EA038279B1 - Method and system for grasping an object by means of a robotic device - Google Patents

Abstract

The invention generally relates to the field of robotic devices and in particular to methods and systems for providing grasping different objects by means of robotic devices. A method of grasping an object by means of a robotic device includes the steps of training the robotic device with the aid of at least one machine learning algorithm for identifying and remembering object grasp points, wherein the training is performed on data describing photographic images of objects and corresponding 3D models of said objects in different perspectives; forming at least one grasp point of an object represented on a graphic image of the object; obtaining a photographic image of the object using a camera of the robotic device, and determining a perspective that depicts the object grasp point formed; obtaining a three-dimensional object point cloud in the determined perspective by means of a depth sensor of the robotic device; determining the location of the object grasp point according to the point cloud obtained; determining the orientation and position of a grasping device of the robotic device in the grasp point found, wherein the determining is performed by means of calculating the rotation matrices of the robotic device; grasping the object by means of the robotic device on the basis of a calculation of an average rotation matrix for the given grasp point. The technical result is an increase in the accuracy with which a grasp region is identified on an object by a robotic device by means of determining the orientation of a grasping device on the basis of a calculation of a rotation matrix to grasp the object for the given grasp point.

Description

Technology area

The claimed technical solution generally relates to the field of robotic devices, and in particular to methods and systems for ensuring the capture of various objects using robotic devices.

State of the art

The use of robotic devices for manipulating various objects is widely known from the state of the art and is widely used in the field of industrial robots in various industries.

The main difficulty in the process of using robotic devices when manipulating various types of objects is the need for a complex learning process, which consists in creating graphic images of objects with their exact geometric parameters and subsequent debugging of robotic manipulators to interact with them.

Currently, for the learning process of object capture, various approaches are proposed based on algorithms and machine learning models, in particular, using neural networks, with the help of which robotic manipulators are trained to capture objects using pre-trained capture points. This approach is disclosed in Dense Object Nets: Learning Dense Visual Object Descriptors By and For Robotic Manipulation (Peter R. Florence et al., CSAIL, Massachusetts Institute of Technology, 09/07/2018). The known solution proposes to use point clouds of objects in the process of teaching robotic devices in order to form object descriptors based on photographic and 3D models of such objects for the purpose of further specifying points of capture and interaction with objects at these points.

The main disadvantage of the known solution is to limit the ability to grip various types of objects, which is due to the lack of analysis of the rotation of the gripper of the robot. Taking into account the fact that the object can be in different positions during the operation of the robotic device, it is necessary to evaluate the ideal position of the gripper for more accurate and correct interaction with the object.

Disclosure of invention

The technical problem or technical problem to be solved is to provide a new process of interaction with objects by means of their capture by robotic devices.

The technical result achieved by solving the above technical problem is to improve the recognition accuracy of the object gripping area by the robotic device by determining the orientation of the gripping device based on the calculation of the rotation matrix for the purpose of gripping the object at the desired gripping point.

The claimed technical result is achieved by means of a method for capturing an object using a robotic device, containing the stages at which the robotic device is taught using at least a machine learning algorithm for recognizing and memorizing the capture points of objects, and the training is performed on data characterizing photographic images of objects and corresponding 3D models of objects in various angles;

form at least one gripping point of the object represented on the graphical image of the object;

obtaining a photographic image of the object using the camera of the robotic device and determining the angle displaying the formed gripping point of the object;

obtaining a three-dimensional cloud of points of the object in the identified perspective using the depth sensor of the robotic device;

determine by the obtained cloud of points the location of the point of capture of the object;

determining the orientation and position of the gripping device of the robotic device at the detected gripping point, and the determination is carried out by calculating the matrices of rotations of the robotic device;

the object is gripped by a robotic device based on the calculation of the averaged rotation matrix for a given gripping point.

In a particular embodiment of the method, the robotic device is a robotic arm or a robotic arm.

In another particular embodiment of the method, at the stage of training, two or more points of gripping of the object are additionally indicated on the corresponding 3D model of the object.

In another particular embodiment of the method, at the training stage, the position of the robotic device at the gripping point is additionally indicated and stored depending on the position of the object.

In another particular embodiment of the method, the location of the capture point in the image is determined using a machine learning model.

In another particular embodiment of the method, the principle of gripping the object is additionally determined depending on the type of gripping device and / or the type of object.

- 1 038279

In another particular embodiment of the method, for several points of object gripping, the optimal gripping point of the object is indicated based on its relative position relative to other gripping points and relative to the gripping device at the moment of gripping.

In another particular embodiment of the method, the robotic arm is calibrated against the surface of interaction with the object.

In another particular embodiment of the method, calibration is performed by applying graphical augmented reality codes to the surface.

The claimed technical result is also achieved by a system for gripping an object using a robotic device, which contains a robotic device containing at least one camera, at least one depth sensor and a gripping device;

a computing device connected to the robotic device, the computing device interacting with a machine learning model that has been trained on photographic images of objects and corresponding 3D models of objects in various angles to recognize and store gripping points of objects;

to form at least one gripping point of the object represented on the graphical image of the object;

receive a photographic image of the object from the camera of the robotic device and determine an angle displaying the generated gripping point of the object;

receive a three-dimensional cloud of points of the object in the identified angle using the depth sensor of the robotic device;

determine the location of the capture point of the object by the received cloud of points;

determine the orientation and position of the gripping device at the detected gripping point, and the determination is carried out by calculating the matrices of the rotations of the robotic device;

generate a command for gripping an object using a robotic device based on the calculation of the averaged rotation matrix for a given gripping point.

In a particular implementation of the system, the robotic device is a robotic arm or a robotic arm.

In another particular implementation of the system, the computing device is a computer, system on a chip (SoC), or server.

In another particular implementation of the system, the computing device is wired or wirelessly connected to a robotic device.

In another particular implementation of the system, the machine learning model is at least one neural network.

In another particular embodiment of the system, the machine learning model is made with the possibility of automatic retraining based on new data about objects.

In another particular implementation of the system, new data about objects is transmitted from a cloud server.

In another particular implementation of the system, the neural network is located on a cloud server.

Brief Description of Drawings

The features and advantages of the present technical solution will become apparent from the following detailed description and accompanying drawings.

FIG. 1 illustrates a general view of the claimed system.

FIG. 2 illustrates a schematic view of a manipulator of a robotic device.

FIG. 3 illustrates a flow diagram of a general robotic device teaching procedure.

FIG. 4 illustrates an example of teaching a robotic device using images and 3D models of an object.

FIG. 5 illustrates an example of teaching a robotic device using multiple gripping points on site.

FIG. 6 illustrates a block diagram of the implementation of the claimed method.

FIG. 7 illustrates an example of centering the gripper to determine the desired gripping point on an object.

FIG. 8 illustrates an example of a real-time imaging system.

FIG. 9 illustrates an example of a computing system.

Detailed description of the implementation of the invention

This technical solution can be implemented on a computer, in the form of an automated information system (AIS) or a computer-readable medium containing instructions for performing the above method. The technical solution can be implemented as a distributed computer system that can be installed on a centralized server (set of servers). User access to the system is possible both from the Internet and from the internal network of an enterprise / organization through a mobile communication device on which software with a corresponding graphical user interface is installed, or a personal computer

- 2 038279 pa with access to the web version of the system with the corresponding graphical user interface. Below will be described the terms and concepts necessary for the implementation of this technical solution.

In this solution, the system means a computer system, a computer (electronic computing machine), CNC (numerical control), PLC (programmable logic controller), computerized control systems and any other devices capable of performing a given, well-defined sequence of computational operations (actions, instructions ). A command processing device means an electronic unit or an integrated circuit (microprocessor) that executes machine instructions (programs). A command processor reads and executes machine instructions (programs) from one or more storage devices. The role of data storage devices can be, but are not limited to, hard disks (HDD), flash memory, ROM (read only memory), solid state drives (SSD), optical drives.

A program is a sequence of instructions intended for execution by a computer control device or a command processing device.

As shown in FIG. 1, the system (100) for implementing the claimed solution comprises a computing device (110) that provides the necessary computational processing of the algorithm for the operation of the robotic device (200), a data transmission channel (130) that provides processes for receiving / transmitting data between the elements of the system (100) as well as a remote computing device (120), such as a cloud server for specialized tasks.

As a computing device (110), with which the user (10) interacts, can be used, for example, a personal computer, laptop, tablet, server, smartphone, server cluster, mainframe, supercomputer, thin client, system on a chip (English System on a Chip, abbreviated SoC), etc. The specific hardware implementation of the computing device (110) depends on the requirements and must provide the necessary computational processing of digital information to ensure the proper functioning of the device (110) as part of the system (100).

As a data transmission channel (130), a different type of communication can be used, for example, a wired and / or wireless type, in particular LAN, WAN, WLAN, Bluetooth, Wi-Fi, GSM / GPRS / LTE / 5G, Ethernet, USB, RS232, RJ45 etc. The remote computing device (120), as a rule, is a cloud storage or a server for solving specialized tasks or program logic. The description of the operation of the device (120) will be described later in the present description.

As a robotic device (200) can be used as a stationary robotic arm for industrial or collaborative purposes, for example, Universal Robotics ™, KUKA ™, Fanuc ™, ABB ™, etc., and robotic manipulators installed on mobile autonomous robotic devices, as limbs, in particular robotic arms.

FIG. 2 is a schematic diagram of an example of a robotic device (200) in the form of a manipulator equipped with a gripper (210) and a module containing an RGB camera (211) and a camera with a depth sensor (212) (also called a depth camera or RGB-D camera). The RGB camera (211) is used to obtain two-dimensional graphic images of the object (20), with which it is supposed to interact using a robotic device (200). The depth camera (212) is used to obtain 3D models of the object (20), as well as to center the gripper (210) and determine the distance to the object (20).

As a gripping device (210), various types of grips can be used, for example, in the form of pincers, anthropomorphic hands, etc. The gripper typically provides 6 degrees of freedom for manipulating an object.

As shown in FIG. 1, the object (20) with which the robotic device (200) interacts can be placed on a surface on which augmented reality markers (201) are applied, which are used to calibrate the robotic device (200) in order to form its area of action (205), in which the object (20) can be captured. Augmented reality markers (201) are graphical augmented reality codes, also called AR-codes, which are used to limit the surface of interaction with an object (20). Object (20) can represent a wide range of products having a soft (deformable) or hard shape. FIG. 3 shows the learning process (300) of a robotic device (200). For training, the approach from the above scientific publication is applied. At the first step (301), the object (20) is captured using an RGB camera (211). Shooting is carried out from various angles. Then, using the depth camera (212), a cloud of points (302) corresponding to the angle is created at each of the obtained angles. Based on the point clouds formed at step (302), a 3D model of the object (20) is built by combining the obtained point clouds for each of the survey angles at step (301). The resulting set of camera positions (211), correlated with the images of the object obtained at the corresponding angles at step (301), as well as the final 3D model of the object (20), obtained at step (303), are used for automatic preparation

- 3,038279 training set for training a machine learning model at stage (304), in particular, a convolutional neural network (CNN).

FIG. 4 shows an example of obtaining views (P1) - (P4) of images of an object (20). For each angle (P1) - (P4), using the depth camera (212), a corresponding model (M1) - (M4) is built based on the resulting point cloud. All point clouds (M1) - (M4), merged into a single whole, form a 3D model of the object (21). As a result, according to the obtained models of the angles (M1) - (M4), the final 3D model of the object (21) is created by merging all the obtained point clouds (M1) - (M4) by the angles (P1) - (P4) of the object images. It should be noted that the number of views of images of the object (20) can be any and not limited to the given example of the number of views. This approach allows for a selected point on the image of an object, for example a photograph or a three-dimensional model, to find the corresponding 3D point on the surface of the object model to capture it.

Using a machine learning model, a mapping is performed from a set of images of a given size to a set of descriptors selected during training in such a way that for two points from two different images, the distance between the corresponding descriptors will be less than a certain value if these points are projections of the same points on the surface of the 3D model of the object (21). This operation is performed several times for different positions of the object (20) in the reach (205) of the robotic device (200). When constructing a training sample for a machine learning model, for each photograph of an object (20) from the corresponding angle, one point is selected, for each of which there is a corresponding 3D point from the surface of the object model, obtained from the point cloud of the corresponding angle of shooting of the object (20). As an example, we can consider pairs of photographs of one object (20) with the corresponding points. If a pair of 3D points on the corresponding 2D images belong to the same voxel / vertex of the 3D model of the object (21), then these points are considered to be corresponding (coinciding), otherwise inconsistent (mismatched). Thus, from all photos, you can get millions of pairs of pixels, both coinciding and non-coinciding. Based on this information, a training sample is constructed for training a neural network.

In the operating mode, photographs of the object (20), consisting of pixels, are fed to the input of the neural network, at the output to each pixel there corresponds a descriptor (a sequence of numbers of the same length n, or a point in n-dimensional space on which the metric is set - the function of calculating the distance between the points of this space). The neural network is trained on a set of pairs of pixels from the training sample so that for matching pixels, the distance between their corresponding descriptors is less than a certain threshold, and for non-matching ones, more. FIG. 5 shows an example of teaching positions of gripping an object (20) by specifying several gripping points (25, 26) of an object (20) using a gripper (210). When teaching the robotic device (200) to interact with objects (20) by the gripping points (25, 26), a mutual orientation of the segments (22) is formed between the selected gripping points (25, 26) and the orientation of the gripping device (210). Further, for pairs of segments (22) between the capture points on the initial data and the corresponding found capture points (25, 26) on the object (20) when shooting it with the RGB camera (211) and the depth camera (212), the rotation matrices are calculated. For all pairs of segments (22), an averaged rotation matrix is calculated, which is applied to the saved configuration corresponding to the previously specified gripping point, and the gripper (210) is set in motion to grip the object at the selected point.

At the stage of training (300) of the robotic device (200), the gripper (210) is additionally trained to interact with objects (20), depending on their shape and degree of deformation, as well as depending on the configuration of the gripper (210). The configuration and gripping principle of the device (210) play an important role in the analysis of the position and the possibility of interaction with the object (20), depending on its location in the reach (205) of the device (200). In the case of working with soft or easily deformable objects (20), for example, plush toys, bags, tissue packages, etc., the capture principle can be acceptable for working with an object (20) by capturing it in any possible way at the specified capture point ( 25).

If the interaction is carried out with a solid object (20), for example, a book, telephone, box, etc., then the gripper (210) is trained to interact with an admissible part of the object (20) (in order to avoid violation of its integrity or unacceptable deformation), and also depending on the configuration of the installed gripper (210). For example, the anthropomorphic grips (210) in the form of a robotic arm have a greater degree of ability to interact with the desired grip point (25) of the object (20) compared to the pincer gripper shown in FIG. 2.

The learning process of the robotic device (200) can also be performed using an external controller, for example, a glove controller, which makes it possible to memorize the principle of interaction with an object (20) by manipulating the operator's movements (10).

FIG. 6 shows a process of performing a method (400) for gripping an object (20) by a robotic device (200). A description of the operating process (400) of the robotic device (200) will also be described with reference to FIG. 7. After passing the above-described learning process (300) of the robotic device (200), its operation can be carried out in an automated

- 4 038279 nominal mode. The operator (10), using the computing device (110), generates (step 401) the capture point (25), for example, by pointing to the object (23) on the image using a graphical user interface (GUI). Thus, a pixel or a group of pixels corresponding to the capture point (25) or the vicinity of the location of this point (25) on the object (20) is marked on the image (23).

Based on the received information about the capture point (25), the robotic device (200) reviews and photographs (step 402) using the RGB camera (211) of the capture object (20). The robotic device (200) takes a series of photographs (no more than 10 shots) from several different positions (angles) of the camera (211). Using a machine learning model, in particular a trained SNS, from a series of photos received from the camera (211), one with the most probable coordinates of the location of a given point of interest (25) is selected, after which the robotic device (200) places the camera (211) in the position with which photo was taken (step 403). Based on the processing of the obtained photographs of the object (20) for each photograph from this series, the SNS calculates descriptors, among which the search and finding of the descriptor is carried out, the distance from which to the pixel coordinate of the capture point (25) is minimal.

Next, the point of interest is localized (step 404) on the surface of the 3D model of the object (21) using the point cloud obtained from the depth camera (212), as shown in FIG. 7. To accurately localize the capture point (25) (step 405), several point clouds are integrated over a certain period of time, and the point in the obtained point cloud is located closest to the straight line drawn from the optical center (27) of the camera (212) and passing through the plane camera images in the projection of the point of interest (25).

Next, at step (406), the calculation of the rotation matrices for the capture of the object (20) is performed, and the position and trajectory of the movement of the device (210) for the capture (step 407) of the object (20) are analyzed by calculating the averaged rotation matrix of the gripper (210).

As shown in FIG. 5, in the process of learning the options for gripping by the device (210), several points (25, 26) are set on the object (20), relative to which the position and orientation of the gripping device (210) are memorized. In the case of gripping an object (20) located in an arbitrary position, the specified points (25, 26) are recognized on the object (20), and the rotation matrices of the gripper (210) are calculated so that the relative orientation of the gripping points (25, 26) and gripper (210) matched the stored values during training.

In the case when grippers (210) with a large number of degrees of freedom are used, a series of positions, orientations and configurations of the gripper (210) are stored during the formation of the gripper at a given point at the training stage. After calculating the rotational matrices of the gripper (210) for each element of this series using inverse kinematics algorithms, the robotic device (200) is brought to the appropriate position. When planning the trajectory of movement of the robotic device (200), the appropriate algorithm takes into account possible collisions both between the nodes of the device itself (200) and with surrounding objects.

The specified machine learning model can be contained both directly on the device (110) associated with the robotic device (200) and located on a remote device (120), for example, a server, and perform data processing on request generated from the computing device (110) on based on data received from the robotic device (200). As shown in FIG. 8, in a particular example of the implementation of the system (100), several computing devices can be used, each of which is designed to implement the required functionality. For example, an image obtained from a smartphone camera (110) can be used as an initial image (23) for forming a gripping point (25). After acquiring the image (23), the user (10) specifies the desired gripping point (25) for the corresponding object (20) with which the robotic device (200) interacts. The device (200) can contact the server (120) to initiate the processing of data about a similar type of object (20), which is in the area of operation of the robotic device (205).

Data processing for object recognition (20) and identification of the required capture point (25) is performed using a trained neural network, which can be located on the server (120) and updated as new data on the corresponding types of objects (121) and options for interaction with them arrive. ... The model library (121) can be stored directly on the server (120) or a third-party resource. At the same time, upon receipt of new information about the models of objects, the SNS can be automatically retrained to expand the classification options and increase the process of autonomous operation of a robotic device (200). This process can further improve the interactions of a robotic device (200) with specified types of objects, which will expand its application in specified industries.

The robotic device (200) can also be configured to automatically work with a given type of object by training gripping for a certain part of the object (20). As shown in FIG. 8, a circle can be used as an object of interaction (20). Depending on the type of gripper (210) installed, the robotic device (200) can be

- 5 038279 the phenomenon of the object (20) in the area of its operation (205) to automatically recognize the location of the gripping area of the object (20), for example a handle or a body. Depending on the type of installed gripper (210), possible models and positions are determined for carrying out the procedure for gripping an object (20).

FIG. 9 shows a general view of the computing system (500). On the basis of the presented system (500), a different range of computing devices can be implemented, for example, a device (110), a server (120), etc. In general, the system (500) contains one or more processors (501) united by a common bus of information exchange, memory means such as RAM (502) and ROM (503), input / output interfaces (504), input / output devices (505 ) and a device for networking (506).

The processor (501) (or multiple processors, multi-core processor, etc.) can be selected from a range of devices currently widely used, for example, such manufacturers as Intel ™, AMD ™, Apple ™, Samsung Exynos ™, MediaTEK ™, Qualcomm Snapdragon ™ etc. Under the processor or one of the processors used in the system (500), it is also necessary to take into account a graphics processor, for example, NVIDIA GPU or Graphcore, the type of which is also suitable for full or partial execution of methods (300, 400), and can also be used for training and applying models machine learning in various information systems.

RAM (502) is a random access memory and is intended for storing machine-readable instructions executed by the processor (501) for performing the necessary operations for logical data processing. RAM (502), as a rule, contains executable instructions of the operating system and corresponding software components (applications, software modules, etc.). In this case, the available memory of the graphics card or graphics processor can act as RAM (502).

ROM (503) is one or more means for permanent storage of data, such as a hard disk drive (HDD), solid state data storage device (SSD), flash memory (EEPROM, NAND, etc.), optical storage media (CD-R / RW, DVD-R / RW, BlueRay Disc, MD), etc.

To organize the operation of system components (500) and to organize the operation of external connected devices, various types of I / O interfaces (504) are used. The choice of appropriate interfaces depends on the specific design of the computing device, which can be, but are not limited to, PCI, AGP, PS / 2, IrDa, FireWire, LPT, COM, SATA, IDE, Lightning, USB (2.0, 3.0, 3.1, micro, mini, type C), TRS / Audio jack (2.5, 3.5, 6.35), HDMI, DVI, VGA, Display Port, RJ45, RS232, etc. To ensure user interaction with the computing system (500), various means (505) of I / O information are used, for example, a keyboard, display (monitor), touch display, touchpad, joystick, mouse manipulator, light pen, stylus, touch panel, trackball, speakers , microphone, augmented reality, optical sensors, tablet, light indicators, projector, camera, biometric identification (retina scanner, fingerprint scanner, voice recognition module), etc.

The networking tool (506) provides data transmission via an internal or external computer network, such as an Intranet, Internet, LAN, and the like. As one or more means (506), an Ethernet card, a GSM modem, a GPRS modem, an LTE modem, a 5G modem, a satellite communication module, an NFC module, a Bluetooth and / or BLE module, a Wi-Fi module, and etc. Additionally, satellite navigation means can also be used as part of the system (500), for example, GPS, GLONASS, BeiDou, Galileo.

The specific choice of system elements (500) for the implementation of various software and hardware architectural solutions can vary while maintaining the required functionality provided from a particular type of device. The presented materials of the invention disclose preferred examples of the implementation of the technical solution and should not be construed as limiting other, particular examples of its implementation, which do not go beyond the scope of the claimed legal protection, which are obvious to specialists in the relevant field of technology.

Claims (17)

CLAIM 1. A method of capturing an object using a robotic device, comprising the steps of teaching a robotic device using at least one machine learning algorithm to recognize and memorize capture points of objects, and training is performed on data characterizing photographic images of objects and corresponding 3D models objects in different angles; at least one gripping point of the object represented on the graphical image of the object is formed; taking a photographic image of the object using the camera of the robotic device and determining the angle displaying the formed gripping point of the object; a three-dimensional cloud of points of an object is obtained in a detected perspective using a depth sensor of a robotic device; - 6 038279 determine the location of the capture point of the object by the obtained point cloud; determining the orientation and position of the gripping device of the robotic device at the detected gripping point, and the determination is carried out by calculating the matrices of the rotations of the robotic device; the object is gripped by a robotic device based on the calculation of the averaged rotation matrix for a given gripping point. 2. The method according to claim 1, characterized in that the robotic device is a robotic arm or a robotic arm. 3. The method according to claim 1, characterized in that at the training stage, two or more points of gripping of the object are additionally indicated on the corresponding 3D model of the object. 4. The method according to claim 1, characterized in that at the stage of training, the location of the robotic device at the gripping point is additionally indicated and stored depending on the location of the object. 5. The method according to claim 1, characterized in that the location of the capture point in the image is determined using a machine learning model. 6. The method according to claim 5, characterized in that the principle of gripping the object is additionally determined depending on the type of gripping device and / or the type of object. 7. The method according to claim 3, characterized in that for several points of object gripping, the optimal gripping point of the object is indicated based on its relative position relative to other gripping points and relative to the gripping device at the moment of gripping. 8. The method according to claim 2, characterized in that the robotic arm is calibrated against the surface of interaction with the object. 9. The method according to claim 8, characterized in that the calibration is performed by applying graphical augmented reality codes to the surface. 10. A system for gripping an object using a robotic device, comprising a robotic device containing at least one camera, at least one depth sensor and a gripping device; a computing device connected to a robotic device, and the computing device interacts with a machine learning model that is trained on photographic images of objects and corresponding 3D models of objects in various angles to recognize and memorize gripping points of objects, configured to form at least one gripping point the object represented on the graphical image of the object; to take a photographic image of the object from the camera of the robotic device and determine the angle that displays the formed gripping point of the object; to acquire a three-dimensional cloud of points of the object in the identified angle using the depth sensor of the robotic device; determine the location of the capture point of the object by the obtained cloud of points; determine the orientation and position of the gripper at the detected gripping point, the determination being made by calculating the rotation matrices of the robotic device; generate a command for gripping an object using a robotic device based on the calculation of the averaged rotation matrix for a given gripping point. 11. The system of claim 10, wherein the robotic device is a robotic arm or a robotic arm. 12. The system of claim 10, wherein the computing device is a computer, a system on a chip (SoC), or a server. 13. The system of claim 12, wherein the computing device is wired or wirelessly connected to the robotic device. 14. The system of claim 10, wherein the machine learning model is at least one neural network. 15. The system according to claim 14, characterized in that the machine learning model is made with the possibility of automatic retraining based on new data about objects. 16. The system according to claim 15, characterized in that new data about objects is transmitted from a cloud server. 17. The system according to claim 14, characterized in that the neural network is located on a cloud server.