阅读说明：本技术 生成用于机器人场景的三维模型的方法和系统 (Method and system for generating three-dimensional model for robot scene ) 是由 张飚 R·博卡 C·莫拉托 C·马蒂内兹 汪建军 滕舟 黄金苗 M·瓦尔斯特罗姆 J 于 2019-06-27 设计创作，主要内容包括：本公开的实施例涉及生成用于机器人场景的三维模型的方法和系统。机器人被配置为使用用于生成足以确定无碰撞路径并且识别工业场景中的对象的3D模型的方法来在对象上执行任务。该方法包括确定预定义的无碰撞路径并且扫描机器人周围的工业场景。工业场景的被存储的图像从存储器中被检索并且被分析以构建新的3D模型。在新的3D模型中检测到对象之后,机器人可以在沿无碰撞路径移动的同时进一步扫描工业场景中的图像,直到对象以预定义的确定性级别被识别。然后机器人可以在对象上执行机器人任务。(Embodiments of the present disclosure relate to methods and systems for generating three-dimensional models for robotic scenes. The robot is configured to perform a task on an object using a method for generating a 3D model sufficient to determine a collision-free path and identify the object in the industrial scene. The method includes determining a predefined collision-free path and scanning an industrial scene around the robot. Stored images of the industrial scene are retrieved from memory and analyzed to construct a new 3D model. After detecting the object in the new 3D model, the robot may further scan the images in the industrial scene while moving along the collision-free path until the object is identified with a predefined level of certainty. The robot may then perform a robot task on the object.)

1. A method, comprising:

determining a predefined collision-free robot path;

moving a robot along the predefined robot path;

scanning an industrial scene with a scanning sensor positioned on a robot while moving along the predefined robot path;

storing the scanned image of the industrial scene in a memory;

constructing a 3D model of the industrial scene based on the images stored in the memory;

planning a next collision-free robot path based on the 3D model;

moving the robot along the next collision-free robot path;

scanning an industrial scene with the scanning sensor positioned on the robot while moving along the next robot path; and

storing the new scanned image in the memory; and

reconstructing the 3D model of the industrial scene based on the new scanned image.

2. The method of claim 1, further comprising repeating the planning, scanning, storing, and reconstructing steps until a complete 3D industrial scene is constructed.

3. The method of claim 2, further comprising performing a work task with the robot after completing the 3D industrial scene model.

4. The method of claim 1, wherein the scanning sensor is a 3D camera.

5. The method of claim 1, further comprising moving the scanning sensor with respect to the robot when determining the predefined collision-free path.

6. The method of claim 5, wherein the movement of the scanning sensor comprises pan, tilt, rotation, and translation motions with respect to the robot.

7. The method of claim 5, wherein the moving of the scanning sensor comprises moving an arm of the robot while a base of the robot remains stationary.

8. The method of claim 1, further comprising planning the collision-free path with a controller having a collision-free motion planning algorithm.

9. The method of claim 1, wherein the planning of the collision-free path occurs in real-time without off-line computer analysis.

10. A method, comprising:

determining a predefined collision-free robot path;

scanning an industrial scene along the predefined collision-free robot path with a scanning sensor positioned on a robot;

storing the scanned image of the industrial scene in a memory;

constructing a 3D model of the industrial scene based on the images stored in memory;

detecting an object within the 3D model of the industrial scene;

moving the robot along the collision-free robot path to generate a next scanning viewpoint of the detected object;

scanning the industrial scene to obtain a new scanned image of the object;

storing the new scanned image in the memory; and

repeating the moving and scanning steps until the object is identified to a threshold level of certainty.

11. The method of claim 10, wherein the scanning comprises capturing an image with a 3D camera.

12. The method of claim 10, further comprising performing a robotic task on the object after the object has been identified to the threshold level of certainty.

13. The method of claim 12, wherein the robotic task comprises grasping the object.

14. The method of claim 10, further comprising panning, tilting, and rotating the scanning sensor to capture images from different vantage points to generate a new 3D model of the industrial scene.

15. The method of claim 14, further comprising planning a next scan path prior to generating the new 3D model of the industrial scene.

16. The method of claim 15, wherein the planning comprises analyzing the new 3D model with a controller having a collision-free motion planner algorithm.

17. The method of claim 16, further comprising determining the next scan path based on results from a collision-free motion planning algorithm.

18. A method, comprising:

determining a predefined collision-free robot path;

scanning an industrial scene close to the robot by using a scanning sensor;

storing the scanned image of the industrial scene in a memory;

constructing a 3D model of the industrial scene based on the images stored in memory;

detecting an object within the industrial scene;

determining whether the object is identified with sufficient accuracy;

determining whether a robotic task can be performed on the object; and

after the object is identified with sufficient certainty, one or more robotic tasks are performed on the object.

19. The method of claim 18, further comprising: if the object is not identified with sufficient certainty, a next scanning viewpoint is generated and the industrial scene is rescanned.

20. The method of claim 18, further comprising: if the 3D model of the industrial scene is insufficient for a gripping analysis, a next scanning viewpoint is generated and the industrial scene is rescanned.

21. The method of claim 18, further comprising: if the 3D model of the industrial scene is incomplete, a next scanning viewpoint is generated and the industrial scene is rescanned.

22. The method of claim 18, wherein the scanning of the industrial scene comprises panning, tilting, rotating, and translating the scanning sensor with respect to the robot.

23. The method of claim 18, further comprising planning, by a controller having a collision-free motion planning algorithm, the collision-free path.

24. The method of claim 23, wherein the planning occurs in real-time without off-line computer analysis.

Technical Field

The present application relates generally to modeling industrial scenes by robots, and more particularly, but not exclusively, to building 3D models of industrial scenes using scanning with visual sensors associated with robots.

Background

With the continued development in the field of robotics, more and more attention has been directed to the development of techniques that allow robots to determine collision-free paths and the position of workpieces or other objects in real time. Randomly placed objects within a robot work area or industrial scene may interfere with certain movements of the robot and prevent work tasks from being completed. Some existing systems have various drawbacks with respect to certain applications. Therefore, there is still a need for further contributions in this area of technology.

Disclosure of Invention

One embodiment of the present application is a unique system and method for generating a real-time 3D model of a robotic work area or industrial scene. Other embodiments include apparatuses, systems, devices, hardware, methods, and combinations for generating collision-free paths for robotic operation in an industrial setting. Further embodiments, forms, features, aspects, benefits, and advantages of the present application will become apparent from the description and drawings provided herein.

Drawings

FIG. 1 is a schematic view of a robotic system according to an exemplary embodiment of the present disclosure;

fig. 2 is a schematic illustration of an industrial robot scenario according to an exemplary embodiment of the present disclosure;

FIG. 3 is a flow diagram for a method for generating a scan path for collision free robot motion according to one embodiment of the present disclosure;

FIG. 4 is a flow diagram for a method for identification of objects in an industrial scene;

FIG. 5 is a flow diagram for a method for generating a 3D model sufficient to reduce ambiguity of an object such that a work task may be performed on the object by a robot;

FIG. 6 is a flow chart for a method for improving object recognition and planning a next scan path for collision-free motion so that the robot can perform tasks on detected objects; and

fig. 7A and 7B define a flow chart for a method for improving object recognition, improving grasping confidence, and planning a next scan path for collision-free motion.

Detailed Description

For the purposes of promoting an understanding of the principles of the application, reference will now be made to the embodiments illustrated in the drawings and specific language will be used to describe the same. It will nevertheless be understood that no limitation of the scope of the application is thereby intended. Any alterations and further modifications in the described embodiments, and any further applications of the principles of the application as described herein are contemplated as would normally occur to one skilled in the art to which the application relates.

With the continued development of the field of robotics, more and more attention is being paid to the development of technologies that allow for more tightly coupled human-machine interaction (HRI). The application of HRI techniques helps the robot understand information about its surroundings and allows the operator to understand or receive feedback about the level of understanding the robot has reached. An initial understanding of the work environment or industrial scenario may be obtained before interaction between the operator and the robot occurs. As the robot's understanding of the scene increases, the level of human interaction may decrease (i.e., the operator does not have to program all information into the robot before operation, which minimizes setup time).

The robot system disclosed herein provides a control method to retrieve valid information from an industrial scene that can be identified by the memory of the robot. The control method enables the robot to obtain information and understand elements or objects in the scene while optimizing robot motion to perform tasks within the industrial scene. The robot path reacts to changes in the scene and helps the robot understand the surrounding boundaries. The ability of the robot to autonomously retrieve information from the scene facilitates detection of objects, constructive robot motion generation, reduction of time required for the overall discovery process, and minimizes human involvement setup and robot programming. Industrial robots may use teach boxes and joysticks to "jog the robot" and teach the robot points, but this can be cumbersome, time consuming and dangerous if the operator is very close to the robot as it moves through the industrial scene. 3D vision and implicit programming can be used to improve robot path programming by generating 3D models of unknown scenes around the robot, which requires time to teach the robot to hold the 3D sensor to scan an industrial scene. It may be difficult to manually generate scan paths to collect sufficient data from scenes and objects without robot collisions and/or causing accessibility issues.

The present disclosure includes methods and systems for automatically generating a complete 3D model of an industrial scene for robotic applications, which reduces engineering time and costs compared to manually programmed robots. The method and system automatically generate scan paths to collect sufficient data about the scene and object locations without causing collision or reachability problems. The robot scanning system is operatively connected to the robot path planning algorithm, to the 3D object recognition algorithm, for providing control inputs to the robot to perform tasks on objects within the industrial scene.

Referring now to FIG. 1, an exemplary robotic system 10 is shown in an exemplary work environment or industrial setting. It should be understood that the robotic systems illustrated herein are exemplary in nature, and that variations in robotic and/or industrial scenarios are contemplated herein. The robotic system 10 may include a robot 12 having a vision system 36, the vision system 36 having one or more cameras 38. In one form, one or more of the cameras 38 may be mounted on one of the movable arms 16a, 16b of the robot 12. In other forms, one or more cameras 38 may be positioned separate from the robot 12. A control system 14, including an electronic controller having a CPU, memory, and input/output system, is operatively coupled to the robot 12 and to the vision system 36. Control system 14 is operable to receive and analyze images captured by vision system 36 and other sensor data for operation of robot 12. In some forms, the control system 14 is defined within a portion of the robot 12.

The robot 12 may include a movable base 20 and a plurality of movable portions coupled thereto. The movable portion may translate or rotate in any desired direction. By way of example and not limitation, the exemplary robot 12 may employ movable portions illustrated by arrows 18, 26, 28, 30, 32, and 34. A bin 40 for holding workpieces or other objects 42 to be retrieved and/or to be manipulated by the robot 12 may constitute at least a portion of an exemplary industrial scenario. An end effector 24, such as a grasping or gripping mechanism or other work tool, may be attached to the movable arm 16a and used to grip the object 42 and/or perform other work tasks on the object 42 as desired. It should be understood that the term "case" is exemplary in nature and, as used herein, refers to, but is not limited to, any container, carton, box, tray, or other structure capable of receiving and/or holding a workpiece, part, or other object. Additional components 44 may be associated with vision system 36. These components 44 may include an illumination system, reflectors, refractors, and beam expanders, among others.

Referring now to FIG. 2, another exemplary industrial scenario 50 is shown. The vision system 36 is operable with the robot 12 to determine the location of various items within the vicinity of the robot 12. By way of example, and not limitation, scenario 50 includes: pick- up bins 60a, 60b, potential obstacles 62a, 62b, and drop bins 64a, 64 b. Vision system 36 may be used to determine the position and orientation of objects in pick bins 60a, 60b and/or drop bins 64a, 64 b. Vision system 36 may identify the object within pickup box 60a and communicate the position and orientation of the object to robot controller 14 (see fig. 1). A robot controller (not shown) then provides commands to the robot 12 to control the movement of the portion of the robot 12 including the grasping mechanism (not shown) in order to grasp and move the selected object and avoid potential obstacles.

Teaching and training robots to autonomously discover and understand industrial scenarios and perform robotic work tasks (such as extracting randomly arranged objects from bins) is a complex task. Given 2D RGB (red, green, blue) sensor images, the robot 12 may be trained to recognize pick-up bins 60a and drop bins 64a and obstacles 62a, 62b within the scene 50. Furthermore, the 2D images may be combined with calibrated 3D camera sensors to generate precise spatial locations of certain identified objects. In some aspects, 3D point cloud based scene reconstruction may be time consuming, but some portions of the 3D point cloud data may be used to reconstruct a scene for a robot trained in a virtual environment from the point cloud data to reconstruct a virtual 3D scene that may be used to train the robot in real time. The method for determining such robot commands in an autonomous manner is described in more detail below.

The control method for defining the automatic robot path starts from an initial viewpoint of the robot 12. The control method directs the robot 12 to pan, in one exemplary form, a 3D sensor that may be mounted on the tip of the robot gripper 24. Next, an initial 3D model of the scene around the robot 12 is reconstructed using the 3D sensor data and robot 12 movement information (3D sensor pose, time stamp, etc.). The initial 3D model of the scene will be analyzed for parameters such as occupancy, missing information, occluded objects, unknown regions, etc. Scene reconstruction is constantly updated during operation of the robot 12. The computer processing calculations for scene reconstruction may be done independently and in parallel with the work tasks of the robot 12, such as pick or place work tasks. If the system is unable to identify the object 42 based on previously collected information, a new scan angle may only be generated. The new scan and robot 12 movement path calculations may or may not be performed separately to complete the scene reconstruction process. A new scan angle is generated based on prior knowledge from scene understanding. Based on the observed data, a posterior prediction distribution can be calculated, and a maximum likelihood estimate of the probability can be used to determine a new perspective that will yield the most useful information to facilitate understanding of the robot 12.

Referring now to fig. 3, a flow diagram illustrates a control method 100 for generating a scan path for collision-free robot motion according to one embodiment of the present disclosure. The control method 100 starts at step 102, where the robot starts with a predefined 3D scan of the collision free path. At step 104, the robot reconstructs the scene and the environment around the robot using the 3D scanning device. The scanning device sends the scanned image to the memory so that the control system can then analyze the 3D scene model at step. Then, at step 108, the control method 100 determines whether the 3D model of the scene surrounding the robot is sufficiently complete for the robot to perform the desired task. If the answer to the query at step 108 is "no," then the control method 100 will generate the next scanning viewpoint for the 3D scanner to obtain additional images. At step 112, the control method 100 plans the next scan path with the robot collision free motion planner based on the latest 3D scene model. At step 114, the robot will scan the industrial scene along the next collision-free path to obtain scanned images from different viewpoints. At step 116, control method 100 generates a complete 3D scene model around the robot. Referring back to step 108, if the answer to the query is "yes," then the control method 100 moves to step 116 and generates a complete 3D scene model around the robot without processing steps 110, 112 and 114.

Referring now to FIG. 4, a control method 200 is shown whereby a robot learns to identify objects within an industrial scene. The control method 200 begins at step 202, where the robot determines a predefined collision-free path with a 3D scanning device. At step 204, the robot reconstructs a scene including the environment surrounding the robot. At step 206, control method 200 runs an object detection and recognition program. At step 208, the query of the control method 200 is such that it is determined whether the 3D model of the scene is sufficient to detect and identify objects to the extent necessary to perform the robotic task thereon. If the answer to query 208 is "no," then control method 200 will generate a next scan viewpoint at step 210 to reduce the ambiguity of the object using the detection and recognition procedure. At step 212, the robot will perform a 3D scan along the new collision-free path. Referring back to step 208, if the answer to the query is "yes," the control method 200 moves to step 214, where the control method 200 generates a local scene model for the identified object for the robot to perform a robot task on.

Referring now to fig. 5, a control method 300 is shown beginning at step 302, where a robot uses a 3D scan to define a predefined collision-free path. At step 304, the control method 300 directs the robot to reconstruct an industrial scene around the robot. The control method 300 then analyzes the scene construct obtained by the robot using an object detection and recognition program at step 306, and the system determines whether the scene is sufficient to positively detect and recognize the object at query 308. If the answer to query 308 is "no," then at step 310, control method 300 will direct the robot to generate additional scan viewpoints in order to reduce the ambiguity of the object. At step 312, the control method 300 plans the next scan path with the robot collision free motion planner based on the latest 3D scene model. The robot then performs a new 3D scan path at step 314. At step 316, a new local scene model is generated and the type of object is determined using the latest 3D scan on the new path. Referring back to query 308, if the answer is yes as to whether the 3D model is sufficient to detect and identify the object, then control method 300 moves down to step 316 to allow the robot to perform a robot work task on the identified object.

Referring now to fig. 6, a control method of operation 400 is provided for a robot to perform a work task (such as picking or grasping) on an object. At step 402, the robot obtains a predefined collision-free path from the 3D scan. The robot then reconstructs the industrial scene around the robot to determine whether an object is detected within the scene at step 404. At step 406, the robot will use a 3D object detection and recognition program to determine if the 3D scene surrounding the robot has the desired objects located therein, and if so, at step 408, the control method 400 runs an object gripping and positioning analysis to determine if the 3D model of the scene is sufficient for gripping objects at query 410. If the answer to query 410 is "no," then at step 412, control method 400 will generate a next scan viewpoint to improve the confidence of the grasp analysis. At step 414, control method 400 will determine a collision-free path for the robot to move along so that the vision system can obtain a 3D scan of the environment. At step 416, after determining that the 3D model is sufficient to perform such a task, the robot will perform a work task on the object (such as grasping or picking up the object). If the 3D model of the scene at 410 is sufficient for the robot to perform a work task, the control method 400 moves to step 416 and then directs the robot to perform a task on the detected object.

Referring now to fig. 7A and 7B, a control method 500 for improving object recognition, improving grip confidence, and for planning a scan path for collision-free motion of a robot is shown. The control method 500 begins at step 502, where the robot obtains a predefined collision-free path from a previous 3D scan of the work environment. The robot then reconstructs the industrial scene around the robot at step 504, and performs 3D object detection and recognition analysis at step 506. After the object is identified, the robot will perform object grasping and positioning analysis at step 508. At query 510, control method 500 will determine whether the 3D model of the industrial scene is sufficient to detect and identify objects. If not, control method 500 will generate a new scan viewpoint to reduce the ambiguity of the object at step 512. At step 514, the robot will plan a new path and based on the scan viewpoint at step 512, the robot will move along the path and generate additional 3D scan images. Control method 500 will then return to step 504 and repeat steps 504, 506, and 508 until the scene model is sufficiently complete at query 510. If the 3D model of the scene is sufficient to detect and identify objects, then at step 516, the object grasp location analysis is run again. At step 518, a query is performed to determine whether the 3D model of the industrial scene is sufficient for the gripping analysis to direct the robot to grip the object. If the answer to query 518 is "no," then at step 520 control method 500 will generate a new scanning viewpoint and move the robot along a new path at step 522 to provide a new 3D scan at a different viewpoint. The control method 500 then loops back to step 504 and repeats steps 504, 506, 508, 510, 516, and 518 until the answer for each query at steps 510 and 518 is yes. Control method 500 analyzes the 3D scene model at step 524 and determines whether the 3D model of the scene surrounding the robot is complete at query 526. If not, the control method 500 generates a next scanning viewpoint at step 528, directs the robot to scan a new collision-free path at step 530, and then returns to step 504 to repeat until the answer to query 526 is yes. At that time, at step 532, the robot is instructed to grasp and pick up the detected object or perform other work tasks on the object.

In one aspect, the present disclosure includes a method comprising: determining a predefined collision-free robot path; moving the robot along a predefined robot path; scanning an industrial scene with a scanning sensor positioned on a robot while moving along a predefined robot path; storing the scanned image of the industrial scene in a memory; constructing a 3D model of the industrial scene based on the images stored in the memory; planning a next collision-free robot path based on the 3D model; moving the robot along the next collision-free robot path; scanning the industrial scene with a scanning sensor positioned on the robot while moving along the next robot path; and storing the new scanned image in a memory; and reconstructing a 3D model of the industrial scene based on the new scanned image.

In a refined aspect, the method further comprises repeating the planning, scanning, storing, and reconstructing steps until a complete 3D industrial scene is constructed; further comprising performing a work task with the robot after completing the 3D industrial scene model; wherein the scanning sensor is a 3D camera; further comprising moving the scanning sensor with respect to the robot when the predefined collision-free path is determined; wherein the movement of the scanning sensor comprises panning, tilting, rotating and translating movements with respect to the robot; wherein the movement of the scanning sensor comprises moving an arm of the robot while a base of the robot remains stationary; further comprising planning a collision-free path with the controller having a collision-free motion planning algorithm; where the planning of collision-free paths occurs in real time without off-line computer analysis.

Another aspect of the present disclosure includes a method comprising: determining a predefined collision-free robot path; scanning an industrial scene along a predefined collision-free robot path with a scanning sensor positioned on the robot; storing the scanned image of the industrial scene in a memory; constructing a 3D model of the industrial scene based on the images stored in the memory; detecting an object within a 3D model of an industrial scene; moving the robot along the collision-free robot path to generate a next scanning viewpoint of the detected object; scanning an industrial scene to obtain a new scanned image of an object; storing the new scan image in a memory; and repeating the moving and scanning steps until the object is identified to a threshold level of certainty.

In a refinement aspect of the disclosure, wherein the scanning comprises: capturing an image with a 3D camera; further comprising performing a robotic task on the object after the object has been identified to a threshold level of certainty; wherein the robotic task comprises grasping an object; further comprising pan, tilt and rotate scan sensors to capture images from different vantage points to generate a new 3D model of the industrial scene; further comprising planning a next scan path before generating a new 3D model of the industrial scene; wherein planning comprises analyzing the new 3D model with a controller having a collision-free motion planner algorithm; and further comprising determining a next scan path based on results from the collision free motion planning algorithm.

Another aspect of the disclosure includes a method comprising: determining a predefined collision-free robot path; scanning an industrial scene close to the robot by using a scanning sensor; storing the scanned image of the industrial scene in a memory; constructing a 3D model of the industrial scene based on the images stored in the memory; detecting an object within an industrial scene; determining whether the object is identified with sufficient accuracy; determining whether a robot task can be performed on an object; and performing one or more robotic tasks on the object after the object is identified with sufficient certainty.

In a refinement aspect of the disclosure, the method further comprises: if the object is not identified with sufficient certainty, generating a next scanning viewpoint and rescanning the industrial scene; further comprising: generating a next scanning viewpoint and rescanning the industrial scene if the 3D model of the industrial scene is insufficient for the gripping analysis; further comprising: if the 3D model of the industrial scene is incomplete, generating a next scanning viewpoint and rescanning the industrial scene; wherein the scanning of the industrial scene comprises panning, tilting, rotating, and translating the scanning sensor with respect to the robot; further comprising: planning a collision-free path by a controller having a collision-free motion planning algorithm; and wherein planning occurs in real-time without off-line computer analysis.

While the application has been illustrated and described in detail in the drawings and foregoing description, the same is to be considered as illustrative and not restrictive in character, it being understood that only the preferred embodiment has been shown and described and that all changes and modifications that come within the spirit of the application are desired to be protected. It should be understood that while the use of words such as preferable, preferabl preferred, preferred or more preferred in the description above indicate that the feature so described may be more desirable, embodiments in which it may not be necessary and in which the same feature is lacking are contemplated as within the scope of the application, the scope being defined by the claims that follow. In reading the claims, it is intended that when words such as "a," "an," "at least one," or "at least a portion" are used, unless specifically stated otherwise in the claims, it is not intended that the claims be limited to one item only. When the language "at least a portion" and/or "a portion" is used, the item can include a portion and/or the entire item unless specifically stated otherwise.

Unless specified or limited otherwise, the terms "mounted," "connected," "supported," and "coupled" and variations thereof are used broadly and encompass both direct and indirect mountings, connections, supports, and couplings. Further, "connected" and "coupled" are not restricted to physical or mechanical connections or couplings.