Method and arrangement for compensating non-geometric error influences on a robot absolute accuracy by means of a laser sensor system

Abstract

Method for compensating non-geometric error influences on an absolute accuracy of a robot (110) by means of a laser sensor system, wherein the robot (110) comprises a plurality of elastic elements (111, 112, 113, 114, 115, 116, 117) and a control unit (140), wherein an elastic element (111, 112, 113, 114, 115, 116, 117) is a rigid body (111, 112, 113) or a joint (114, 115, 116) or an effector (117) or a robot base (120), wherein at least one radiation pattern generator (131, 132) is arranged stationary in an environment of the robot (110) within a workspace (200) or outside the workspace (200), wherein at least one radiation pattern is radiated through the working space (200) of the robot (110) by means of the at least one radiation pattern generator (131, 132), wherein the at least one radiation pattern comprises at least one laser light beam and / or at least one laser light plane, wherein at least one sensor (130) with at least one light-sensitive surface is arranged on the effector (117) of the robot (110), wherein the robot (110) is successively controlled by the control unit (140) into a plurality of measuring configurations in which the at least one radiation pattern impinges on the at least one light-sensitive surface, wherein the robot (110) is controlled in accordance with robot structural information stored electronically in the control unit (140), wherein a position of a projection of the at least one radiation pattern onto the at least one light-sensitive surface is detected by the at least one sensor (130) and measurement information describing the position is forwarded from the at least one sensor (130) to a computing unit (150), wherein torques act on the elastic elements (111, 112, 113, 114, 115, 116, 117, 120) due to an external force depending on a respective joint configuration and where the robot structural information is inaccurate due to non-geometric error influences, characterized by that the at least one radiation pattern is radiated through the working space (200) and the plurality of measuring configurations (I, II, III, IV) is selected such that for at least one elasticity element (111, 112, 113, 114, 115, 116, 117, 120) from the plurality of elasticity elements (111, 112, 113, 114, 115, 116, 117, 120) there is at least one pair of measuring configurations (I, II, III, IV) for which the absolute value of the difference between the torques of the at least one pair of measuring configurations on the at least one elasticity element (111, 112, 113, 114, 115, 116, 117, 120) is greater than a target value, that on the light-sensitive surface of the at least one sensor (130) a deviation from a straight line and / or plane implicitly predetermined by the at least one radiation pattern and its beam direction as well as its radiation pattern orientation is taken into account, by implicitly comparing the position of the projection detected by at least one sensor (130) for each of the plurality of measurement configurations (I, II, III, IV) with a position of the projection determined on the basis of the faulty robot structure information, and that corrected robot structural information is generated to compensate for the non-geometric error influences from the deviation.

Description

[0001] The invention relates to a method for compensating non-geometric error influences on a robot's absolute accuracy by means of a laser sensor system according to the preamble of claim 1 and a corresponding arrangement.

[0002] In the current state of the art, the term "mechanism" encompasses various types of robots, particularly industrial robots. These mechanisms are generally universally programmable machines that operate largely autonomously within a given framework, for example, for handling, assembling, or processing workpieces. With appropriate programming, a mechanism is capable of consistently and autonomously repeating a given workflow.

[0003] To better understand the following explanations, some basic terms are first defined: 1. Robot: A robot is a system of so-called rigid bodies connected to one another by rotary, sliding, or screw joints. The term "robot" within the meaning of the invention is defined broadly here (namely in the sense of the term "mechanism" commonly used in the literature) and includes, in particular, industrial robots, gantries, machine tools, hexapods, and mechanical frameworks. The inventive method described below is suitable and intended for use with all forms of robots in this sense. For the sake of clarity, however, it will be illustrated below only using so-called kinematic chains. 2. Effector: An effector is a possibly complex, rigid or movable element of a robot to which a tool (e.g., a gripper for gripping a workpiece, a milling cutter, a drill, a sensor such as a camera, etc.) can be attached to perform a specified activity. The so-called TCP (tool center point) is a freely determined point of action of a tool or workpiece attached to the effector, for example, the focus of a picked-up laser or the center of a held object. The effector is occasionally referred to in the literature as a "robot hand." 3. Position: The term position refers to the position and orientation of an object in three-dimensional space. 4. Joint configuration: The joint configuration is the totality of all control values of the joints of a mechanism or robot, which determines the position of all rigid bodies relative to each other, including the effector. 5. Radiation pattern generator: A radiation pattern generator generates directed laser beams or directed laser radiation patterns such as single beams or bundles of isolated single beams, line or cross-shaped radiation patterns or any other patterns. 6. Sensor: A sensor can register the position and, depending on its specific design, also the orientation of a radiation pattern striking the sensor relative to a coordinate system permanently assigned to the sensor. For this purpose, the sensor has at least one light-sensitive surface. For example, the sensor can have a light-sensitive surface on each of two or more sides. 7. Calibration object: A calibration object is a generic term for a single sensor and a single radiation pattern, including a single associated radiation pattern generator. Connected images of radiation patterns on a sensor surface of the radiation pattern generator, such as points, lines, or crosses, are considered a single calibration object. Disjointed radiation patterns generated by a radiation pattern generator, e.g., using splitting optics, are considered multiple different calibration objects. 8. Calibration object pair: A calibration object pair is defined as a coherent radiation pattern including the associated radiation pattern generator with a sensor, i.e. it consists of two so-called complementary calibration objects. 9. Laser sensor system: A laser sensor system is a calibration system for various types of robots, based on the following principle: One calibration object of a pair of calibration objects is mounted on the effector and is referred to below as the effector object. The other calibration object of the pair of calibration objects is positioned stationary in or near the workspace and is referred to below as the reference object. The robot moves the effector object into a variety of positions in which at least one radiation pattern from the radiation pattern generator strikes the sensor. The sensor forwards the detected measured values, i.e. the position of the radiation pattern relative to the sensor coordinate system, to a computing unit, which derives global correction information for the specific, individual, usually slightly deformed robot specimen from the measured values and the associated joint configurations.Each radiation pattern generator can irradiate different sensors during a robot calibration, and each sensor can be irradiated by different radiation pattern generators. 10. Workspace: The workspace is the space or the set of positions that the robot can reach. 11. Robot absolute accuracy: Robot absolute accuracy indicates the deviation between a specified target position of the effector and the position actually reached by the calibrated or uncalibrated robot controller, or between a specified target trajectory of the effector within the robot's workspace and the actual trajectory traveled. Robot absolute accuracy can be improved by calibrating the robot. 12. Mathematical model: A robot consisting of joints and rigid segments can be described by a mathematical model that contains the robot parameters. 13. Robot parameters: Robot parameters are divided into geometric (usually so-called "Denavit-Hartenberg" or "Hayati parameters") and non-geometric robot parameters. Geometric parameters specify the constant lengths and angles that make up the kinematic structure of the robot. Non-geometric robot parameters specify, for example, joint elasticity or beam elasticity. The more precisely the values of the robot parameters of a given robot are known, the more precisely it can be controlled using the mathematical model. The totality of all parameters involved in the mathematical model of robot calibration is referred to as (calibration) model parameters. 14. Model-based mathematical parameter identification: In principle, various methods for mathematical, i.e., computational, parameter identification are known in the state of the art, for example, well-known methods of nonlinear optimization or the so-called (iterative) regression calculation. The goal of parameter identification is to determine the actual parameters of the robot so that the residual error, or the residue, is as small as possible. The residue is defined by the deviation between the effector positions calculated using the mathematical model (i.e., the respective target position), and the actual effector positions (i.e., the respective actual position). 15. General terms: In general, when no model is used, the model parameters are called global correction information, the robot parameters are called robot structure information, and the calibration object parameters are called the specification of the posture of the calibration objects. 16. The global correction information allows the correct effector position to be precisely calculated for each joint position of the robot, and conversely, all exact joint configurations for a given effector position. If the calibration procedure is not model-based but rather based on artificial intelligence, the global correction information is contained in the weighting matrices, which essentially comprise the information of the mathematical model without the model parameters being explicitly included in the matrices.

[0004] It is also known in the state of the art that for a successful robot calibration, in addition to the kinematic structure of the robot itself, the following elements must also be taken into account: 1) the position of all stationary reference objects in the workspace or relative to the robot base, e.g. the position of radiation pattern generators and sensors as well as 2) the position of all effector objects relative to the robot flange or relative to the last joint of the robot in front of the effector, e.g. a position of a reflector matching the laser tracker, a sphere matching the measuring pot, a sensor matching the radiation pattern generator or a radiation pattern generator matching the sensor.

[0005] To ensure that robots can be controlled and operated as precisely as possible in their workspace at all times, they are calibrated when accuracy decreases, i.e. all influences on the accuracy of the effector position are identified as precisely as possible and compensated by the robot control system.

[0006] Furthermore, it is known to perform indirect measurements for robot calibration using a laser tracker, or "tracking interferometer," as already mentioned. This is a measuring device that can record the 3D coordinates of object points with high precision. A laser tracker consists of an interferometer whose laser beam can automatically follow a reflector. Likewise, theodolite systems consisting of several individual theodolites can also be used to calibrate the robot. Laser trackers and theodolite systems are referred to as "global measurement systems" because they can record and measure every effector position in the robot's workspace in at least three dimensions. In contrast, so-called "local measurement methods" are also known, which - in contrast to global measurement methods - can, by design, only record measurements in a locally limited (usually low-dimensional) part of the workspace, e.g.They can only record measurement points along one or more straight lines or planes, such as laser lines, on a compact reference object in the workspace, or they can only perform measurements on a spherical reference object or within a compact measuring chamber, or they can only perform distance measurements using an extendable cable. The resulting absolute accuracy of calibrations based on such local measurement methods is usually subject to limitations and is often inferior to global measurement methods; however, the required equipment is significantly more cost-effective. Local measurement methods differ from one another in the measuring devices used.

[0007] In this context, FR 2 696 969 A1 discloses a calibration method in which a laser beam source is attached to the hand of a robot to be calibrated, and a measurement plane near the robot is used as a reference. The robot moves to a series of unspecified hand positions in which the laser beam strikes the measurement plane. The coordinates of the impact points on the measurement plane are identified, and from these, together with the associated joint configurations, the estimated values for the position of the measurement plane, and the position of the laser beam source relative to the hand, the robot parameters are determined using mathematical parameter identification.

[0008] WO96 / 30171 A1 describes a local measurement method and a calibration device for calibrating the motion axes of industrial robots. The calibration device consists of a calibration beam, e.g., a laser beam, in the robot's workspace and an associated interruption detector. The interruption detector, e.g., a sphere with a known radius, is mounted on the robot hand. The robot then moves to a series of unspecified hand positions in which the calibration beam is interrupted by the sphere. The calibration parameters are then calculated from the associated joint configurations, the estimated robot position, and the position of the laser relative to the hand.

[0009] DE 10 2012 016 106 A1 discloses an arrangement for model-based local calibration of a robot in a workspace using directed laser radiation patterns. The laser radiation patterns are generated by laser radiation pattern generators and detected by radiation pattern position sensors. The laser radiation pattern generators and radiation pattern position sensors interact in such a way that, when a laser radiation pattern strikes the radiation pattern position sensor, measured values containing position information are forwarded to computing devices, which use these measured values to determine the parameters of the mathematical robot model. At least two groups of laser radiation pattern generators and radiation pattern position sensors are provided.

[0010] WO 00 29 / 175 A1 discloses a device and a local measuring method for measuring mechanisms such as robots and their positions, which are used to recalibrate such mechanisms. The local measuring method described in WO 29 / 175 A1 is necessarily preceded by at least one global measurement taking into account the reference objects required for the subsequent local measuring method. For the local method, one or more effector objects are arranged on the robot and one or more reference objects are mounted stationary relative to the robot base. The position accuracy, i.e. parameters of the robot that influence the positioning and orientation accuracy, or the relative position of the effector and reference objects, are recorded using internal signal and information processing elements in the robot as well as external signal and information processing elements.The effector objects interact with the reference objects in defined positions. Upon detection of an interaction, the corresponding joint configuration of the robot is forwarded to an information processing system, which evaluates it. WO29 / 175 A1 uses a mathematical robot model in the form of a geometric system of equations to calculate the parameters. A mathematical parameter identification method is used to determine the geometric mechanism parameters, as well as non-geometric phenomena such as joint elasticities. Determining the non-geometric robot parameters is only possible here because a precise global measurement of the robot and the reference objects of the local measurement system was previously performed using a laser tracker.The reference objects of the local measuring system may no longer be moved after measurement by the global measuring system and remain stationary in the robot's workspace.

[0011] DE 10 2004 026 813 A1 relates to a method for controlling a plurality of handling devices such as multi-axis industrial robots. At least one handling device acts as a reference handling device and is moved to a number of predetermined poses within its workspace, at which internal position values are determined as first target poses. For each target pose, a first actual pose of the reference handling device is subsequently determined by an external measuring system. Subsequently, at least one further handling device moves to specific poses of the reference handling device as second target poses, wherein for each of these poses, an actual pose of the further handling device is determined by an external measuring system.Based on the actual-target deviations between the target and actual poses of the two handling devices determined in this way, a parameter model for the other handling device is then determined, in which both the handling device's own errors and the errors of the reference handling device can be compensated simultaneously.

[0012] EP 4 052 861 A1 discloses an automatic calibration method for a calibration device connected to a camera arranged on the end effector of a robot. First, a robot-based coordinate system and an image of a marker marked in the robot's workspace are acquired by the camera and a robot controller, and the acquired image and the robot-based coordinate system are recorded while the end effector is moved to a plurality of sample coordinates. Furthermore, the position of a marker based on the robot coordinate system is estimated using the acquired image and the robot-based coordinate system.

[0013] EP 3 974 779 A1 discloses a method for hand-eye calibration for robots. The method comprises a coordinate recording step for controlling a rear end of a robot arm to move it sequentially to at least three predetermined positions above a calibration plate. At each predetermined position, a laser provided on the robot arm is controlled to project onto the calibration plate, coordinates of an end point of the rear end of the robot arm are recorded in a robot coordinate system during the projection, and a camera at the rear end of the robot arm is controlled to photograph the projection on the calibration plate. Furthermore, the coordinates of the projection are recorded in the camera coordinate system.In a transformation matrix calculation step, a calibration transformation matrix is calculated according to the recorded coordinates of the at least three projections on the calibration plate in the camera coordinate system and the coordinates of the end point of the rear end of the robot arm in the robot coordinate system during each projection.

[0014] CN 1 09 900 251 A1 discloses a device and method for robot positioning based on visual technology. The device comprises an automatic scanning device with line-structured light, a robot module, and a higher-level computer. The automatic scanning device with line-structured light comprises an industrial camera, a laser, and a galvanometer. The robot module, in turn, comprises a robot body and a robot controller. The method enables the three-dimensional positioning of a target object in any position by an industrial robot and can be used for the robot's detection and transport tasks.

[0015] Furthermore, it is generally known that the global correction information of a mechanism can also be determined using artificial intelligence. In this case, the determination is not based on a mathematical model and without explicitly identifying the model parameters.

[0016] In the specific arrangements of the comparatively inexpensive, locally measuring systems used to date, some non-geometric robot parameters or robot structural information cannot be identified with sufficient precision due to conceptual, unavoidable, algebraic dependencies. These algebraic dependencies are fundamentally due to the existing local measurement principles, not to a lack of measurement accuracy or inadequacies in artificial intelligence or mathematical parameter identification. Therefore, the state of the art is not able to identify and compensate for certain variables due to fundamental obstacles. Therefore, some non-geometric robot parameters cannot be determined using existing local measurement methods.

[0017] Elastic deformations, in particular, have not yet been consistently captured by robot manufacturers and taken into account in robot controllers, and are usually only modeled very simply. Elasticities are a significant source of errors in robots. If they are not identified and taken into account in the controller, the achievable absolute accuracy of the robot is significantly compromised.

[0018] If non-geometric parameters are nevertheless determined using a local measurement method in the prior art, as described, for example, in WO29 / 175 A1, it is not only necessary to perform numerous measurements in different joint configurations and under different loads on a large number of effector objects and reference objects. Furthermore, it is a mandatory requirement that the precise position of the reference objects and the effector objects relative to each other be precisely measured in advance, so that in this case, too, a global measuring system must be used at least once. However, such laser trackers or theodolite systems are associated with high acquisition costs.

[0019] It is an object of the present invention to propose a comparatively cost-effective and locally measuring method by means of which, in particular, non-geometric parameters of a robot can be precisely identified.

[0020] This object is achieved according to the invention by the method for compensating non-geometric error influences on the absolute accuracy of a robot by means of a laser sensor system according to claim 1. Advantageous embodiments and further developments of the invention emerge from the dependent claims. The invention relates to a method for compensating non-geometric error influences on the absolute accuracy of a robot by means of a laser sensor system, wherein the robot comprises a plurality of elastic elements and a control unit, wherein an elastic element is a rigid body or a joint or an effector or a robot base, wherein at least one radiation pattern generator is arranged stationary in an environment of the robot within a workspace or outside the workspace, wherein at least one radiation pattern is radiated through the workspace of the robot by means of the at least one radiation pattern generator,wherein the at least one radiation pattern comprises at least one laser light beam and / or at least one laser light plane, wherein at least one sensor with at least one light-sensitive surface is arranged on the effector of the robot, wherein the robot is controlled by means of the control unit successively into a plurality of measurement configurations in which the at least one radiation pattern impinges on the at least one light-sensitive surface, wherein the robot is controlled in accordance with robot structural information electronically stored in the control unit, wherein a position of a projection of the at least one radiation pattern onto the light-sensitive surface is detected by the at least one sensor and measurement information describing the position is forwarded from the at least one sensor to a computing unit,where torques act on the elastic elements due to an external force depending on a respective joint configuration and where the robot structural information is subject to errors due to non-geometric error influences.

[0021] The method according to the invention is therefore carried out as a so-called laser sensor calibration, in which all measurement information required for calibration is recorded and processed locally on sensor surfaces.

[0022] In addition to a robot base, which is stationary in the workspace, the robot to be calibrated has a plurality of joints and rigid bodies, at least one effector, and a control unit. Typically, the robot has exactly one effector, but the method according to the invention is not limited to robots with only one effector.

[0023] The joints can take different forms, for example, as rotary, sliding, or screw joints. Furthermore, the joints preferably have electric actuators for moving the joints and the rigid bodies attached to them.

[0024] The rigid bodies represent rigid and immobile connecting bodies in the sense of arm segments between two joints.

[0025] The at least one effector is preferably arranged via a joint on the outermost rigid body, i.e., the rigid body furthest from the robot base, and can, for example, pick up or operate a tool, a gripper, or even a workpiece, in particular, pick up a workpiece by means of the gripper. For example, this makes it possible, among other things, for the robot to use the gripper to guide the workpiece to a stationary tool for processing.

[0026] Since the joints, the rigid bodies, the robot base and the effector have an inherent, unavoidable elasticity, they are summarized under the generic term “elasticity elements” in the sense of the invention.

[0027] Finally, the control unit can be connected to the remaining components of the robot exclusively at the data level, for example, via one or more data cables or wirelessly. A connection beyond this is possible, but not required. The control unit includes an electronic processing unit that enables real-time control of the robot by converting movement requests to the robot into corresponding joint control commands.

[0028] At least one radiation pattern generator is arranged in the robot's workspace, spaced apart from the robot base or spaced apart from the robot, and additionally or alternatively outside the robot's workspace. The at least one radiation pattern generator is stationary, i.e., fixed and immobile at least for the duration of the implementation of the method according to the invention. In contrast to the known generic methods, however, the invention does not require the exact position of the associated radiation pattern generator relative to the robot base, i.e., advantageously, no laser tracker or theodolite system is required to determine this position.

[0029] Advantageously, the at least one radiation pattern generator can remain in its assigned position in the workspace or outside the workspace even after the method according to the invention has been carried out, in order to be able to be reused for later, repeated implementations of the method according to the invention.

[0030] Preferably, several radiation pattern generators are arranged in the working space, for example two, three or four radiation pattern generators.

[0031] The at least one radiation pattern generator is advantageously designed as a laser, for example, a semiconductor laser diode or a gas laser, in particular a HeNe laser, with upstream optics for generating a specific radiation pattern. The optics can consist of one or more optical lenses or optical gratings.

[0032] The light emitted by the at least one radiation pattern generator then represents the generated and radiated radiation pattern. Within the meaning of the invention, it is not necessary for the radiation pattern to have a wavelength perceptible to the human eye. The wavelength can also be in the infrared or ultraviolet range.

[0033] At least one sensor with at least one light-sensitive surface is arranged on the effector. The sensor is therefore an optical sensor and can, for example, be a so-called CCD sensor or CMOS sensor, wherein the at least one light-sensitive surface advantageously represents the associated light-sensitive sensor array. The light-sensitive surface of the at least one sensor preferably has a matrix structure that not only detects the impact of the radiation pattern but also allows the shape and position of the projection of the radiation pattern to be determined.

[0034] However, the light-sensitive surface can also be essentially a one-dimensional line. The sensor would then be designed as a so-called line sensor. In this case, the radiation pattern is preferably a light plane.

[0035] The at least one sensor is configured to detect the position of the projection of the at least one radiation pattern as it strikes the at least one light-sensitive surface. Accordingly, the at least one light-sensitive surface is configured to detect all wavelengths of all radiation pattern generators used. The at least one sensor can be attached to the effector in any suitable manner.

[0036] The position of a projection on the light-sensitive surface is detected by the at least one sensor. Within the meaning of the invention, the "position of the projection" refers to the shape, orientation, and precise position of the at least one radiation pattern on the at least one light-sensitive surface. In the case of a non-point-shaped cross-section of the radiation pattern, the shape and orientation of the at least one radiation pattern on the at least one light-sensitive surface contain not only the position but also information about the orientation of the at least one sensor relative to the at least one radiation pattern generator. The orientation information can advantageously be used for calibration.For example, a projection of a square radiation pattern is displayed as a distorted rectangle on a light-sensitive surface if the radiation pattern does not impinge perpendicularly on the light-sensitive surface. Likewise, a circular radiation pattern becomes a distorted ellipse in this case. Based on the deformation of the projection compared to the radiation pattern generated by the at least one radiation pattern generator, an angle of inclination of the at least one sensor relative to the at least one radiation pattern generator can be determined. Based on a rotation of the projection compared to the radiation pattern generated by the at least one radiation pattern generator, an orientation of the at least one sensor relative to the at least one radiation pattern generator can additionally be determined.Instead of having a continuous cross-section, such as the aforementioned square or circular cross-section, the radiation pattern can also consist of two or more spaced-apart individual beams, thereby achieving essentially the same advantages. If the at least one sensor is designed as a substantially one-dimensional line sensor, the radiation pattern advantageously has a linear, one-dimensional cross-section, i.e., is designed as at least one light plane.

[0037] Measurement information from the at least one sensor, which describes the position of the projection on the light-sensitive surface, is then forwarded to an advantageously external processing unit. The measurement information preferably includes a position of the projection on the light-sensitive surface, a shape of the projection on the light-sensitive surface, and an orientation of the projection on the light-sensitive surface. The external processing unit can, for example, be a known computer with software designed to evaluate or process the measurement information. The external processing unit also has an electronic memory in which the robot's measurement configurations for the acquired measurement information are stored.

[0038] Preferably, multiple sensors are arranged on the effector, for example, two, three, or four sensors. Likewise, multiple radiation patterns from different radiation pattern generators can be detected synchronously, simultaneously, sequentially, or following any desired system, by a single sensor by moving the sensor into the respective radiation pattern. To facilitate differentiation between the different radiation patterns in this case, it is conceivable for different radiation patterns to have different shapes or cross-sections, or for the different radiation patterns to be switched.

[0039] The at least one radiation pattern comprises at least one laser light beam with, for example, a circular, rectangular, or oval cross-section. Any other cross-sectional geometries are also conceivable. The radiation pattern can be generated, for example, by a so-called diffractive optical element. The diffractive optical element can split a single beam into several separate, unconnected laser light beams, which, within the meaning of the invention, are understood as independent and distinct laser light beams. The laser light beam then preferably consists of individual light beams with no or only a very small expansion angle.Alternatively or additionally, the at least one radiation pattern can also be designed as a laser light plane, in which case the generated laser light beam is expanded in one dimension by means of the optics to form a laser light plane or is continuously pivoted in a plane by means of a suitable device, for example by means of an electric motor and a pivotable mirror, so that the laser light beam is constantly guided along the plane. In any case, the at least one radiation pattern generator is arranged in the workspace such that the at least one radiation pattern runs through the workspace of the robot, i.e. that the sensor on the effector of the robot can be steered into the radiation pattern so that the at least one radiation pattern strikes the sensor on the effector. The robot is advantageously capable of a large number of joint configurations orTo adopt measurement configurations in which at least one radiation pattern hits the sensor.

[0040] Preferably, the at least one radiation pattern generator generates more than one laser light beam, wherein the laser light beams are parallel to one another or have a specific, fixed angle to one another. For this purpose, the radiation pattern generator can, for example, comprise more than one light source, for example, two or more lasers, or comprise a diffractive optical element as described above.

[0041] When implementing the method according to the invention, the robot is controlled by the control unit into a plurality of joint configurations in which the at least one radiation pattern is detected by the at least one sensor, i.e., in which the at least one radiation pattern impinges on the light-sensitive surface. Such joint configurations are referred to as measurement configurations within the meaning of the invention.

[0042] Advantageously, the at least one sensor is moved into the radiation pattern multiple times in different measurement configurations and stopped there to acquire measured values, or moved continuously or in sections along the radiation pattern. Preferably, however, multiple measured values can also be acquired in identical measurement configurations, for example, to detect measurement errors.

[0043] Due to unavoidable external forces, such as gravity, all elastic elements of the robot are subject to varying or equal torques. Since the elastic elements exhibit a certain degree of elasticity, they react to the torques caused by the external force with slight elastic deformation, which leads to inaccuracies in the robot's operation. These elastic deformations are classified as non-geometric error influences. Accordingly, the uncorrected robot structural information is subject to errors due to ignored or incorrect non-geometric error influences.

[0044] The method according to the invention is characterized in that the at least one radiation pattern is radiated through the working space in such a way and the plurality of measuring configurations is selected in such a way that for at least one elasticity element from the plurality of elasticity elements there is at least one pair of measuring configurations for which the absolute value of the difference in the torques of the at least one pair of measuring configurations on the at least one elasticity element is greater than a target value, that on the light-sensitive surface of the at least one sensor a deviation from a straight line and / or plane implicitly predetermined by the at least one radiation pattern and its beam direction as well as its radiation pattern orientation is taken into account,by implicitly comparing the position of the projection detected by at least one sensor for each of the plurality of measurement configurations with a position of the projection determined on the basis of the erroneous robot structural information, and by generating corrected robot structural information to compensate for the non-geometric error influences from the deviation.

[0045] In order for parameter identification to assign the correct elasticities to the different elasticity elements, sufficient measurement data containing sufficient information must be available. As an example, consider the parallel rotary joints of typical six-axis industrial robots. Assume that the second and third rotary joints are always arranged parallel, as usual. If the joint position or joint angle of the third joint were constant across all measurement configurations, the torques on the second and third joints would be proportional to each other in all measurement configurations. In this case, no known method would be able to distinguish between the elasticities of joints two and three and determine the correct elasticities of the different joints.To enable a precise determination of the global correction information, it is advantageous if the entire set of measurement configurations includes different joint configurations in which the torques on the elastic elements vary sufficiently. Each of the different elasticities should be excited as strongly as possible and brought into opposite load states if possible. Furthermore, each elasticity element should be brought into as different load states as possible relative to the other elasticity elements. If the spectrum of load states of the elasticity elements is as broadly distributed and as heterogeneous as possible, a largely exact assignment of the elasticities to the individual elasticity elements is possible. The measurement configurations are preferably selected such that for each elasticity element, many pairs of measurement configurations have a large difference in the applied torques.Smaller torque differences lead to poorer calibration results. A clever arrangement of the calibration objects can promote the frequent occurrence of large torque differences.

[0046] The target value mentioned is not a fixed value but rather a flexibly adjustable value.

[0047] Furthermore, according to the invention, the deviation is determined by comparing the recorded projection position for the measurement configurations with a projection position determined based on the erroneous robot structural information. The deviations determined in this way include the error influence of the elasticity of the elastic elements as well as, in general, the non-geometric and geometric parameter values during robot operation when the robot is controlled according to the erroneous robot structural information.

[0048] Finally, to compensate for the non-geometric and, advantageously, also the geometric error influences, corrected robot structural information is calculated from the totality of the deviations. This corrected robot structural information is used for future robot control.

[0049] The invention therefore describes a method which, without the need for a costly laser tracker or a theodolite system or similar costly global measuring devices, makes it possible to directly identify the non-geometric parameters such as the elasticities of joints or rigid bodies of the robot on the basis of the obtained measured values and to take them into account accordingly during a calibration of the robot or to eliminate the error influences in the corrected robot structural information, so that the working accuracy of the robot can be significantly improved compared to generic, locally measuring methods.

[0050] A further observation is that the precise determination of elasticities and other non-geometric error influences results in an additional increase in accuracy because the totality of all other robot structural information can therefore be determined even more precisely.

[0051] According to a preferred embodiment of the invention, it is provided that a model-based parameter identification is carried out, wherein the corrected global structural information is corrected model parameters of a mathematical model of the robot and of the at least one radiation pattern generator and of the at least one sensor, wherein the model parameters comprise robot parameters describing the robot and calibration object parameters describing the at least one radiation pattern generator and the at least one sensor, wherein the plurality of measurement configurations consists of at least one measurement series, wherein a measurement series comprises all measurement configurations that are recorded with a selected calibration object pair.

[0052] To calculate the corrected robot structural information, a mathematical model or a mathematical description of the robot's kinematic structure as well as the non-geometric and geometric error influences is implemented in the computing unit. In model-based robot calibration, the robot structural information is referred to as the model's robot parameters. The calibration object parameters describe the position of the sensor relative to the effector and of the radiation pattern generator relative to the robot base. The mathematical model describes the position of the effector and thus of at least one sensor for each given joint configuration. Once the position of the radiation pattern generator is known, the position of the projection on the sensor in the sensor coordinate system can be calculated using the mathematical model.It is therefore possible to calculate the location on the light-sensitive surface of the at least one sensor and the orientation in which the at least one radiation pattern hits the sensor. The calculated position of the projection is more or less subject to error, depending on whether and how well the model parameters were previously determined using the method according to the invention.

[0053] However, the projection position is initially calculated based on the erroneous robot structural information, resulting in the previously described deviation between the calculated position and the position detected by the sensor. The calculated, initially erroneous, projection position is used together with the measured, i.e., actual, position in the mathematical model to computationally determine the corrected robot parameters.

[0054] According to a particularly preferred embodiment of the invention, the corrected robot parameters are calculated iteratively by the computing unit. The goal of the iterative calculation method is, in particular, to minimize the residual or average residual error or deviation between the calculated positions of the respective projection and the positions detected by the sensor.

[0055] The iterative procedure generates improved model parameters in each iteration step, which are increasingly closer to the real or finally corrected model parameters, but are still subject to errors.

[0056] According to another particularly preferred embodiment of the invention, the corrected robot parameters are calculated by the computing unit using a characteristic system of equations, which can also be represented in the form of a characteristic matrix equation. The characteristic system of equations is derived from a general kinematic system of equations and additionally includes the position of calibration objects. In contrast to the kinematic system of equations, the characteristic kinematic calibration equation system—characteristic system of equations for short—also takes into account the calibration method used, along with the associated calibration objects and their arrangement. The characteristic system of equations can be represented as a characteristic (matrix) equation using homogeneous matrices.Preferably, each side of the characteristic equation describes the position of the radiation pattern on the light-sensitive sensor surface, which is explained in more detail below.

[0057] For each calibration method, a characteristic system of equations or a characteristic matrix equation can be established. The different characteristic equations of the different methods form a classification of the calibration methods, which allows a precise, sharp separation into different and closely related methods.

[0058] According to a particularly preferred embodiment of the invention, it is provided that the characteristic system of equations is formulated as P*L=G0*G1*…Gn*S wherein P describes a position of the at least one radiation pattern generator relative to the robot (110), wherein L describes a beam direction of the at least one radiation pattern, wherein G0 * G1 * ... G n starting from the robot base, describes a spatial transition from the robot base to the effector and the G i describe a transition from one rigid body of the robot to a next rigid body including the joint belonging to the respective transition or from one joint to a next joint including the intermediate rigid body, where n denotes the number of joints of the robot and where S describes a spatial transition from the effector of the robot to an arbitrary but fixed coordinate system on the light-sensitive surface of the at least one sensor.

[0059] The characteristic equation is obtained by successively mathematically describing the transformations or transitions from the robot base to the first joint, then from joint to joint, and finally from the last joint to the sensor. Linking them together or concatenating the transformations appropriately, the result is then compared to the position of at least one radiation pattern on the sensor. According to the aforementioned Denavit-Hartenberg specification, each of these transitions is preferably described uniformly using four spatial elementary transformations, which always consist of two pure rotations and two pure translations. The translations correspond to lengths in transitions, and the rotations correspond to rigid angles in transitions. All transitions or transformations can be represented, for example, using so-called homogeneous 4 x 4 matrices.Their mathematical connection is realized by the usual, simple matrix product of these matrices. If, for a given arrangement of calibration objects, a suitably chosen position of the radiation pattern—or its representation using a homogeneous matrix—is denoted by P and the orientation of the radiation pattern at position P by L, then P * L specifies the directional propagation of the radiation pattern. If the homogeneous matrices of the transitions in a given measurement configuration are denoted by G0, G1, ..., G. n and let S be the matrix that describes the transition from the last joint to the position of the projection of the radiation pattern on the sensor, then for a robot with n joints we obtain as a result of multiplying the transitions G0 * G1 * ... G n * S is a homogeneous 4 x 4 matrix describing the transformation from the robot base to the sensor.

[0060] The equation of the two matrix products or the two sides of the equation P*L=G0*G1*…Gn*S then specifies the position of the projection of the radiation pattern on the at least one light-sensitive sensor surface.

[0061] For a more in-depth explanation of the relationships, the case of the other basic method of laser sensor technology is mentioned, in which a laser is attached to the effector and the sensor is stationary in or near the workspace. In this case, a different characteristic equation of the type P'*S'=G0*G1*…Gn*L' with identical G i and different matrices P', L', S'.

[0062] The difference between the two characteristic equations shows that the two methods belong to different classes of the classification of calibration methods.

[0063] According to a particularly preferred embodiment of the invention, it is provided that the computing unit creates a Jacobi matrix based on the characteristic system of equations, which mathematically relates an infinitesimal change in the deviations to an infinitesimal change in the robot parameters.

[0064] The use of the Jacobian matrix for iterative parameter identification has proven to be a suitable method for determining optimal approximate solutions for overdetermined nonlinear systems of equations.

[0065] The Jacobian matrix, also known as the functional matrix or derivative matrix, is the mxn matrix of all first partial derivatives of the characteristic system of equations or the corresponding error functional.

[0066] According to a further preferred embodiment of the invention, it is provided that the computing unit forms a pseudo-inverse of the Jacobian matrix.

[0067] The pseudoinverse of a matrix is a generalization of the inverse matrix to non-square matrices, which is why it is often referred to as a generalized inverse. The use of pseudoinverses of matrices is well known from the solution of systems of linear equations and linear fit problems. According to another particularly preferred embodiment of the invention, the corrected robot parameters are calculated by the computing unit using model-based mathematical parameter identification, and this includes at least one calculation step performed as a nonlinear optimization. The nonlinear optimization methods advantageously used for this purpose have proven suitable for identifying precise values of the robot parameters of the characteristic equation system.

[0068] In mathematics, the term nonlinear optimization describes the process of optimizing a scalar objective function of one or more real variables within a restricted range. In other words, nonlinear optimization optimally fits the parameter values of the mathematical model so that the aforementioned residual is minimized. This has been shown to enable the determination of optimal parameter values with a small residual, even in cases where the initial parameter values and the acquired measured values are relatively inaccurate.

[0069] Methods suitable within the meaning of the invention for carrying out the at least one calculation step carried out as a non-linear optimization are, for example, the Gauss-Newton method and the Levenberg-Marquardt method.

[0070] According to an alternatively preferred embodiment of the invention, it is provided that the corrected robot structural information is contained in weighting matrices, wherein the weighting matrices are created and / or parameterized by means of a machine learning method and / or by artificial intelligence.

[0071] In this case, the method according to the invention is not implemented in a model-based manner. The weighting matrices contain the information of the mathematical model without the model parameters explicitly appearing in the matrices.

[0072] By its very nature, machine learning doesn't require the provision of a mathematical model. Instead, a data set of comparative values is collected from the deviation of the radiation pattern from rectilinear propagation or from the position of the projection detected by at least one sensor. This data set is then used to train the artificial intelligence. This then leads to the creation of the weighting matrices via several intermediate steps.

[0073] The more measured values or training data are available for machine learning, the more precisely the robot structural information implicitly contained in the weighting matrices can be determined and subsequently used to directly control the robot.

[0074] According to a further preferred embodiment of the invention, it is provided that the at least one radiation pattern generator is arranged such that an angle exists between the propagation direction of the at least one radiation pattern and a direction of action of the force of gravity, the angle being between 30° and 150°.

[0075] If the angle is 90°, the radiation pattern is projected horizontally through the room. This ensures that high torques act on the robot's elastic elements in many of the measurement configurations. The higher the torques acting during the measurements, the more pronounced the errors resulting from incorrect values of the elastic parameters or missing parameters become in the laser beam deviations. This allows the elasticities of the different elastic elements to be better differentiated and identified more precisely. Once the individual elasticities are precisely identified, their impact on the robot's positioning accuracy can be compensated.

[0076] According to a further preferred embodiment of the invention, the at least one radiation pattern comprises at least two rigidly connected laser beams with an angle of less than 5 degrees or at least two crossed light planes. This results in the advantage that the orientation of the projection onto the at least one light-sensitive surface of the sensor can be determined very easily and precisely. In this case, the spacing of the parallel laser beams should be close enough so that, in suitable measurement configurations, both can strike the light-sensitive surface of the sensor simultaneously. These radiation patterns provide more information per measurement than measurements with a simple laser beam.

[0077] According to a further preferred embodiment of the invention, it is provided that the target value is 5% of the mathematically maximum possible absolute value of all pairwise differences of the torques which any pairs of practically or theoretically measurable measuring configurations have on at least one elasticity element.

[0078] This is a preferred minimum requirement for the measurement series or measured values. For example, the elasticity element can be a rigid segment which, in a first measurement configuration, is stretched horizontally in a first direction so that a first maximum torque acts, and in a second measurement configuration is stretched horizontally in a second direction opposite to the first direction so that a second maximum torque acts. In other words, the elasticity element is elastically deformed to a maximum in a first direction by the external force and elastically deformed to a maximum in a second, opposite direction in the other measurement configuration. The absolute value of the difference between these maximum torques is the maximum expected absolute value.

[0079] The requirement is that in the totality of the measurement configurations there is at least one pair for which the absolute value of the difference between the two torques is greater than 5% of the maximum expected absolute value of the difference between the torques under any pair of measurement configurations.

[0080] This requirement places very low demands on the measurement series and the arrangement of the calibration objects. If the calibration objects are arranged inappropriately or if the measurement series fail to meet this requirement due to an inappropriate selection of the measurement configurations, the given measurement series are unsuitable for identifying the elasticity of the given elastic element with sufficient accuracy according to the method claimed here.

[0081] In practice, as already mentioned, one should ensure that there is not just a single such pair. Rather, the calibration will be more successful the more measurement configurations there are for each of the elasticity elements, where the spectrum of loading states of a given elasticity element is broadly distributed and as heterogeneous as possible.

[0082] According to a further preferred embodiment of the invention, the external force is a gravitational force, a compressive force, or a torsional force. These are the external forces usually acting on the robot. The force of gravity acts on the robot constantly and in every joint configuration, whereby the torque resulting from the force of gravity on one or more elastic elements is essentially determined by the orientation of the respective elastic element(s). If, for example, for a selected elastic element, the remaining arm - i.e. the remaining arm whose torque acts on the selected elastic element - is extended horizontally, a maximum torque results in this position of the remaining arm. If, on the other hand, the remaining arm is extended vertically, the resulting torque with regard to the elastic element disappears. The compressive force acts on the robot orIts elastic elements are compressed when the robot presses a workpiece or the effector against another body, such as a tool, causing the elastic elements to compress. This can also result in a torque being exerted. The torsional force acts, for example, when the robot rotates a workpiece against resistance, such as when tightening a screw into a thread with a specific torque. The identical torque then acts on the robot's elastic elements.

[0083] According to a further preferred embodiment of the invention, it is provided that the robot is repeatedly controlled with different additional weights or payloads into the plurality of measuring configurations and / or in subsets thereof.

[0084] The different payloads can, for example, correspond to the full weight or half the weight that the robot lifts or moves in production. Another important special case is that no additional payload is carried.

[0085] Since the robot typically lifts and moves workpieces or at least tools used to process the workpieces during operation, the torque acting on it due to gravity is not only determined by the weight of the robot or its elastic elements, but also by the lifted weight of the workpieces or tools. The totality of the corrected robot structural information can be determined even more precisely if the weight of the workpieces or tools acting on the elastic elements is varied as much as possible within the entire measurement series.

[0086] According to a further preferred embodiment of the invention, it is provided that the corrected robot structural information is generated for a predeterminable subset of elasticity elements of the robot.

[0087] For the method according to the invention, the elasticity elements do not need to be considered individually or together; rather, they can also be combined into different and changing groups. This provides additional information that can be used to determine the corrected robot structural information. This further improves the accuracy of the corrected robot structural information.

[0088] The invention also relates to an arrangement for compensating non-geometric error influences on the absolute accuracy of a robot by means of a laser sensor system, comprising a robot with a plurality of elastic elements and with a control unit, at least one radiation pattern generator, at least one sensor with at least one light-sensitive surface, wherein an elastic element is a rigid body or a joint or an effector or a robot base, wherein at least one radiation pattern generator is arranged stationary in an environment of the robot within a workspace or outside the workspace, wherein the at least one radiation pattern generator is designed to radiate at least one radiation pattern through the workspace of the robot, wherein the at least one radiation pattern comprises at least one laser light beam and / or at least one laser light plane,wherein at least one sensor with at least one light-sensitive surface is arranged on the effector of the robot, wherein the control unit is designed to control the robot, in accordance with robot structural information electronically stored in the control unit, successively into a plurality of measurement configurations in which the at least one radiation pattern impinges on the at least one light-sensitive surface, and wherein the sensor is further designed to detect a position of a projection of the at least one radiation pattern onto the at least one light-sensitive surface and to forward measurement information describing the position to a computing unit. According to the invention, the arrangement is designed to carry out the method according to the invention. This results in the advantages already described in connection with the method according to the invention also for the arrangement according to the invention,

[0089] According to a preferred embodiment of the invention, it is provided that the computing unit is functionally and structurally integrated into the control unit.

[0090] This results in the advantage that the method according to the invention can be executed by the robot without an additional, external processing unit. Furthermore, if the radiation pattern generators remain in place in the workspace after their initial arrangement, the method according to the invention can be executed again at any time and without special preparation, which can avoid or reduce downtime. Another advantage is that overall hardware savings can be achieved, since the processing unit and the control unit can share certain components, such as a hard drive, RAM, and the processor.

[0091] The invention is explained below by way of example with reference to embodiments shown in the figures.

[0092] They show: Fig. 1 shows, by way of example and schematically, a possible embodiment of an arrangement according to the invention for compensating non-geometric error influences on the absolute accuracy of a robot by means of a laser sensor system, Fig. 2 shows an example and schematic view of the robot in two different measuring configurations, Fig. 3 shows an example and schematic representation of the robot Fig. 2 in two further different measurement configurations and Fig. 4 shows, by way of example and schematically, another possible embodiment of an arrangement according to the invention.

[0093] Identical objects, functional units, and comparable components are designated by the same reference symbols throughout the figures. These objects, functional units, and comparable components are identical in terms of their technical features, unless explicitly or implicitly stated otherwise in the description.

[0094] Fig. Figure 1 shows, by way of example and schematically, a possible embodiment of an arrangement 100 according to the invention for compensating non-geometric error influences on the absolute accuracy of a robot 110 using a laser sensor system. The arrangement 100 comprises, for example, the robot 110 and the laser sensor system.

[0095] The robot 110 is designed, for example, as an industrial robot 110 which is used in a productive operation for processing workpieces (not shown in Fig. 1) is provided.

[0096] The robot 110 comprises, for example, a plurality of elastic elements 111, 112, 113, 114, 115, 116, 117, 120 and further, not in Fig. 1 explicitly shown elastic elements, of which the ball joints 114, 115 and 116 are composed.

[0097] The robot 110 is arranged stationary in the workspace 200 with its robot base 120, which is also elastically deformable under the action of force and is thus an elastic element 120.

[0098] The further elastic elements 111, 112, 113, 114, 115, 116, 117 are partly designed as rigid bodies 111, 112, 113 and partly as joints 114, 115, 116. The joints 114, 115 and 116 are designed, for example, as ball joints 114, 115, 116, which each consist of two elementary rotation joints (not shown in Fig. 1). The robot 110 thus comprises six elementary rotational joints.

[0099] The rigid bodies 111, 112, 113 represent, for example, arm segments 111, 112, 113 of the robot 110, which connect the joints 114, 115, 116 to each other. The joints 114, 115, 116 are, for example, ball joints 114, 115, 116. Another elastic element 117 is the effector 117, on which an optical sensor 130 with, for example, two light-sensitive surfaces is arranged. The effector 117 can be moved within the workspace according to the joint configurations of the robot 110. All elastic elements 111, 112, 113, 114, 115, 116, 117, 120 are elastically deformable at least to a small extent under the influence of an external force, for example gravity.

[0100] The arrangement 100 according to the invention also includes, for example, a control unit 140, which is connected to the robot 110 via a wired data connection and is arranged outside the workspace 200. The control unit 140 is designed as a computer device 140, which, using suitable control software, is designed to control the robot 110 or to specify joint configurations of the robot 110 and to read out joint positions. The control unit 140 further has a human-machine interface via which a human operator can make various inputs, for example for controlling or maintaining the robot 110.

[0101] The arrangement 100 according to the invention further comprises the already mentioned optical sensor 130 with the two light-sensitive surfaces and two radiation pattern generators 131, 132.

[0102] The two radiation pattern generators 131, 132 are in the example of Fig. 1 are arranged stationary and with a fixed orientation in the workspace 200. For example, the radiation pattern generators 131, 132 are laser diodes 131, 132 that emit a simple beam with a point-shaped cross-section. The radiation pattern generators 131, 132 emit, for example, at a wavelength of approximately 700 nm.

[0103] The arrangement of the radiation pattern generators 131, 132 in the work space 200 is such that a first radiation pattern of a first radiation pattern generator 131 is emitted at an angle of 30° against the direction of gravity (illustrated by an arrow 160) and a second radiation pattern of a second radiation pattern generator 132 is emitted at an angle of 25° against gravity.

[0104] The control unit 140 is configured to successively control the robot 110 into a plurality of measurement configurations based on robot structural information stored electronically in the control unit 140. To this end, the control unit 140 uses a mathematical model of the robot, which mathematically describes the robot 140 via its robot parameters. The mathematical model contains the robot structural information as so-called robot parameters, with the robot parameters describing the non-geometric and geometric properties of the robot 110. The robot parameters, like the calibration object parameters, are part of the model parameters, with the calibration object parameters describing the positions and orientations of the radiation pattern generators 131, 132 relative to the robot base 120.

[0105] A measurement configuration is a joint configuration in which at least one of the two radiation patterns impinges on one of the two light-sensitive surfaces of the sensor 130. When a radiation pattern impinges on a light-sensitive surface of the sensor 130, the sensor 130 detects the precise position of the radiation pattern on the light-sensitive surface. In other words, the sensor 130 detects the projection of the radiation pattern impinging on the light-sensitive surface.

[0106] The sensor 130 then wirelessly transmits measurement information describing the position of the projection to a computing unit 150, for example. The control unit 140 also transmits information describing the measurement configuration to the computing unit 150. However, since the robot structural information stored in the control unit 140 and used for the mathematical model has not yet been corrected, for example, it is erroneous and leads to a deviation between a target position of the projection of the radiation pattern on the sensor 130 at the effector 117 and an actual position of the projection of the radiation pattern on the sensor 130 at the effector 117.This deviation is significantly influenced by a gravitational force acting on the exemplary elasticity elements 111, 112, 113, 114, 115, 116, 117, 120, which leads to an elastic deformation of the elasticity elements 111, 112, 113, 114, 115, 116, 117, 120 that depends on the respective joint configuration. This influence of gravity on the elasticity of the elasticity elements 111, 112, 113, 114, 115, 116, 117, 120 is not reflected in the uncorrected robot parameters and leads to a so-called non-geometric error when controlling the robot without taking these elasticities into account.

[0107] The two radiation patterns are now radiated through the workspace 200 in such a way that a plurality of measurement configurations is possible, in which for each elasticity element 111, 112, 113, 114, 115, 116, 117, 120 to be identified. In the totality of all measurement configurations, at least one pair should be included for which the absolute value of the difference in the torques caused by gravity on at least one elasticity element 111, 112, 113, 114, 115, 116, 117, 120 is greater than a target value, wherein the target value is, for example, 5% of the maximum expected absolute value of the difference in the torques of the at least one pair of measurement configurations on that same elasticity element 111, 112, 113, 114, 115, 116, 117, is 120.

[0108] The computing unit 150 then calculates the corrected robot structural information or the corrected model parameters in an iterative process from the deviations between the detected positions of the projection on the light-sensitive surface of the sensor 130 and the respective calculated impact points. The corrected model parameters are calculated, for example, based on a characteristic system of equations.

[0109] According to another, also in Fig. In the embodiment shown in Figure 1, the control unit 140 is designed as a neural network that generates the corrected robot structural information using artificial intelligence. In this case, the robot structural information is contained in weighting matrices, which the control unit 140 generates from as much measurement data as possible. The weighting matrices thus contain the information of the mathematical model without the model parameters explicitly appearing in the matrices.

[0110] Machine learning eliminates the need for a mathematical model. Instead, a data set of comparative values is collected directly from the deviations of the radiation pattern from rectilinear propagation or from the position of the projection detected by at least one sensor, which is then used to train the artificial intelligence.

[0111] Fig. 2 shows, by way of example and schematically, the robot 110 in two different measurement configurations I, II. The measurement configuration I is shown in dashed lines, while the measurement configuration II is shown with solid lines. In both measurement configurations I, II, a gravity-induced torque acts on the elastic elements 111, 112, 113, 114, 115, 116, 117, 120. The special feature here is that the joints 115 and 116 in the two measurement configurations I, II shown are each subjected to opposite maximum loads and the absolute value of the difference between the two torques is maximum, since the rigid body 113, the joint 116 and the effector 117 are held perpendicular to the gravity 160 in both measurement configurations I, II. These torques act on the individual elastic elements 111, 112, 113, 114, 115, 116, 117, 120 differently and depending on the respective measurement configuration I, II.For example, the joint 114 in the measurement configuration I is acted upon by a force shown in the illustration in the . Fig. 2 anti-clockwise torque, which is caused by the weight of the elastic elements 112, 113, 115, 116, 117 following the joint 114 being held by the joint 114. The joint 114 is therefore subjected to the weight of the elastic elements 112, 113, 115, 116, 117 and the resulting torque in the representation of the Fig. 2 slightly bent to the bottom left.

[0112] In measurement configuration I, the joint 115 also experiences a left-handed torque, which is caused by the weight of the elastic elements 113, 116, 117 following the joint 115. As already described, the torque acting on the joint 115 is shown in the diagram of the Fig. 2 is maximum, since the elastic elements 113, 116, 117 following joint 115 are stretched perpendicular to gravity 160. The torque acting on joint 115 acts downward to the left. Joint 116 also experiences a maximum torque in measurement configuration I. The torque acting on joint 116 is also maximum and is caused by the weight of effector 117, which is stretched perpendicular to gravity 160. The torque acting on joint 116 also acts downward to the left.

[0113] If the robot 110 is brought into the measuring configuration II, which, for example, corresponds to the measuring configuration I mirrored on a vertical axis, the torques from the measuring configuration I act on the joints 114, 115 and 116, but with the opposite sign.

[0114] The resulting absolute value of the difference in torque acting on joints 115 and 116 in measurement configurations I and II is therefore, as shown in the example, maximum. The absolute value of the difference in torque at joint 114, on the other hand, is Fig. 2 not maximum.

[0115] Fig. 3 shows an example and schematically the robot 110 of the Fig. 2 in two further, different measurement configurations III, IV. Measurement configuration III is shown in dashed lines, while measurement configuration IV is shown in solid lines. In contrast to Fig. 2 is according to Fig. 3 also the joint 114 is subjected to opposite maximum load in the two measurement configurations III, IV shown.

[0116] In the measurement configuration III shown in dashed lines, the elastic elements 112, 113, 115, 116, 117 following the joint 114 are stretched perpendicular to the force of gravity, so that a maximum torque acting downwards to the left results at the joint 114.

[0117] In measurement configuration IV, the maximum torque also acts on joint 114, since here too the elastic elements 112, 113, 115, 116, 117 following joint 114 are stretched perpendicular to the force of gravity, although in this case the torque is opposite in its direction of action to the torque at joint 114 in measurement configuration III.

[0118] Measurement configuration IV corresponds to measurement configuration III mirrored on a vertical axis.

[0119] Therefore, the absolute value of the difference in torques at the joint 114 for the measuring configurations III and IV according to the embodiment of the Fig. 3 maximum.

[0120] The torques on the joints 115 and 116 in both measurement configurations III, IV are identical to those already determined in Fig. 2 described torques.

[0121] Fig. 4 shows, by way of example and schematically, another possible embodiment of an arrangement 100 according to the invention. According to the example, the arrangement 100 comprises the robot 110 and the radiation pattern generator 131, which are arranged at a distance from one another in the work space 200.

[0122] Fig.Figure 4 illustrates the representation of the robot 110 and the radiation pattern generator 131 in the mathematical model or in the characteristic equation system. The robot 110, for example, again comprises the three ball joints 114, 115, and 116. The corresponding joint transitions are each described as homogeneous 4 x 4 matrices and designated G1 to G6. The origin of the robot base coordinate system is, for example, in the upper left corner of the pedestal-shaped robot base 120. The transition from this upper left corner to the rigid segment 111 is represented, for example, via the transfer matrix G0.

[0123] The radiation pattern generator 131 emits a radiation pattern. The exit point of the radiation pattern from the radiation pattern generator 131 relative to the upper left corner of the robot base 120 is given by the vector p or the transition matrix P. The radiation pattern generator 131 radiates in the direction r. The radiation pattern strikes the optical sensor 130 at a point that is given relative to a local sensor coordinate system on the light-sensitive surface of the sensor 130 by two plane coordinates or a transition matrix S. The distance between the exit point of the radiation pattern and the point of incidence of the radiation pattern is a distance I. The transition matrix from the exit point of the radiation pattern to the point of incidence of the radiation pattern at a distance l is denoted by the transition matrix L. Using these notations, the characteristic system of equations can be written, for example, as follows: P*L=G0*G1*G2*G3*G4*G5*G6*S

[0124] The straightness of the radiation pattern, which is a laser beam and thus completely straight, is specified, for example, by the transition matrix L in the model, which specifies a straight propagation with a constant direction vector r and a scalar variable. The straightness of the radiation pattern is thus integrated into the mathematical model, as is the specific arrangement of the calibration objects 130, 131. Together with the elasticities of the elasticity elements 111, 112, 113, 114, 115, 116, 117, 120, which are also described in the mathematical model, the mathematical model, together with corresponding measurements, can now be subjected to nonlinear optimization in order to, among other things, determine the elasticities or correct erroneous model parameters. List of reference symbols 100 arrangement 110 robots, industrial robots 111, 112, 113 Elastic element, rigid body 114, 115, 116 Elastic element, joint 117 Elasticity element, effector 120 robot base 130 Optical Sensor 131 radiation pattern generator, semiconductor laser diode 132 radiation pattern generator, semiconductor laser diode 140 control unit 150 computing units 160 Gravity 200 workspace r direction vector l Distance G0, G1, G2, G3, G4, G5, G6, transition matrix S, P, LI, II, III, IV measurement configuration

Claims

[1] Method for compensating non-geometric error influences on an absolute accuracy of a robot (110) by means of a laser sensor system, wherein the robot (110) comprises a plurality of elastic elements (111, 112, 113, 114, 115, 116, 117) and a control unit (140), wherein an elastic element (111, 112, 113, 114, 115, 116, 117) is a rigid body (111, 112, 113) or a joint (114, 115, 116) or an effector (117) or a robot base (120), wherein at least one radiation pattern generator (131, 132) is arranged stationary in an environment of the robot (110) within a workspace (200) or outside the workspace (200), wherein at least one radiation pattern is radiated through the working space (200) of the robot (110) by means of the at least one radiation pattern generator (131, 132), wherein the at least one radiation pattern comprises at least one laser light beam and / or at least one laser light plane, wherein at least one sensor (130) with at least one light-sensitive surface is arranged on the effector (117) of the robot (110), wherein the robot (110) is successively controlled by the control unit (140) into a plurality of measuring configurations in which the at least one radiation pattern impinges on the at least one light-sensitive surface, wherein the robot (110) is controlled in accordance with robot structural information stored electronically in the control unit (140), wherein a position of a projection of the at least one radiation pattern onto the at least one light-sensitive surface is detected by the at least one sensor (130) and measurement information describing the position is forwarded from the at least one sensor (130) to a computing unit (150), wherein torques act on the elastic elements (111, 112, 113, 114, 115, 116, 117, 120) due to an external force depending on a respective joint configuration and where the robot structural information is inaccurate due to non-geometric error influences, characterized by , that the at least one radiation pattern is radiated through the working space (200) and the plurality of measuring configurations (I, II, III, IV) is selected such that for at least one elasticity element (111, 112, 113, 114, 115, 116, 117, 120) from the plurality of elasticity elements (111, 112, 113, 114, 115, 116, 117, 120) there is at least one pair of measuring configurations (I, II, III, IV) for which the absolute value of the difference between the torques of the at least one pair of measuring configurations on the at least one elasticity element (111, 112, 113, 114, 115, 116, 117, 120) is greater than a target value, that on the light-sensitive surface of the at least one sensor (130) a deviation from a straight line and / or plane implicitly predetermined by the at least one radiation pattern and its beam direction as well as its radiation pattern orientation is taken into account, by implicitly comparing the position of the projection detected by at least one sensor (130) for each of the plurality of measurement configurations (I, II, III, IV) with a position of the projection determined on the basis of the faulty robot structure information, and that corrected robot structural information is generated to compensate for the non-geometric error influences from the deviation. [2] Method according to claim 1, characterized bythat a model-based parameter identification is carried out, wherein the corrected global structural information is corrected model parameters of a mathematical model of the robot (110) and of the at least one radiation pattern generator (131, 132) and of the at least one sensor (130), wherein the model parameters comprise robot parameters describing the robot (110) and calibration object parameters describing the at least one radiation pattern generator (131, 132) and the at least one sensor (130), wherein the plurality of measurement configurations consists of at least one measurement series, wherein a measurement series comprises all measurement configurations recorded with a selected calibration object pair (130, 131, 132). [3] Method according to claim 2, characterized by that the corrected robot parameters are calculated iteratively by the computing unit (150). [4] Method according to at least one of claims 2 and 3, characterized bythat the corrected robot parameters are calculated by the computing unit using a characteristic system of equations, which can also be represented in the form of a characteristic matrix equation, whereby the characteristic system of equations is derived from a general kinematic system of equations and additionally includes the position of calibration objects. [5] Method according to claim 4, characterized by that the characteristic system of equations is formulated P*L=G0*G1*…Gn*S wherein P describes a position of the at least one radiation pattern generator relative to the robot (110), wherein L describes a beam direction of the at least one radiation pattern, wherein G0 * G1 * ... G n starting from the robot base, describes a spatial transition from the robot base to the effector (117) and the G idescribe a transition from one rigid body of the robot to a next rigid body including the joint belonging to the respective transition or from one joint to a next joint including the intermediate rigid body, where n denotes the number of joints of the robot and where S describes a spatial transition from the effector of the robot to an arbitrary but fixed coordinate system on the light-sensitive surface of the at least one sensor. [6] Method according to at least one of claims 4 and 5, characterized by that the computing unit creates a Jacobi matrix based on the characteristic system of equations, which mathematically relates an infinitesimal change in the deviations to an infinitesimal change in the robot parameters. [7] Method according to claim 6, characterized by that the computing unit (150) forms a pseudoinverse of the Jacobian matrix. [8] Method according to at least one of claims 2 to 7, characterized by that the corrected robot parameters are calculated by the computing unit (150) by means of model-based mathematical parameter identification and this comprises at least one calculation step carried out as a non-linear optimization. [9] Method according to claim 1, characterized by that the corrected robot structural information is contained in weighting matrices, wherein the weighting matrices are created and / or parameterized by means of a machine learning method and / or artificial intelligence. [10] Method according to at least one of claims 1 to 9, characterized by that the at least one radiation pattern generator is arranged such that an angle exists between the propagation direction of the at least one radiation pattern and a direction of action of gravity, the angle being between 30° and 150°. [11] Method according to at least one of claims 1 to 10, characterized by that the at least one radiation pattern comprises at least two rigidly connected laser beams with an included angle of less than 5 degrees or at least two crossed light planes. [12] Method according to at least one of claims 1 to 11, characterized by that the target value is 5% of the mathematically maximum possible absolute value of all pairwise differences of the torques which any pairs of practically or theoretically measurable measuring configurations have on at least one elasticity element. [13] Method according to at least one of claims 1 to 12, characterized by that the external force is a gravitational force, a compressive force or a torsional force. [14] Method according to at least one of claims 1 to 13, characterized bythat the robot (110) is repeatedly controlled with different additional weights or payloads into the plurality of measuring configurations and / or subsets thereof. [15] Method according to at least one of claims 1 to 14, characterized by that the corrected robot structure information is generated for a predeterminable subset of elasticity elements (111, 112, 113, 114, 115, 116, 117, 120) of the robot (110). [16] Arrangement (100) for compensating non-geometric error influences on an absolute accuracy of a robot (110) by means of a laser sensor system, comprising a robot (110) with a plurality of elastic elements (111, 112, 113, 114, 115, 116, 117, 120) and with a control unit (140), at least one radiation pattern generator (131, 132), at least one sensor (130) with at least one light-sensitive surface, wherein an elastic element (111, 112, 113, 114, 115, 116, 117) is a rigid body (111, 112, 113) or a joint (114, 115, 116) or an effector (117) or a robot base (120), wherein at least one radiation pattern generator (131, 132) is arranged stationary in an environment of the robot (110) within a workspace (200) or outside the workspace (200), wherein the at least one radiation pattern generator (131, 132) is designed to radiate at least one radiation pattern through the working space (200) of the robot (110), wherein the at least one radiation pattern comprises at least one laser light beam and / or at least one laser light plane, wherein at least one sensor (130) with at least one light-sensitive surface is arranged on the effector (117) of the robot (110), wherein the control unit (140) is designed to control the robot (110) in accordance with robot structural information stored electronically in the control unit (140) one after the other into a plurality of measurement configurations (I, II, III, IV), in which the at least one radiation pattern strikes the at least one light-sensitive surface and wherein the sensor (130) is further configured to detect a position of a projection of the at least one radiation pattern onto the at least one light-sensitive surface and to forward measurement information describing the position to a computing unit (150), characterized by , that the arrangement (100) is designed to carry out a method according to at least one of claims 1 to 15. [17] Arrangement according to claim 16 characterized by , that characterized by that the computing unit (150) is functionally and structurally integrated into the control unit (140).

Images