DE102022123243B3 - Artificial intelligence for precise determination of output torques of collaborative robots - Google Patents

Abstract

The invention creates a new artificial intelligence, in particular a machine learning model or machine learning algorithm, for the precise determination of an output-side torque, in particular an output-side torque of an actuator gear of a joint of a collaborative robot, which is set up to output one or more output variables based on input variables , wherein the input variables of the artificial intelligence include: a first angle specification, which corresponds to an angular position on the input side, a second angle specification, which corresponds to an angular position on the output side, and wherein further the output variables include the outgoing torque determined by the artificial intelligence, the artificial intelligence furthermore doing so is set up to output one or more secondary output variables based on secondary input variables, the secondary input variables of the artificial intelligence for each of two or more, in particular six, joint actuators of a collaborative robot comprising: a first angle specification, which corresponds to an input-side angular position at the respective joint corresponds to a second angle specification, which corresponds to an output-side angular position at the respective joint, and wherein optionally the secondary output variables also include the six-axis determination of forces and torques at the end effector provided by the artificial intelligence, in particular a three-component force and a three-component torque at the end effector.

Description

TECHNICAL FIELD

The present invention relates generally to the technical field of robotics, autonomous mobile robots (AMRs) and in particular collaborative robots (cobots) and safety and precision technology in the workplace and in assembly plants.

BACKGROUND

Cobot systems have been used in more and more facilities in recent years. This has led to human workers increasingly sharing the same space with cobot systems. Such robots are designed to be particularly collaborative. For example, movements should initially be stopped if objects or workers are in the way in order to prevent damage and injuries. So-called voltage wave transmissions (harmonic drive / harmonic gear box) are often used as transmissions.

For solid and precise control of a cobot and its articulated motors, which makes it particularly suitable for use in precision environments as well as working environments with human workers, it is necessary to measure and/or know an outgoing/output torque as precisely as possible. This is particularly the case for the locally occurring torques at a joint of a robot arm.

A local torque sensor is often used on a joint for this purpose. However, this is often expensive, required for each joint, is often very heavy, hinders the robot's flexibility during use and increases energy consumption.

Robots can also be equipped with six-axis force-torque sensors. Such a sensor can, for example, be arranged on an end effector on a multi-jointed robot arm. In particular, however, the potential of the data provided by such a sensor is not fully exploited.

Six-axis force-torque sensors are already used in robots or can often be retrofitted.

The patent specification EP 2 231 369 B1 discloses a robot and a method for monitoring the moments on such a robot. What is described is a robot with at least two joints and parts that can move relative to one another by means of at least one joint, with at least one sensor detecting torque being arranged on at least one movable part, and two sensors and redundant evaluation units for redundant evaluation being provided for the redundant detection of a torque and wherein a device is provided for switching off the robot or for triggering a safe state if measured values of the same torque recorded by the at least two sensors deviate from one another outside a predetermined tolerance range, the two sensors being in the form of full bridges having strain gauges on a transmission of the robot are arranged so that they detect the same torque and are connected by means of two computer units in the form of differently designed integrated circuits with microcontrollers within a transmitting unit in which a first check of the measured torque values is carried out. However, the solution described here has, among other things, the disadvantage that dual sensors are required to detect torque, which is expensive and also has a detrimental effect on the dynamics and weight of the robot.

The DE 10 2021 212 128 B3 describes a method for controlling a load-bearing robot, comprising determining a one- or multi-dimensional parameter of the load based on transmission input-side and transmission output-side joint coordinate values of the joints that have been determined in a first position of the robot. Furthermore, a first transformation takes place between joint coordinates on the transmission input side and a pose of a robot-fixed reference and a second transformation between joint coordinates on the transmission output side and a pose of the robot-fixed reference.

The DE 102 45 594 A1 describes a method for collision detection for a drive system with a motor and a moving part driven by the motor. The method includes the measurement of a drive-side movement variable of the motor, the measurement of an output-side movement variable of the movable part and the determination of an error signal depending on the two movement variables. From another prior art there is an indirect determination of torques, for each individual joint, via the motor current flow known. This solution is inexpensive, but highly imprecise, and therefore no longer meets today's quality standards for high-performance cobot systems.

It is therefore an object of the present invention to further develop collaborative robot systems in such a way that they work precisely and safely ten, especially in shared work environments with people. A high level of precision and flexibility of the robots should be guaranteed and costs should be minimized as best as possible without compromising on the quality of the product and the results achieved with it.

This object is achieved by the method according to claim 1. According to the invention, an artificial intelligence, in particular a machine learning model or machine learning algorithm, is provided for the precise determination of an output-side torque, in particular an output-side torque of an actuator gear of a joint of a collaborative robot, which is set up for this purpose is to output one or more output variables on the basis of input variables, the input variables of the artificial intelligence comprising: a first angle specification, which corresponds to an angular position on the input side, a second angle specification, which corresponds to an angular position on the output side, and furthermore the output variables that are determined by the artificial intelligence include certain outgoing torque, wherein the artificial intelligence is further configured to output one or more secondary output variables based on secondary input variables, the secondary input variables of the artificial intelligence for each of two or more joint actuators, particularly preferably six joint actuators, of a collaborative robot include: a first angle specification, which corresponds to an input-side angular position at the respective joint, a second angle specification, which corresponds to an output-side angular position at the respective joint, and in particular the secondary output variables further represent the six-axis determination of forces and torques at the end effector provided by the artificial intelligence, in particular a three-component force and a three-component torque at the end effector.

According to the invention, the artificial intelligence determines the forces and torques at the joints of the robot particularly precisely. However, to train this artificial intelligence, forces and torques on the end effector, in particular the three-component force and the three-component torque, are used. In particular, training can continue in continuous operation based on data from the six-axis force-torque sensor, which further improves the determination of the individual joint torques during operation and dynamically adapts it to the current state of the robot (e.g. when wear occurs). . The precision is higher than with conventional methods. As a result, this precise measurement allows the cobot control and the motor control to work and act more precisely.

According to the invention, the artificial intelligence is trained with the help of a feedback control loop using a measured value of an end effector-side torque of a six-axis force-torque sensor, which compares the output-side torque with this and terminates the training if it is compromised.

According to the invention, a precise indirect six-axis determination of forces and torques on an end effector is carried out using the following data sets as input data: data sets comprising data relating to an input-side angular position and an output-side angular position of a respective joint actuator gear for six joint actuators of a robot arm,

According to the invention, a measured value of an end effector-side torque, measured by a six-axis force-torque sensor, is also used as a target value as part of the feedback control loop in which the artificial intelligence is arranged.

The additional six-axis force-torque sensor and the once again redundant determination of torques/forces further increase the precision. The direct output of the torque sensor can train the AI excellently. In addition, the solution according to the invention can also determine six-component torques/forces independently of the torque sensor, which is particularly advantageous, for example, if the latter fails. This avoids dead time and incorrect and/or uncontrolled (and therefore often dangerous) behavior of the cobot.

Further according to the invention, training is carried out continuously, but training is aborted and/or interrupted if it is determined that the value of the output torque measured by the six-axis force-torque sensor is compromised or the corresponding signal breaks off.

Necessary calculations can take place locally and/or non-locally, in particular also in a distributed manner, for example in the context of cloud computing. A robot according to the invention, which is generally local, benefits from the invention through improved behavior and improved work safety, also through non-local calculations.

In a preferred embodiment, the artificial intelligence is set up for unsupervised learning with a multi-layer perceptron.

In a further preferred embodiment, the input variables for each joint further include: angular velocity, velocity, acceleration and/or temperature.

In a further embodiment, a method for precisely determining an output-side torque, in particular an output-side torque of an actuator gear of a joint of a collaborative robot, is provided by means of an artificial intelligence, in particular a machine learning model or machine learning algorithm, which is set up to do so on the basis of input variables to output one or more output variables, the input variables of the artificial intelligence comprising: a first angle information, which corresponds to an angular position on the input side, a second angle information, which corresponds to an angular position on the output side, and wherein the output variables further include the outgoing torque determined by the artificial intelligence , wherein the artificial intelligence is and/or was trained using a method according to a training method according to the invention, in particular is continuously trained in continuous operation.

In addition, the invention creates a training method suitable for artificial intelligence. According to the invention, a method is provided for training an artificial intelligence, in particular a machine learning model or machine learning algorithm, according to one of the preceding claims, in particular through unsupervised learning, by means of a precise indirect six-axis determination of forces and torques on an end effector using the following Data sets as input data: Data sets comprising data relating to an input-side angular position and an output-side angular position of a respective joint actuator gear (203), in particular for six joint actuators of a robot arm. Training is carried out using the six-axis determination by the sensor. The particular goal is to train the AI in such a way that it enables improved torque determination at the individual joints during operation.

The Kl can be trained continuously during operation, which ensures an extremely high level of training and also takes into account changes that occur over time (e.g. changed environmental conditions or wear).

Preferably, training takes place taking into account a direct torque and force measurement as a target value (for example a measured value from a six-axis force-torque sensor).

Unsupervised learning has proven to be particularly suitable. Good results were achieved in particular when using a multi-layer perceptron as a neural network. The properties of the MLP ensure sufficient feedback.

The AI can be arranged in a kind of control loop and/or feedback loop. In this way, unsupervised learning can be effectively supplemented by six-axis force/torque measurement as a target value.

The invention enormously increases the dynamic usability of cobots in changing conditions. For example, a high level of precision after a replacement part on the cobot (or, for example, refurbished cobots) is guaranteed by the high flexibility of the system.

For example, the AI can take into account angular velocity, speed, acceleration and temperature as additional features (input variables), especially for each joint, to take these effects into account. However, the invention is by no means limited to this.

The invention uses artificial intelligence to make effects such as friction, wear, material fatigue, reversal range, hysteresis, counter-reactions, other non-linear effects, manufacturing tolerances and/or individual differences in production accessible to efficient evaluation and consideration in the cobot control. These effects are often difficult to simulate and even more rarely accessible to a direct and adequate mathematical description. These problems will be overcome using artificial intelligence.

The invention is by no means limited to the effects mentioned here (in the sense of an exhaustive list). On the contrary, a particular advantage of using artificial intelligence in the present case is that the individual effects do not have to be modeled manually, but rather all effects that occur - and all of them at the appropriate scale - are reproduced simultaneously and directly by the AI .

Through the precise and, if necessary, redundant determination of the three-component force on the effector side and the three-component torque For the purpose of training the multifunctional Kl, which can provide precise torques of individual joints during operation solely on the basis of angular positions - and therefore without direct measurement - the invention makes a lasting contribution to technical progress in the technical field of collaborative robots.

BRIEF DESCRIPTION OF THE DRAWINGS

• 1: ein Gelenk eines Roboters mit eingangsseitigem und ausgangsseitigem/ausgabeseitigem Winkellagekodierer
• 2: ein Gelenk eines Roboters ausgangsseitigem/ausgabeseitigem Drehmomentsensor und ausgangsseitigem/ausgabeseitigem Winkellagekodierer
• 3: eine schematische Zeichnung eines Antriebsstranges für ein Robotergelenk mit ausgangsseitigem/ausgabeseitigem Drehmomentsensor und beidseitigen Winkellagekodierern
• 4: einen Roboterarm mit Sechs-Achsen-Kraft-Drehmoment-Sensor am Endeffektor und beidseitigen Winkellagekodierern an den Getrieben, geeignet für einen Einsatz der vorliegenden Erfindung
• 5: eine schematische Zeichnung eines Maschinenlernmodells zum Einsatz im Rahmen der vorliegenden Erfindung.

Preferred embodiments of the present disclosure are described below with reference to the following figures:

• 1 : a joint of a robot with an input and output/output angular position encoder
• 2 : a joint of a robot output/output side torque sensor and output/output side angular position encoder
• 3 : a schematic drawing of a drive train for a robot joint with an output/output torque sensor and angular position encoders on both sides
• 4 : a robot arm with a six-axis force-torque sensor on the end effector and bilateral angular position encoders on the gears, suitable for use of the present invention
• 5 : a schematic drawing of a machine learning model for use in the context of the present invention.

DETAILED DESCRIPTION

The 1-3 describe how a joint works. The invention is then based on the 4 and 5 described.

The 1 shows a joint 100 of a robot with an input-side and output-side/output-side angular position encoder.

An input-side angle encoder or angular position encoder 1 is provided in the present case by the absolute value encoder 1.

Furthermore, an output-side/output-side angle encoder or angular position encoder 1 is provided by the incremental rotary encoder 1.

This cobot of the invention is accessible due to the double configuration with angular position encoders.

The 2 shows a joint 100 of a robot with an output-side/output-side torque sensor 210 and an output-side/output-side angular position encoder 202.

A tension shaft gear, wave gear or sliding wedge gear 203 is used for reduction in the drive train of the robot joint.

The 3 shows a schematic drawing of a drive train for a robot joint with an output/output torque sensor and angular position encoders on both sides.

An electric motor 330 drives a shaft 320 to rotate. A voltage converter 340 provides the necessary energy, with motor control electronics 350 controlling the motor 330. The shaft 320 opens into a speed reducer 360 on the output side. The gearbox effects a reduction, for example.

A rotary encoder or angular position encoder 301 is arranged on the input side (input-side angular position encoder 301). Another rotary encoder or angular position encoder 302 is arranged on the output side (output side angular position encoder 302). According to the invention, a force acting on the effector side and a torque acting on the effector side are determined by means of artificial intelligence from the measured values of the angular position sensors 301, 302 (and/or the difference in the measured values). In particular, this is done taking into account the various multiple joints of the robot arm.

The 4 shows a robot arm with a six-axis force-torque sensor 410 on the end effector and bilateral angular position encoders on the gears, also suitable for use of the present invention.

According to the invention, a force acting on the effector side and a torque acting on the effector side can be determined for the six joints, for example, using artificial intelligence from the measured values of the angular position sensors 301, 302 (and/or the difference in the measured values). This is a possibility for a six-axis determination of forces and torques on the end effector. This is preferably used for AI training.

The torque can also be determined directly by a six-axis force-torque sensor 410 on the effector side, provided such a six-axis force-torque sensor 410 is available or has been retrofitted. This creates a data set that is suitable for training the AI. In particular, a redundant six-axis determination of forces and torques is possible. This is particularly precise. Also A failure of the six-axis force-torque sensor is prevented.

The six-axis determination of forces and torques measured directly by the six-axis force-torque sensor can be used primarily to train the muscles.

For example, in continuous operation, the AI is further trained using the six-axis force-torque sensor.

Later, for example, the six-axis force-torque sensor can be removed, for example temporarily. But it can also remain there. If it fails at some point, the training is stopped and the excellent and extensively trained AI can even determine the six-axis determination of forces and torques on its own (based on the output values of the two rotary encoders per joint).

The invention has the further advantage that environmental effects (e.g. temperature) as well as signs of wear, material fatigue, manufacturing tolerances of the components, etc. are dynamically taken into account through the training, in particular through the continuous training.

The invention thus contributes to an improved cobot control, since it can work with significantly more precisely determined joint torques during operation. Through the continuous training of the systems based on the six-axis force-torque sensor, any changes that may occur are quickly learned and taken into account.

A further technical effect of the invention also occurs when the six-axis determination of forces and torques is carried out redundantly (by rotary encoder + KI as well as directly by torque sensor, in particular six-axis force-torque sensor), as well as when a torque sensor , particularly a six-axis force-torque sensor, fails. In the latter case, dead time and incorrect behavior of the cobot are avoided.

This brings economic benefits.

The main advantage, however, is the improved calculation of the joint torques on-the-fly during operation based on the angular positions at the joints.

The 5 shows a schematic drawing of a machine learning model for use in the context of the present invention. In particular, the model can be trained as shown.

For each joint (joint), the input and output angular positions are given as input to the Kl. By using the Kl, 6 torques or forces are determined, represented here by the numbering 0 - 5. This is done using a common Kl for the 6 joints, so that correlations between the behavior of the individual joints can be taken into account. This increases the accuracy of the determinations. From the 6 torques or forces, a further (mathematically coordinate-transformed) form of the six-axis determination of forces and torques can be derived, for example by transformation using a Jacobi matrix, which enables a direct comparison with a measurement result of a six-axis force-torque sensor .

In this way, a control loop can be created for improved training (error/feedback) of the AI, which particularly complements the unsupervised learning of the AI.

The Kl trained in this way can, for example, be used to precisely determine the torque at individual joints during operation, based on local angular position encoders. In particular, this is also possible without local torque sensors being present on the joints, since the training can already take place entirely on the basis of the signals from the six-axis force-torque sensor.

In a further embodiment, individual torques at the 6 joints can also be determined directly by joint-local torque sensors. This can particularly affect a subgroup of the 6 gears. In this way, the precision of the determination of individual joints can be improved again.

In this way, up to threefold redundancy can be created for the six-axis determination of the forces and torques, since firstly the AI determines from angle encoder signals, secondly through local torque sensors on the respective joints, thirdly through the effector-side six-axis force-torque system. Sensor, which can also be retrofitted.

The precision created in this way is unprecedented and convincing in practice! It meets the highest precision requirements.

The embodiments described schematically here can be further developed by numerous details of the invention, in particular those already described above.

Although some aspects have been described in the context of a device, it is clear that these aspects also represent a description of the corresponding method, where a block or a device corresponds to a method step or a function of a method step. Analogously, aspects that are described as part of a method step also represent a description of a corresponding block or element or a property of a corresponding device.

Embodiments of the invention can be implemented in a computer system. The computing system may be a local computing device (e.g., personal computer, laptop, tablet computer, or mobile phone, or embedded microcontroller) with one or more processors and one or more storage devices, or may be a distributed computing system (e.g., a cloud computing system with one or more processors or one or more storage devices distributed at different locations, for example at a local client and/or one or more remote server farms and/or data centers). The computer system may include any circuit or combination of circuits. In one embodiment, the computer system may include one or more processors, which may be of any type. As used herein, processor may mean any type of computing circuit such as, but not limited to, a microprocessor, a microcontroller, a complex instruction set microprocessor (CISC), a reduced instruction set microprocessor (RISC), a very long instruction word (Very Long Instruction Word; VLIW) microprocessor, a graphics processor, a digital signal processor (DSP), a multi-core processor, a field-programmable gate array (FPGA), or any other type of processor or processing circuit. Other types of circuits that may be included in the computer system may include a custom-built circuit, an application-specific integrated circuit (ASIC), or the like, such as one or more circuits (e.g., a communications circuit) for use with wireless devices such as . B. cell phones, tablet computers, laptop computers, walkie-talkies and similar electronic systems. The computer system may include one or more storage devices, which may include one or more storage elements suitable for the particular application, such as a main memory in the form of a random access memory (RAM), one or more hard drives, and/or a or multiple drives that handle removable media such as CDs, flash memory cards, DVDs, and the like. The computer system may also include a display device, one or more speakers, and a keyboard and/or controller, which may include a mouse, trackball, touch screen, voice recognition device, or any other device that allows a system user to enter information into the computer system and information to receive from the same.

Some or all of the method steps may be performed by (or using) a hardware device, such as a processor, a microprocessor, a programmable computer, or an electronic circuit. In some embodiments, one or more of the key method steps may be performed by such a device.

Depending on particular implementation requirements, embodiments of the invention may be implemented in hardware or software. The implementation may be performed with a non-volatile storage medium such as a digital storage medium such as a floppy disk, a DVD, a Blu-Ray, a CD, a ROM, a PROM and EPROM, an EEPROM or a FLASH memory in which electronically readable control signals are stored, which interact (or can interact) with a programmable computer system in such a way that the respective method is carried out. Therefore, the digital storage medium can be computer readable.

Some embodiments according to the invention include a data carrier with electronically readable control signals that can interact with a programmable computer system to perform one of the methods described herein.

In general, embodiments of the present invention may be implemented as a computer program product with program code, the program code being effective for executing one of the methods when the computer program product is running on a computer. The program code can, for example, be stored on a machine-readable medium.

Further exemplary embodiments include the computer program for carrying out one of the methods described herein, which is stored on a machine-readable medium.

In other words, an embodiment of the present invention is therefore a computer program with a program code for carrying out tion of any of the methods described herein when the computer program is running on a computer.

Therefore, another embodiment of the present invention is a storage medium (or a data carrier or a computer-readable medium) that includes, stored thereon, a computer program for carrying out one of the methods described herein when executed by a processor. The data carrier, digital storage medium or recorded medium is usually tangible and/or non-seamless. Another embodiment of the present invention is an apparatus as described herein that includes a processor and the storage medium.

A further exemplary embodiment of the invention is therefore a data stream or a signal sequence that represents the computer program for carrying out one of the methods described herein. The data stream or the signal sequence can, for example, be configured so that they are transmitted via a data communication connection, for example via the Internet.

Another embodiment includes a processing means, for example a computer or a programmable logic device, configured or adapted to carry out any of the methods described herein.

Another embodiment includes a computer on which the computer program for executing one of the methods described herein is installed.

Another embodiment according to the invention includes an apparatus or system configured to transmit (e.g., electronically or optically) to a receiver a computer program for carrying out one of the methods described herein. The receiver may be, for example, a computer, a mobile device, a storage device, or the like. The device or system may, for example, include a file server for transmitting the computer program to the recipient.

In some embodiments, a programmable logic device (e.g., a field programmable gate array, FPGA) may be used to perform some or all of the functionality of the methods described herein. In some embodiments, a field programmable gate array may cooperate with a microprocessor to perform any of the methods described herein. In general, the methods are preferably performed by each hardware device.

Embodiments may be based on using artificial intelligence, in particular a machine learning model or machine learning algorithm. Machine learning can refer to algorithms and statistical models that computer systems can use to perform a specific task without using explicit instructions, rather than relying on models and inference. In machine learning, for example, instead of a rule-based transformation of data, a transformation of data can be used that can be derived from an analysis of historical and/or training data. For example, the content of images can be analyzed using a machine learning model or using a machine learning algorithm. In order for the machine learning model to analyze the content of an image, the machine learning model can be trained using training images as input and training content information as output. By training the machine learning model with a large number of training images and/or training sequences (e.g. words or sentences) and associated training content information (e.g. labels or annotations), the machine learning model “learns” the content of the images recognize so that the content of images not included in the training data can be recognized using the machine learning model. The same principle can be used for other types of sensor data as well: By training a machine learning model using training sensor data and a desired output, the machine learning model "learns" a transformation between the sensor data and the output, which can be used to create a Provide output based on non-training sensor data provided to the machine learning model. The provided data (e.g. sensor data, metadata and/or image data) can be pre-processed to obtain a feature vector, which is used as input for the machine learning model.

Machine learning models can be trained using training input data. The examples above use a training method called “supervised learning.” In supervised learning, the machine learning model is trained using a plurality of training samples, where each sample may include a plurality of input data values and a plurality of desired output values, that is, each training sample is assigned a desired output value. By specifying both training samples and desired output values, the machine learning model “learns” which output value based on an input sample that is similar to the sample values provided during training. In addition to supervised learning, semi-supervised learning can also be used. In semi-supervised learning, some of the training samples lack a desired output value. Supervised learning can be based on a supervised learning algorithm (e.g. a classification algorithm, a regression algorithm, or a similarity learning algorithm). Classification algorithms can be used when the outputs are restricted to a limited set of values (categorical variables), i.e. the input is classified as one from the limited set of values. Regression algorithms can be used when the outputs show any numerical value (within a range). Similarity learning algorithms can be similar to both classification and regression algorithms, but are based on learning from examples using a similarity function that measures how similar or related two objects are. In addition to supervised learning or semi-supervised learning, unsupervised learning can be used to train the machine learning model. In unsupervised learning, input data may be provided (only) and an unsupervised learning algorithm may be used to find structure in the input data (e.g. by grouping or clustering the input data, finding commonalities in the data). Clustering is the assignment of input data comprising a plurality of input values into subsets (clusters) such that input values within the same cluster are similar according to one or more (predefined) similarity criteria, while they are dissimilar to input values comprised in other clusters.

Reinforcement learning is a third group of machine learning algorithms. In other words, reinforcement learning can be used to train the machine learning model. Reinforcement learning involves training one or more software actors (so-called “software agents”) to perform actions in an environment. Based on the actions taken, a reward is calculated. Reinforcement learning is based on training the one or more software agents to select the actions in such a way that the cumulative reward is increased, resulting in software agents that become better at the task they are given (as by increasing rewards proven).

Furthermore, some techniques can be applied to some of the machine learning algorithms. For example, feature learning can be used. In other words, the machine learning model may be at least partially trained using feature learning and/or the machine learning algorithm may include a feature learning component. Feature learning algorithms, called representation learning algorithms, can preserve the information in their input but transform it in such a way that it becomes useful, often as a preprocessing stage before performing classification or prediction. For example, feature learning can be based on principal component analysis or cluster analysis.

In some examples, anomaly detection (i.e., outlier detection) may be used, which aims to provide identification of input values that raise suspicion because they differ significantly from the majority of input and training data. In other words, the machine learning model may be trained at least in part using anomaly detection, and/or the machine learning algorithm may include an anomaly detection component.

In some examples, the machine learning algorithm may use a decision tree as a prediction model. In other words, the machine learning model can be based on a decision tree. In a decision tree, the observations about an item (e.g., a set of input values) can be represented by the branches of the decision tree, and an output value corresponding to the item can be represented by the leaves of the decision tree. Decision trees can support both discrete values and continuous values as output values. When discrete values are used, the decision tree can be called a classification tree; when continuous values are used, the decision tree can be called a regression tree.

Association rules are another technique that can be used in machine learning algorithms. In other words, the machine learning model can be based on one or more association rules. Association rules are created by identifying relationships between variables in large sets of data. The machine learning algorithm may identify and/or utilize one or more ratio rules that represent the knowledge derived from the data. The rules can e.g. B. be used to store, manipulate or apply knowledge.

Machine learning algorithms are usually based on a machine learning model. Different In other words, the term “machine learning algorithm” may mean a set of instructions that can be used to create, train, or use a machine learning model. The term “machine learning model” may mean a data structure and/or a set of rules that represents the learned knowledge (e.g. based on the training performed by the machine learning algorithm). In embodiments, the use of a machine learning algorithm may imply the use of an underlying machine learning model (or a plurality of underlying machine learning models). The use of a machine learning model may imply that the machine learning model and/or the data structure/set of rules that is/are the machine learning model is trained by a machine learning algorithm.

For example, the machine learning model can be an artificial neural network (ANN). ANNs are systems inspired by biological neural networks such as those found in a retina or brain. ANNs include a plurality of interconnected nodes and a plurality of connections, called edges, between the nodes. There are usually three types of nodes, input nodes that receive input values, hidden nodes that are (only) connected to other nodes, and output nodes that provide output values. Each node can represent an artificial neuron. Each edge can send information from one node to another. The output of a node can be defined as a (nonlinear) function of the inputs (e.g. the sum of its inputs). A node's inputs can be used in the function based on a "weight" of the edge or node providing the input. The weight of nodes and/or edges can be adjusted in the learning process. In other words, training an artificial neural network may include adjusting the weights of the nodes and/or edges of the artificial neural network, i.e. H. to achieve a desired output for a given input.

Alternatively, the machine learning model can be a support vector machine, a random forest model or a gradient boosting model. Support Vector Machines are supervised learning models with associated learning algorithms that can be used to analyze data (e.g. in a classification or regression analysis). Support vector machines can be trained by providing input with a plurality of training input values belonging to one of two categories. The support vector machine can be trained to assign a new input value to either category. Alternatively, the machine learning model may be a Bayesian network, which is a probabilistic directed acyclic graphical model. A Bayesian network can represent a set of random variables and their conditional dependencies using a directed acyclic graph. Alternatively, the machine learning model may be based on a genetic algorithm, which is a search algorithm and heuristic technique that mimics the process of natural selection.

Claims (11)

Artificial intelligence, in particular a machine learning model or machine learning algorithm, for precisely determining an output-side torque of an actuator gear (203) of a joint (100) of a collaborative robot, which is set up to output one or more output variables based on input variables, wherein the Input variables of the artificial intelligence include: a first angle specification, which corresponds to an angular position on the input side, a second angle specification, which corresponds to an angular position on the output side, and furthermore the output variables include the outgoing torque determined by the artificial intelligence, the artificial intelligence also being set up to do this: to output one or more secondary output variables based on secondary input variables, the secondary input variables of the artificial intelligence for each of six joint actuators (100) of a collaborative robot comprising: a first angle specification, which corresponds to an input-side angular position at the respective joint, a second angle specification, which corresponds to an output-side angular position at the respective joint, and wherein the secondary output variables further comprise the six-axis determination of forces and torques at the end effector provided by the artificial intelligence, namely a three-component force and a three-component torque at the end effector, the artificial intelligence using a Feedback control loop is trained by a measured value of an end effector-side torque of a six-axis force-torque sensor, which compares the output-side torque with this and, if compromised, stops the training, with a precise indirect six-axis determination of forces and torques on one End effector using the following data sets as input data: Data sets comprising data relating to an input-side angular position and an output-side angular position of a respective joint actuator gear (203), for six joint actuators of a robot arm, wherein a measured value of an end effector-side torque, measured by a six-axis -Force-torque sensor, further used as a target value as part of the feedback control loop in which the artificial intelligence is arranged, the training being operated continuously, but the training being aborted and/or interrupted when it is determined that that the value of the output torque measured by the six-axis force-torque sensor is compromised or the corresponding signal breaks off. Artificial intelligence after Claim 1 , set up for unsupervised learning with a multi-layer perceptron. Artificial intelligence after Claim 1 or 2 , wherein the input variables, per joint, further include: angular velocity, speed, acceleration and / or temperature. Method for training an artificial intelligence, namely a machine learning model or machine learning algorithm, according to one of the preceding claims through unsupervised learning or reinforcement learning by means of a precise indirect six-axis determination of forces and torques on an end effector using the following data sets as input data: Data sets comprising data relating to an input-side angular position and an output-side angular position of a respective joint actuator gear (203) for six joint actuators of a robot arm. Procedure according to Claim 4 , wherein a measured value of an end effector torque measured by a six-axis force-torque sensor is further used as a target value as part of a feedback control loop in which the artificial intelligence is arranged. Procedure according to Claim 5 , wherein the training is operated continuously, but the training is aborted and / or interrupted if it is determined that the value of the output torque measured by the six-axis force-torque sensor is compromised or the corresponding signal breaks off. Method for precisely determining an output-side torque of an actuator gear (203) of a joint (100) of a collaborative robot using an artificial intelligence, namely a machine learning model or machine learning algorithm, which is set up to output one or more output variables based on input variables, wherein the input variables of the artificial intelligence include: a first angle specification, which corresponds to an angular position on the input side, a second angle specification, which corresponds to an angular position on the output side, and wherein further the output variables include the outgoing torque determined by the artificial intelligence, the artificial intelligence using a method according to one of the Claims 4 until 6 is and/or has been trained, particularly in continuous operation. Procedure according to Claim 7 , wherein the first angle information was provided by an angular position encoder (301) on the input side of a gear, namely an absolute value encoder, and the second angle information was provided by an angular position encoder (302) on the output side of a gear, namely an incremental rotary encoder. Procedure according to Claim 7 or 8th , wherein the gear (203) comprises a voltage wave gear, wave gear and / or sliding wedge gear. A computer program comprising instructions which, when the program is executed by a computer, cause it to carry out the method according to one of the Claims 1 until 9 to carry out. Computer, computer system or computer network that are set up to locally or non-locally use the method according to one of the Claims 1 until 9 to effect.

Images