in a second step, the results of the separate filtering processes are combined to generate a filtered indentation image. Here, in this case, firstly for each pixel to be filtered, i.e. for each pixel having a non-zero pixel value Pw within the displacement step k, a copy of the relevant indentation image is generated, in which the pixel values ​​Pw of all other pixels Pij are set to zero in the copy of the indentation image. Then, in each copy of the indentation image, a corresponding filter, for example a blur filter, is applied to each remaining one pixel having a non-zero pixel value Pw. Here, an imitation of an impact on an adjacent body surface is performed by a respectively selected contact point in the form of a pixel having a non-zero pixel value Pw. Finally, from the copy of the filtered indentation image, i.e. the filtered copy, a filtered indentation image is generated. For this, a maximum value Pw is selected as the value Pw of each pixel Pij of the filtered indentation image, and this maximum value Pw occurs for all pixels Pij assigned to the same field Fij in all copies of the filtered indentation image. In a second step of filtering, the value Pw of each pixel Pij in the filtered indentation image is compared with the value Pw of the corresponding pixel in the original indentation image, and the value Pw of each pixel in the filtered indentation image is set to zero if the two values ​​differ from each other by more than a predetermined tolerance, and is left at the value Pw if the two values ​​differ from each other by the tolerance or less than a predetermined tolerance, i.e. are the same or substantially the same. This filtered indentation image is then used in a subsequent step instead of the original indentation image.

[0026] Thus, here, at filter level 1, the individual pixel values ​​in the indentation image are smoothed in a separate filtering process. The copying or filtering step is performed for all pixels in the indentation image with value >0, so that if there are n pixels with value >0, there are also n copied or filtered indentation images, which contain the filter results of the individual filtered pixels. Then, from all n filtered indentation images, i.e. intermediate images, the pixel at position i, j with the highest pixel value is taken over to the filtered indentation image. Then, if the value of pixel Pij in the filtered indentation image corresponds to the value of pixel Pij in the original indentation image, it can be assumed that the hit object really touches the human body position at this position. At positions where there is no contact, the pixel in the result image gets the value 0, since the associated pixel in the filtered indentation image does not fulfill the condition.

[0027] The above filtering therefore has the advantage that the quantification of collisions is more precise and in particular an overestimation of the forces occurring is prevented, which allows, for example, a correspondingly controlled robotic device to operate at higher working speeds and still be safe.

[0028] In a further advantageous embodiment, it is provided that an additional factor matrix image with pixel values ​​Vij is generated, where an edge detection is performed on the original indentation image and the additional factor matrix image is adapted to have pixel values ​​>1 at edge locations and pixel values ​​= 1 at non-edge locations. This can be done, for example, by adding the value 1 to each pixel value in the result image of the edge detection. When determining the respective stress value for the indentation pixel Pij, or mathematically equivalently, when calculating the force F acting on a given body position, i.e. before the assessment of the impact is performed, the determined stress value for the indentation pixel is multiplied with the respectively assigned pixel value Vij of the additional factor matrix image. Correspondingly, the force calculated for step k is calculated according to the formula F k =Σ i,j Vij k×σ(Pij k ) × Aij.

[0029] This has the advantage that the localization effects of objects with hard edges are taken into account, since typically higher mechanical stresses act on body parts at the edges, which allows the mechanical stresses in the edge regions determined in the remaining method steps to be subsequently increased, generating a more realistic assessment.

[0030] In another advantageous embodiment, it is provided that a further factor matrix image with pixel values ​​Wij is generated, and the determined stress values ​​for the indentation pixels Pij are multiplied with the respectively assigned pixel values ​​Wij of the further factor matrix image, after which an assessment of the impact takes place. Based on the value Pw of the indentation pixels Pij respectively assigned to the pixel values ​​Wij and on the basis of the preset impact velocity for the impact, the respective strain rates for the indentation pixels Pij and thus for the assigned fields Fij are determined, and based on the respective strain rates, the pixel values ​​Wij are determined with the strain rate characteristic curves preset for the fields Fij assigned to the respective indentation pixels Pij, whereby the further factor matrix image is in particular generated. The values ​​of the strain rate characteristic curves and thus the pixel values ​​Wij are here ≧1. Here, the further factor matrix image, like the additional factor matrix image, is respectively individually preset for one, several or all steps k. The strain rate characteristic curves of the fields Fij may be identical or may be individually preset for each field Fij or group of fields Fij. Correspondingly, the force calculated for the step k is therefore calculated according to the formula F k =Σ i,j Wij k ×σ(Pij k ) × Aij, or if further factor matrix images are applied in combination with additional factor matrix images, the formula F k =Σ i,j Wij k ×Vij k ×Aij.

[0031] This has the advantage that the strain rate dependence of the sensitivity of biological tissues, i.e. the dependence of the injury risk on the impact velocity of the impacting object, is simulated and therefore a more accurate assessment of the impact is made using an additional factor matrix that depends on the impact velocity.

[0032] In a further advantageous embodiment, it is proposed to calculate the respective impression depth, i.e. the pixel value Pw of the pixel Pij assigned to the field Fij of the impression image, by setting a straight line parallel to the normal vector through the midpoint of each field Fij, determining whether the straight line hits the surface of the object, and if so, determining the distance between the intersection point of the straight line with the surface of the object and the midpoint of each field Fij as the impression depth if it corresponds to the impression of the object on the organism (for example, in a suitable definition, this distance is less than 0). One such suitable definition is, for example, to select the normal vector so that it points outward relative to the body part. However, if the definition of the direction is reversed, for example, a positive distance corresponds to the impression of the object on the organism. This definition of the impression depth has been found to be advantageous in terms of the amount of calculations and the accuracy of the achievable results.

[0033] In a further advantageous embodiment, it is contemplated that the sides of each field Fij of the mesh grid are of the same length or a natural multiple of the same length, i.e. the size of each field is a natural multiple of the unit size, but not necessarily the same.

[0034] An aspect also relates to a method for controlling a robotic device, in which collisions are calculated, preferably using a force-deformation characteristic curve, the force-deformation characteristic curve having been previously determined using a method according to one of the described embodiments, and the results from the calculation are then used for controlling or monitoring the robotic device, i.e. the robotic device is controlled depending on the evaluation result of the method.

[0035] A further aspect relates to a collision assessment device for aligning in a virtual space a provided 3D model of a given object and a provided shell-mesh grid model of a given body position, the relative arrangement of the two models corresponding to the relative arrangement of the object and the body position at a given collision with each other, each field Fij of the mesh grid of the shell-mesh grid model being square, and a stress-deformation characteristic curve being predefined for each field Fij of the mesh grid. The collision assessment device is further configured to stepwise displace the 3D model and the shell-mesh grid model with respect to each other in the virtual space according to a collision direction determined by the given collision, and for each step k of the stepwise displacement an indentation image with indentation pixels Pij is generated, a respective pixel value Pw of the indentation pixels Pij representing an indentation depth of the 3D model in a field Fij of the mesh grid of the shell-mesh grid model assigned to the respective indentation pixel Pij. Preferably, here exactly one field Fij is assigned to each indentation pixel Pij and vice versa. The stepwise displacement corresponds to a movement of the object, e.g. a robotic device, at the collision point under observation. The motion is present as an initial condition and thus specifies the impact direction and / or impact velocity. In particular under the effect of an increase in the contact force, the motion of the object may change. The change in motion due to the (contact) force calculated by the described method or the described device may be, but does not necessarily have to be, taken into account iteratively in the method or device.

[0036] Furthermore, the collision assessment device is also configured to determine, for the or at least one indentation image, in particular a plurality of indentation images, a respective stress value for an indentation pixel Pij according to the stress-deformation characteristic curve assigned to the corresponding field Fij and the respective pixel value Pw, and further configured to calculate, for at least one step k corresponding to the indentation image, at least one force F acting on a given body position by summing up the products of the stress values ​​determined for the indentation pixel Pij and the area Aij of the assigned field Fij of the mesh grid.

[0037] Further aspects also relate to a robotic device equipped with such a collision assessment device and / or a control device for a robotic device equipped with such a collision assessment device and configured to control the robotic device depending on the assessment results, in particular depending on the calculated forces acting on a given body position and / or depending on the maximum allowable energy and / or depending on the maximum allowable deformation and / or depending on the maximum allowable force.

[0038] Here, the advantages and advantageous embodiments of the crash assessment device correspond to the advantages and advantageous embodiments of the described method and vice versa.

[0039] The features and feature combinations described herein above in the introduction, as well as the features and feature combinations described below in the description of the figures and / or shown only in the figures, can be used not only in the respective combinations presented, but also in other combinations, without departing from the scope of the present invention. Thus, embodiments that are not explicitly shown or described in the figures, but which can be produced by combining separate features from the described embodiments, should also be considered as included and disclosed in the present invention. Also embodiments and feature combinations that do not have all the features of the original independent claims described should also be considered as disclosed. Moreover, also embodiments and feature combinations that go beyond or deviate from the feature combinations described in the claims, especially according to the above embodiments, should also be considered as disclosed.

[0040] The subject matter according to the invention will now be explained in more detail on the basis of the following figures, without being limited to the particular embodiments shown therein. [Brief description of the drawings]

[0041] [Figure 1] FIG. 1 illustrates an exemplary force-deformation characteristic curve. [Diagram 2]FIG. 2 illustrates an exemplary situation regarding a collision between a 3D model of an object and a shell-mesh grid model of a given body position, aligned in virtual space corresponding to the collision. [Diagram 3] FIG. 13 illustrates exemplary steps of filtering an indentation image. [Figure 4] FIG. 13 illustrates exemplary steps of filtering an indentation image. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0042] In the drawings, identical and functionally identical elements are marked with like reference numbers.

[0043] 1 shows an exemplary profile of the force-deformation characteristic curve f, here determined for a forearm muscle under the action of a cylinder with a diameter of 25 mm, where the applied force F (in Newtons) is plotted against the tissue deformation D (in millimeters). Typically, the force-deformation characteristic curve F is exponential up to a limiting value (here a deformation of about 13 mm) and then linear.

[0044] Fig. 2 shows an exemplary arrangement of a 3D model of an object and a shell-mesh grid model of a given body position. Here, the object 1 is represented by a pyramidal 3D model with a vertex at a contact point B on a given body position 2, which is represented by a (planar) shell-mesh grid model. In the illustrated example, the body position 2 is the palm of the hand. The shell-mesh grid model has a number of fields Fij in its mesh grid, which are square and here of equal size. Upon collision, the object 1 moves towards the body position 2 in the collision direction K.

[0045] Each field Fij is assigned here a stress-deformation characteristic curve s, which assigns a respective stress s to a given deformation D. Furthermore, in the illustrated example, a strain-rate characteristic curve d is also assigned for each field, which assigns a value W to a given strain rate D / v, where D represents the deformation and v represents the speed at which the deformation D occurs. The strain rate can also be pre-set in another way, for example ε'=d / dt L(t) / L0, where L0 is the initial length of the material and L(t) is the length at the time t. Naturally, the characteristic curve must be pre-set consistently for the given strain rate. The strain-rate characteristic curve allows the strain-rate dependence of the sensitivity of the biological tissue to be simulated as a factor matrix image.

[0046] 3 shows a first step of an exemplary filtering of the generated indentation print image. The indentation image AB, here with 6x6 pixels Pij, has here, by way of example, four pixels Pij with non-zero values ​​Pw. In the illustrated example, these are the four pixels P33 "O", P34 "▽", P43 "□" and P44 "△". Here, each of the four pixels has a distinct shade representing the respective pixel value Pw, with darker shades corresponding to larger pixel values ​​Pw and therefore larger deformations.

[0047] In the first step of filtering, each individual pixel Pij (here pixels P33, P34, P43 and P44) is filtered in a separate filtering process. For this, for each pixel P33, P34, P43 and P44 to be filtered, a corresponding copy AB-33, AB-34, AB-43, AB-44 of the relevant indentation image AB is generated, and the pixel values ​​of all other pixels Pij are set to zero. Then, in each copy AB-33, AB-34, AB-43, AB-44, a corresponding filter, here a blurring filter, is applied to the remaining pixels Pij having a non-zero value Pw. Then, from the filtered copies AB-33', AB-34', AB-43', AB-44', a filtered indentation image AB' is generated. This is done by selecting as value Pw of each pixel Pij of the filtered impression image AB' the maximum value Pw occurring for all pixels Pij assigned to the same field Fij (i.e. here all pixels Pij with the same indexes i, j) in all filtered copies AB-33', AB-34', AB-43', AB-44' for pixels Pij with the same indexes i, j. Thus, for pixels Pij assigned to the same field Fij, the global maximum of pixel values ​​Pw is identified across the different filtered copies and only this global maximum is further considered for the corresponding pixel Pij. These maxima are represented by "○", "▽", "□" or "△" in the copies AB-33', AB-34', AB-43', AB-44', respectively.

[0048] Thus, in the illustrated example, the filtered impression image AB' contains pixel values ​​Pw from copy AB-33' for pixels P22, P23, pixel values ​​Pw from copy AB-34' for pixels P24, P25, P34, P35, pixel values ​​Pw from copy AB-43' for pixels P32, P33, P42, P43, P52, P53, and pixel values ​​Pw from copy AB-44' for pixels P44, P45, P54, P55.

[0049] 4, an exemplary second step of filtering is now shown, where the pixel value Pw of each pixel Pij in the filtered indentation image AB' is compared with the value Pw of the corresponding pixel Pij in the original indentation image AB to obtain the final filtered indentation image AB". Correspondingly, the filtered indentation image AB' from the first step can be referred to as the intermediate indentation image, and the filtered indentation image AB" from the second step can be referred to as the resultant indentation image.

[0050] For the resultant indentation image AB'', the value Pw of each pixel in the filtered indentation image AB' is set to zero if it differs from the value PW of the respective pixel in the original indentation image AB by more than a predetermined tolerance value, and is left at the value Pw if the two values ​​differ from each other by the tolerance value or by less than the predetermined tolerance value. Pixels whose pixel values ​​Pw of the indentation images AB and AB' differ from each other by the tolerance value or by less than the predetermined tolerance value (i.e. are substantially the same) are represented in FIG. 4 by a diamond "◇". Indeed, only in these pixels Pij (here pixels P34, P43, P44) or in the assigned field Fij, deformations must be due to a direct pressure of the object on the body part, and therefore these positions are important for assessing the collision, in particular with regard to the injury risk.

[0051] In the illustrated example, pixels P43, P43, P44 have pixel values ​​0.5, 1, 0.75, i.e. corresponding to deformations of 0.5 mm, 1 mm, 0.75 mm. Pixels Pij now have area Aij=1 cm 2 Since the contact force F is assigned to a field F i with a stress-strain characteristic curve s, the contact force F can be calculated by the stored stress-strain characteristic curve s. In the illustrated example, the force F = [s(0.5 mm) + s(0.75 mm) + s(1 mm)] N / cm 2 x 1cm 2 =[26+71+160]N / cm 2 x 1cm 2 would be 257N.

[0052] In FIG. 5, exemplary indentation images AB, AB', AB'' are shown for the collision also shown in FIG. 2, with a perspective view and a cross-sectional view of the object 1 submerged in the body position 2. In the indentation images AB, AB', AB'', again darker shading corresponds to a larger deformation. Starting from the original indentation image AB, here in a first step (1) similar to the first step of filtering shown in FIG. 3, an intermediate indentation image AB' is generated. In a second step (2) the result indentation image AB'' is generated accordingly. It is directly evident from the result indentation images, especially in a quantifiable way, that the deformations occurring at the edges of a pyramidal object when it collides with the palm of the hand are important with regard to the assessment of the collision and in particular the possible injury risk.

Claims

1. A method for evaluating a predetermined collision between a predetermined body position (2) of a living being and a predetermined object (1), in particular for evaluating a collision between a predetermined body position of a human and a predetermined part of a robotic device, comprising: a) a method step of providing a 3D model of the predetermined object (1) to a computing unit; b) a method step of providing a shell-mesh grid model of the predetermined body position (2) to a computing unit, wherein each field Fij of the mesh grid of the shell-mesh grid model is square, and stress-deformation characteristic curves (s) are preset for each field Fij of the mesh grid; c) a method step of using the computing unit to align the provided 3D model and the provided shell-mesh grid model in a virtual space, the relative arrangement of the two models corresponding to the relative arrangement of the object and the body position at the predetermined collision; d) a method step of displacing the 3D model and the shell-mesh grid model step by step relative to each other in the virtual space according to a collision direction (K) specified by the predetermined collision by the computing unit, and for each step k of the step-by-step displacement, an indentation image (AB) having indentation pixels Pij is generated, and each pixel value Pw of the indentation pixels Pij represents the indentation depth of the 3D model within the field Fij of the mesh grid of the shell-mesh grid model assigned to each of the indentation pixels Pij; e) a method step of determining, for the indentation image or at least one indentation image (AB, AB', AB'') with respect to the corresponding field Fij, the respective stress values for the indentation pixels Pij according to the stress-deformation characteristic curves (s) assigned to the corresponding fields Fij and the respective pixel values Pw; f) a method step of calculating a force F acting on the predetermined body position (2) by summing the product of the stress value determined for the indentation pixels Pij and the area Aij of the assigned field Fij of the mesh grid for the step k of the displacement corresponding to at least one of the indentation images (AB, AB', AB''); A method comprising the above steps.

2. In method step e), for a plurality of indentation images (AB, AB', AB''), in particular for all indentation images (AB, AB', AB''), respective stress values for the indentation pixels Pij are determined, and in method step f), for a plurality of steps k corresponding to each of the respective indentation images (AB, AB', AB''), in particular for all steps k, a method step of calculating a force F acting on the predetermined body position (2). A method step of generating a force-deformation characteristic curve (f) for the predetermined collision between the predetermined object (1) and the predetermined body position (2), and in particular Based on the generated force-deformation characteristic curve (f), preferably, a method step of evaluating the collision by defining a maximum allowable energy or a maximum allowable force regarding the movement of the object (1) underlying the collision The method according to claim 1, characterized in that

3. Before determining each of the stress values for the indentation pixels Pij in method step e), filtering of the indentation image (AB) or the generated indentation image (AB) is performed, and the filtering is performed using, in particular, at least one image-based filter and / or a filter based on a machine learning method, and the method step e) of determining and the method step f) of calculating are performed based on the filtered indentation images (AB', AB''). The method according to claim 1, characterized in that

4. In the filtering in the first step, individual pixels are filtered in separate filtering processes. First For each pixel to be filtered, a copy (AB-33, AB-34, AB-43, AB-44) of the relevant indentation image (AB) from method step d) is generated, and the pixel values Pw of all other pixels Pij are set to zero. Then In each copy of the indentation image, a corresponding filter, in particular a blur filter, is applied to each pixel. Finally From the copies (AB-33', AB-34', AB-43', AB-44') of the indentation image to which the filter has been applied, a filtered indentation image is created, which is done by selecting the maximum value Pw as the value Pw of each pixel Pij of the filtered indentation image (AB'), and the maximum value Pw occurs for all pixels Pij assigned to the same field Fij within all copies of the indentation image of the pixel Pij to which the filter has been applied. During the filtering, in a second step, the value Pw of each pixel Pij in the filtered indentation image (AB') is compared with the value Pw of the corresponding pixel Pij in the indentation image (AB) from method step d), and the value Pw of each of the pixels in the filtered indentation image (AB'') is set to zero if the two values differ from each other by more than a predetermined tolerance value, and remains the value Pw if the two values differ from each other by only the tolerance value or less than the predetermined tolerance value. The method according to claim 3, characterized in that.

5. Edge detection is performed on the indentation image (AB) from method step d), and an additional factor matrix image is adapted to have a pixel value greater than 1 at the edge position and a pixel value of 1 at the non-edge position, and the determined stress value for the indentation pixel Pij is multiplied by the pixel value Vij assigned to each of the additional factor matrix images, thereby generating an additional factor matrix image having the pixel value Vij. The method according to claim 1, characterized in that.

6. A further factor matrix image having pixel values Wij is generated, and the determined stress value for the indentation pixel Pij is multiplied by the pixel values Wij assigned to each of the further factor matrix images, where the strain rate is determined based on the value Pw of the indentation pixel Pij assigned to each of the pixel values Wij and a predetermined impact velocity, and based on each of the strain rates, the pixel values Wij are determined using a preset strain rate characteristic curve (d) for the field Fij assigned to each of the indentation pixels Pij, thereby generating the further factor matrix image in particular. The method according to claim 1, characterized in that.

7. A straight line is set parallel to the normal vector through the midpoint of each field Fij, and it is determined whether the straight line hits the surface of the object (1). If it hits, the distance between the intersection point of the straight line and the surface of the object (1) and the midpoint of each field Fij is determined as the indentation depth when it corresponds to the indentation of the object (1) into the living being, whereby each indentation depth is calculated. The method according to claim 1, characterized in that.

8. The method according to claim 1, characterized in that after alignment, at at least one contact point (B) of the two models, the distance between the two models is zero.

9. The method according to claim 1, characterized in that the sides of each field Fij of the mesh grid are of the same length or a natural number multiple of the same length.

10. A method for controlling a robot device, wherein a collision is evaluated using the method according to any one of claims 1 to 8, and the robot device is controlled according to the evaluation result of the method.

11. a) In a virtual space, align a provided 3D model of a predetermined object (1) with a provided shell-mesh grid model of a predetermined body position (2), such that the relative arrangement of the two models corresponds to the relative arrangement of the object (1) and the body position (2) at a predetermined collision with each other, each field Fij of the mesh grid of the shell-mesh grid model is square, and a stress-deformation characteristic curve is preset for each field Fij of the mesh grid. d) In the virtual space, displace the 3D model and the shell-mesh grid model step by step relative to each other according to the collision direction (K) specified by the predetermined collision. For each step k of the stepwise displacement, an indentation image (AB) having indentation pixels Pij is generated, and each pixel value Pw of the indentation pixels Pij represents the indentation depth of the 3D model within the field Fij of the mesh grid of the shell-mesh grid model assigned to each of the indentation pixels Pij. e) According to the stress - strain characteristic curve(s) assigned to the corresponding field Fij and each of the pixel values Pw, for the indentation image or at least one indentation image (AB, AB', AB''), determine the respective stress values for the indentation pixel Pij, f) For step k corresponding to at least one of the indentation images (AB, AB', AB''), calculate the force F acting on the predetermined body position (2) by summing the product of the stress value determined for the indentation pixel Pij and the area Aij of the assigned field Fij of the mesh grid A collision assessment device configured as such.

Images