WO2018161305A1 - 抓取质量检测方法及其应用的方法与*** - Google Patents
抓取质量检测方法及其应用的方法与*** Download PDFInfo
- Publication number
- WO2018161305A1 WO2018161305A1 PCT/CN2017/076128 CN2017076128W WO2018161305A1 WO 2018161305 A1 WO2018161305 A1 WO 2018161305A1 CN 2017076128 W CN2017076128 W CN 2017076128W WO 2018161305 A1 WO2018161305 A1 WO 2018161305A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- quality
- grab
- point
- crawling
- grasping
- Prior art date
Links
Images
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J9/00—Programme-controlled manipulators
- B25J9/16—Programme controls
Definitions
- the invention relates to the field of robots, in particular to a method for detecting a quality of grasping, a method for grasping and grasping the method for grasping the quality of the grabbing, a method for grasping, a system for grasping, and a system for grasping.
- the basic implementation of the application is currently limited to a specific object form, so that the preset execution scheme is not truly intelligent and enhances versatility. How to automatically identify objects, and then plan an executable crawling scheme to achieve intelligence and versatility is one of the main research directions of intelligent robots.
- Ferrari and Canny proposed a method based on their ability to evaluate arbitrary external disturbances more than 20 years ago. This method is only for the positive component of the force at the point of contact. This metric became popular because it was intuitive and relatively simple. However, after the method is improved based on practical operations, it is not well recognized and applied due to the large amount of calculation. And the method does not consider local negative curvature.
- the prior art regarding the quality of the grab is based on the grabbed contact points and the weight of the object.
- the surface of the local object that will also be stressed when the grip is performed according to the contact point.
- the object of the present invention is to solve the problem that the local characteristics of the surface of the object are not considered in the prior art to solve the problem of the grab quality.
- a method for detecting a grab quality provided by a specific embodiment of the present invention is implemented in one or more computer systems, including performing steps:
- the contact point is an executable point that meets the quality of the grab.
- a method for crawling planning provided by an embodiment of the present invention is implemented in one or more computer systems, including performing steps:
- the contact point including a pit
- the contact point is an executable point that meets the quality of the grab
- An embodiment of the present invention provides a method for grasping, the method being implemented in one or more computer systems, including performing steps:
- the contact point including a pit
- the contact point is an executable point that meets the quality of the grab
- At least one processor At least one processor
- the data store includes instructions executed by at least one processor to enable the system to have executable functions including:
- Concave detection unit for:
- the contact point including a pit
- the contact point is an executable point that meets the quality of the grab
- a sensing device for acquiring a point cloud of the object to be grasped
- At least one processor At least one processor
- the data store includes data that is accessible by the processor, including:
- a database having an object and a grasping scheme corresponding to the object, the grasping scheme being formed according to a set of contact points obtained by including a pit of the surface of the object;
- Execution unit for:
- the end effector is controlled to perform a grab on the object according to the grasping scheme.
- the grasping quality detecting method provided by the invention, the grasping planning method, the grasping method, the grasping planning system and the grasping system applying the grasping quality detecting method, and the partial features of the surface of the concave object having negative curvature are combined Take quality, thus providing a crawling solution that effectively improves the quality of crawling. Furthermore, the grasping scheme adopts a local feature of negative curvature to facilitate the external interference resistance of the grab and improve the robustness. At the same time, it also effectively improves the calculation efficiency.
- FIG. 1 is a schematic structural diagram of a grab planning system and a grabbing system according to an embodiment of the present invention.
- FIG. 2 is a schematic flow chart of a method for capturing quality detection according to an embodiment of the present invention.
- FIG. 3 is a schematic flowchart of a method for capturing a plan according to an embodiment of the present invention.
- FIG. 4 is a schematic flowchart diagram of a method for grasping and planning according to another embodiment of the present invention.
- FIG. 5 is a schematic flowchart of a method for grasping according to an embodiment of the present invention.
- FIG. 6 is a schematic diagram of contact point analysis provided by an embodiment of the present invention.
- Figure 7 is a schematic illustration of an experiment performed in accordance with an embodiment of the present invention.
- An embodiment of the present invention provides a crawling planning system 10, including an end effector 130, at least one processor 150, and the data storage includes instructions executed by the at least one processor 150 to enable the system to have executable functions including: a concave detecting unit 102.
- the concave surface detecting unit 102 is configured to acquire a contact point that the surface of the object can be grasped, and the contact point includes at least one concave point.
- the grab quality detecting unit 104 is configured to obtain a convex set according to the friction cone of the contact point, and determine whether the grab quality condition is met according to the convex set, and if so, the contact point is an executable point that meets the grab quality.
- the capture plan 106 is configured to acquire a crawling scheme of the end effector that performs the crawl according to the executable point.
- the end effector 130 will be described with a three-finger manipulator as a specific example, but is not intended to limit the solution of the present invention. It can be understood that the end effector can be a jig, which has more degrees of freedom than the jig. Two-finger robots, as well as other forms of multi-finger robots.
- a grasping planning system provided by the present invention provides a high-quality grasping planning system for a concave object, which performs a grasping by utilizing a local feature of a surface of a concave object having a negative curvature.
- the gripping method is performed using a multi-finger robot, the portion having a negative curvature Features are good for grasping against external interference and improving robustness.
- a capture planning system 10 provided by the present embodiment includes a robot 130, a sensing device 140, at least one processor 150, and data storage including instructions executed by at least one processor 150, such that The system has executable functions including: a concave surface detecting unit 102, a grab quality detecting unit 104, a grab planning unit 106, a collision-free detecting unit 108, and an object matching unit 110.
- the data store also includes data that is accessible by the processor 150, including a database 112 having objects and a capture scheme corresponding to the objects.
- the gripping scheme is a gripping planning method according to the present invention, which is formed according to a set of contact points obtained by including pits on the surface of the object.
- the robot 130 adopts a scheme in which three fingers have 9 degrees of freedom.
- the sensing device 140 is configured to acquire identifiable information of the object. That is, according to the sensing device 140 acquiring the data of the object to be grasped, the system can identify the object and perform subsequent execution steps of the corresponding object.
- the point cloud is a set of point data of the surface of the object obtained by the sensing device 140.
- RealSense TM's verification carried out.
- the database 112 is configured to store an object and a grasping scheme corresponding to the object, including object information, and a relative position of the reference point of the end effector 130 and the object.
- the object information may be data corresponding to the identifiable information, including direct correspondence or indirect correspondence.
- the indirect correspondence includes an association relationship converted according to a preset rule.
- the end effector 130 will be exemplified below by the robot 130.
- the reference point of the robot 130 is a defined reference information that can represent the position of the robot 130. For example, it may be a joint of the robot 130 connected to the robot arm, or for a complicated robot, including an angle value of each joint of the robot, and the like. It can be flexibly set according to the specific application scenario and the end effector used.
- the object information includes the object identification, the object name, the object number, and the like, and can match the information of the specific object.
- the units included in the system 10 are communicatively coupled to each other, including direct or indirect communication.
- the execution quality detection method and the acquisition planning method provided by the embodiment of FIG. 2 to FIG. 4 will be exemplified below, and the execution principle of the system 10 will be exemplified.
- the system 10 acquires a point cloud of the object through the sensing device 140 (S402). And reconstructing the surface of the object according to the point cloud of the object.
- the method of triangular mesh reconstruction is specifically illustrated.
- the system 10 constructs a triangular mesh of the surface of the object according to the point cloud (S404). It can be understood that this step can be implemented by the concave detecting unit 102, or can be implemented independently of the concave detecting unit 102 and the object matching unit 110 dedicated to reconstruction.
- the concave surface detecting unit 102 calculates the angle between the co-edges of the adjacent triangular meshes based on the triangular mesh, and obtains the concave line based on the included angle judgment (S406).
- the collision-free detecting unit 108 acquires the collision-free contact point according to the concave line based on at least one groove detected by the concave detecting unit 102. Including, when there is only one groove, the concave is performed Line detection. When there are a plurality of concave lines, the respective concave lines are detected. For convenience of explanation, a point sampled from a concave line is referred to as a pit. Collision-free detection is performed according to the pits included in the groove (S408).
- the specific detection method includes sampling the concave point from the concave line and simulating one of the fingers contacting the concave point. Specifically, the end of the finger can be used to contact the concave point.
- the contact point of the other finger when the finger touches the pit is obtained.
- the end of the finger is simulated to contact the pit, and the other fingers are closed to obtain the contact points of the other fingers on the surface of the object. Since this example is exemplified by a three-finger robot 130, in this example, the contact point includes the pit and a contact point based on the other two fingers below the pit.
- the pit, and other contact points of the robot 130 on the surface of the object obtained based on the pit are taken as a set of contact points (S412).
- the end effector is a robot 130, it has a higher degree of freedom, so that the angle of the joint on the finger of the robot 130 can be changed to change the position of the palm when the end of the finger contacting the pit is unchanged. If other fingers are closed for each palm position, there may be multiple sets of contact points based on one pit, and thus the grab quality detecting unit 104 may select several sets of contact points above a certain threshold, or only one set of contact points may be reserved. .
- the collision detecting unit 108 may further include a step of determining whether there is an undetected pit (S426), when judging When there is an undetected pit, the next pit is acquired, and the above-described detecting step (S408, S410) is repeatedly performed until the pits of all the pits are completed without collision detection.
- the pits may be derived from a defined sampling rule.
- the robot can be judged according to the acquired set of contact points and the operating space parameters. 130 Whether a collision occurs when each finger reaches its respective contact point.
- the operating space parameters include environmental parameters and obstacle parameters, which can be flexibly set according to specific application scenarios and implementation space. If not, the set of contact points is used as a collision-free contact point, and the grab quality detecting unit 104 is provided to determine whether the grab quality is met.
- the middle finger of the robot is used as a finger that contacts the pit.
- the grab quality detecting unit 104 calculates a convex set based on the friction cone of the set of contact points based on a set of contact points obtained by the collision-free detecting unit 108 (S414).
- a convex set based on the friction cone of the set of contact points based on a set of contact points obtained by the collision-free detecting unit 108 (S414).
- Whether the content of the grab quality is met according to the convex set includes: performing a force closed grab quality analysis according to the convex set (S416). That is, the contact force applied by the finger to the object and the external load received by the object have a vector sum of zero. It is judged whether or not the analysis result is higher than the first threshold (S418).
- the set of contact points is an executable point that conforms to the grab quality (S420).
- the setting of the first threshold may be determined according to the requirements of a specific application scenario, such as an empirical value obtained during an experiment or an operation, or a theoretical value.
- the capture planning unit 106 acquires a crawling scheme of the end effector that performs the crawling according to the executable point according to the executable point of the grab quality detected by the grab quality detecting unit 104 (S422). Specifically, when the robot 130 contacts the corresponding executable point, the relative position based on the reference point is saved, and the relative position is saved to the database 112 corresponding to the grasping scheme of the object.
- the obtained executable points that meet the quality of the crawl are in multiple groups, and thus, multiple crawling schemes are obtained.
- the sorting selection may be performed according to the set sorting rule. For example, according to the angle with the Z axis of the world coordinate system, sorted from small to large. According to the sorting, several prioritized schemes are selected, for example, the top ten crawling schemes are sorted. Furthermore, according to the prior scheme, the quality of the crawling of each scheme is compared. In the end, the acquisition plan with the best priority for the best quality is obtained.
- the database 112 stores an object and a capture scheme corresponding to the object, wherein each object may correspondingly include multiple capture schemes. The database 112 can also store the crawl quality corresponding to each crawling scheme. It can be understood that the specific sorting method may be flexibly changed according to a specific application scenario, and is not limited to the above specific implementation examples.
- a high-quality grasping planning system for a concave object which performs a grasping by utilizing a local feature of a surface of a concave object having a negative curvature. It is good for grabbing against external disturbances and improves robustness.
- a gripping quality defined by a friction cone is used to further perform an effective gripping plan.
- the grasping method provided by the embodiment of the present invention is effectively verified, and the grabbing quality is higher than that of the prior art that does not consider the negative curvature feature. And the calculation efficiency is better than the prior art grasping method.
- the method overcomes the consideration that the conventional method does not incorporate the local negative curvature feature into the selected contact point.
- FIG. 6 is a schematic diagram of contact point analysis provided by an embodiment of the present invention.
- the illustration is schematically illustrated in a planar form. It should be noted that the solution of the present invention is applicable to applications of a three-dimensional scene.
- the surface of the object can be reconstructed by a triangular mesh. It is assumed that the boundary of the captured object can be decomposed into a finite set of faces, and the contact points are placed at the differentiable points.
- v be the unit vector of the tangent plane T p
- g v is a one-dimensional function defined according to the direction of v
- h is a given constant
- the gradient function is defined accordingly:
- v f is a unit vector on the plane T p and has the same direction as f 2 and f 3 , then:
- the F(p) defined above demonstrates the case illustrated in Fig. 6, that is, in the direction of negative curvature, the friction cone F(p) is enlarged and has anisotropic expansion characteristics, that is, a friction cone having a direction of greater negative curvature The angle is greater.
- the resistance to external forces e.g., external lateral forces
- the embodiment of the present invention adopts the local negative curvature of the object as a consideration element of the grab quality, and the method of performing the grab planning based on the partial negative curvature rate, which has the beneficial effect of improving the grab quality.
- the implementation adopts negative curvature as the grabbing planning method of the grabbing elements (see the inventive method column in the list), and the aforementioned random crawling method (see the existing method column in the list) performs ten grabs on each of the three objects.
- the results of the success rate of the capture are as follows:
- the average time of execution is as follows:
- the embodiment of the present invention adopts the local negative curvature of the object as the consideration element of the grab quality, and the method of grasping the plan based on the partial negative curvature, which not only has better grab quality, but also has the beneficial effect of improving the computing efficiency.
- the crawling scheme is stored to the database 112 based on the foregoing system 10 executing the crawling planning method.
- System 10 also includes an execution unit 114.
- the executing unit 114 determines whether the object has been stored in the database 112 according to the point cloud of the object, and if so, directly passes the object corresponding to the object.
- Information Obtain a crawling scheme corresponding to the database 112.
- the objects to be captured may be acquired by the above-mentioned crawling planning method, and the corresponding crawling scheme is stored in the database 112.
- the present invention also provides a grab system 20, referring to FIG.
- the data includes: a database 112 having an object and a grasping scheme corresponding to the object, the grasping scheme being formed according to a set of contact points obtained by including the pits of the surface of the object, specifically derived from the concave surface of the grasping planning system 10
- the detecting unit 102, the grabbing quality detecting unit 104, the grabbing planning unit 106, the collision-free detecting unit 108, and the object matching unit 110 may also be executed and constructed.
- the at least one processor 150 can read the data accessible by the database 112, and the data implementation execution unit 114 that can be fetched and executed.
- the grab system 20 acquires a point cloud of the object through the sensing device 140 (S502).
- the execution unit 114 may directly index the capture scheme corresponding to the object in the database 130 according to the point cloud of the object (S504).
- the executing unit 114 is further configured to obtain a position of a reference point corresponding to the relative position in the world coordinate system according to the pose of the object and the relative position included in the grabming scheme.
- the pose includes the positional parameters x, y, z, and the attitude parameters Pitch, Yaw, and Roll.
- the position information of the object in the world coordinate system is obtained, and according to the relative position of the grasping scheme, the position of the reference point corresponding to the relative position of the object in the world coordinate system is calculated.
- the position of the reference point in the world coordinate system is obtained according to the calculation, that is, the end position of the arm is obtained, and then the arm is obtained at the end according to the inverse kinematics algorithm.
- the position of the position that is, the target pose.
- the motion plan of the current pose to reach the target pose is performed.
- the drive module of the robot arm is provided to complete the execution. Specifically, in the world coordinate system, the relative position of the robot 130 and the object with the upper direction of the object as the grasping direction can be calculated to obtain a grasping scheme.
- the capture planning system further includes: when the capture planning system 20 acquires a point cloud of the object according to the sensing device 140, if there is no corresponding object data in the database 112, the data does not exist.
- the object information and the grasping scheme corresponding to the object may include a concave surface detecting unit 102 of the grab planning system 10, a grab quality detecting unit 104, a grab planning unit 106, a collision-free detecting unit 108, and an object matching unit. 110.
- the corresponding function is retrieved and executed by the processor 150, and a crawling scheme for the object that does not have a record is obtained and stored in the database 112.
- the grasping planning system provided by the embodiments of the present invention has various achievable manners, and the foregoing is merely illustrative of the principles, and is not intended to limit the present invention, and is obtained by those skilled in the art based on the principle. The variants are still within the scope of the invention.
- the present invention also provides a capture quality detection method 200, the method 200 being implemented on one or more computer systems, including the steps of:
- S210 obtaining a convex set according to a friction cone of a contact point that can be grasped on the surface of the object, the contact point including the concave point;
- S212 Determine, according to the convex set, whether the quality of the grab quality is met
- the implementation method provides a concave feature that has a negative curvature on the surface of the object as a consideration factor for the detection of the grab quality, and is effectively combined.
- the method of grasping quality detection is beneficial for effective execution of the grabbing plan.
- determining, according to the convex set, whether the content meets the grab quality condition according to the convex set includes:
- S418 Determine whether the first threshold is higher according to the result of the force closed grab quality analysis.
- the present invention also provides a crawling planning method 300 for applying the above-described crawling quality detecting method, the method 300 being implemented on one or more computer systems, including performing steps:
- S310 acquiring a contact point on the surface of the object that can be grasped, and the contact point includes a pit;
- S314 Determine, according to the convex set, whether the quality condition is met
- S318 Acquire a crawling scheme of the end effector that performs the crawl according to the executable point.
- a high-quality grasping planning method for a concave object is proposed, and the grasping is performed by using a local feature of the surface of the concave object having a negative curvature.
- the grasping method is performed by a multi-finger manipulator, the local feature with negative curvature is advantageous for grasping against external interference and improving the robustness.
- the method overcomes the consideration that the conventional method does not incorporate the local negative curvature feature into the selected contact point.
- step S316 is performed to acquire the contact points that the surface of the object can be grabbed:
- S408 Perform collision-free detection according to the pits included in the concave line
- collision-free detection includes:
- the concave line for acquiring the surface of the object includes:
- S406 Obtain a groove according to an angle of an adjacent triangular mesh of the co-edge.
- judging whether the content of the grab quality is consistent according to the convex set includes:
- S418 Determine whether the first threshold is higher according to the result of the force closed grab quality analysis.
- the present invention also provides a capture method 500, the method 500 being implemented on one or more computer systems, including the steps of:
- S502 Acquire identifiable information of the object. Specifically, the point cloud of the object can be acquired by the sensor to identify the object.
- S504 Obtain a crawling scheme corresponding to the object in the database according to the identifiable information.
- the grabbing scheme includes the relative position of the end point of the end effector to the object. It can also include the grab quality of the corresponding crawling scheme.
- the method 300 of the foregoing embodiment may be performed:
- S310 acquiring a contact point on the surface of the object that can be grasped, the contact point including a pit;
- S314 Determine, according to the convex set, whether the quality of the grab quality is met
- S318 Acquire a crawling scheme of the end effector that performs the crawling according to the executable point.
- the method 400 provided by the above embodiments may also be implemented to obtain a grasping scheme.
- the functional unit modules included in the above embodiments include program codes stored in a storage medium, and are read and executed by a processor to implement their functionalities. It can be understood that each unit is not limited to one continuous block or a single tangible physical unit, and the physical storage may be stored in multiple blocks or in the same storage medium. Continuous block.
- the functional unit module can also be implemented in hardware, including individual components, or through multiple component combinations, or in combination with other components.
- the computer readable medium further includes a non-transitory computer readable medium, such as a computer readable medium that stores data for a short period of time, such as memory, a cache of a processor, and random access memory (RAM).
- the computer readable medium may also include a non-transitory computer readable medium storing program code and/or data for a long period of time, such as secondary or persistent long term storage, such as read only memory (ROM), optical or magnetic disks, optical disks only Read memory (CD-ROM), for example.
- ROM read only memory
- CD-ROM optical disks only Read memory
- the computer readable medium can also be other volatile or nonvolatile storage systems.
- the computer readable medium can be considered a computer readable storage medium or a tangible storage device.
Landscapes
- Engineering & Computer Science (AREA)
- Robotics (AREA)
- Mechanical Engineering (AREA)
- Manipulator (AREA)
Abstract
一种抓取质量检测方法，包括执行步骤：根据物体表面可执行抓取的接触点的摩擦锥获得凸集，所述接触点包括凹点（S210）；根据所述凸集判断是否符合抓取质量条件（S212）；若是，则所述接触点为符合抓取质量的可执行点（S214）。所述方法将局部物体表面负曲率特征和抓取质量相结合，采用摩擦锥定义抓取质量，以进一步执行有效的抓取规划。还提供了应用该抓取质量检测方法的抓取规划方法，抓取方法、抓取规划***及抓取***。
Description
本发明涉及机器人领域，具体涉及一种抓取质量检测方法，应用该抓取质量检测方法的抓取规划方法，抓取方法、抓取规划***及抓取***。
关于机器人的抓取方案，目前而言可实施应用的基本是限定于特定物体形态，从而给定预置的执行方案，并没有真正的具备智能化以及提升通用性。如何自动识别物体，进而规划可执行抓取方案，实现智能性、通用性，是目前智能机器人的主要研究方向之一。
对此，已有相关的学术研究方案，例如，最近一项由Roa和Su′arez完成的调查，提供的一个全面的总结包括二十四个指标，并揭示了负曲率通常不考虑。
在实践方面，由法拉利和Canny二十多年前提出的基于其抗任意外部干扰能力评估抓取的方法。该方法仅针对于在接触点的力的正分量。这个度量变得流行，因为它是直观的且相对简单。但是，该方法基于实践操作而进行改进后，由于计算量大并没有被很好的认可与应用。且该方法也并未考虑局部负曲率。
总体而言，现有技术关于抓取质量是根据抓取的接触点及物体重心进行评估。但是没有考量依据接触点执行抓取时，同样会受力的局部物体表面。对于局部表面形成的特征性，如何能进一步提高通用性同时具有高质量的抓取仍未有相关的研究方案。
发明内容
本发明的目的是解决现有技术中未考量物体表面局部特征的特性，来解决抓取质量的问题。
本发明具体实施方式提供的一种抓取质量检测方法，所述方法实施于一个或多个计算机***，包括执行步骤：
根据物体表面可执行抓取的接触点的摩擦锥获得凸集，所述接触点包括凹点；
根据所述凸集判断是否符合抓取质量条件；
若是，则所述接触点为符合抓取质量的可执行点。
本发明实施方式提供的一种抓取规划方法，所述方法实施于一个或多个计算机***，包括执行步骤：
获取物体表面可执行抓取的接触点，所述接触点包括凹点；
根据所述接触点的摩擦锥获得凸集；
根据所述凸集判断是否符合抓取质量条件；
若是，则所述接触点为符合抓取质量的可执行点；
根据所述可执行点获取执行抓取的末端执行器的抓取方案。
本发明实施方式提供的一种抓取方法，所述方法实施于一个或多个计算机***，包括执行步骤：
获取物体的可识别信息；
根据所述可识别信息获取数据库中对应该物体的抓取方案；
若数据库中不存在对应的物体，则：
获取物体表面可执行抓取的接触点，所述接触点包括凹点；
根据所述接触点的摩擦锥获得凸集；
根据所述凸集判断是否符合抓取质量条件；
若是，则所述接触点为符合抓取质量的可执行点；
根据所述可执行点获取执行抓取的末端执行器的抓取方案。
本发明实施方式提供的一种抓取规划***，包括：
末端执行器；
至少一个处理器；
数据存储包括由至少一个处理器执行的指令，以使***具有可执行功能包括：
凹面检测单元，用于：
获取物体表面可执行抓取的接触点，所述接触点包括凹点；
抓取质量检测单元，用于：
根据所述接触点的摩擦锥获得凸集；
根据所述凸集判断是否符合抓取质量条件；
若是，则所述接触点为符合抓取质量的可执行点；
抓取规划单元，用于：
根据所述可执行点获取执行抓取的所述末端执行器的抓取方案。
本发明实施方式提供的一种抓取***，其特征在于，包括：
末端执行器；
感测装置，用于获取被抓物体的点云；
至少一个处理器；
数据存储包括可被所述处理器存取的数据，包括：
具有物体及与物体对应的抓取方案的数据库，所述抓取方案根据包括所述物体表面的凹点得到的一组接触点形成；
执行单元，用于：
根据所述感测装置获取的所述物体的可识别信息，于所述数据库中索引对应该物体的抓取方案；
根据所述抓取方案控制所述末端执行器对所述物体执行抓取。
本发明提供的抓取质量检测方法、应用该抓取质量检测方法的抓取规划方法、抓取方法、抓取规划***及抓取***，利用凹面物体的表面具有负曲率的局部特征，结合抓取质量，从而提供有效提升抓取质量的抓取方案。进而，该抓取方案采用负曲率的局部特征有利于抓取的抗外部干扰力，提升鲁棒性。同时，还有效提高计算效率。
为了更清楚地说明本发明的技术方案，下面将对实施方式中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是发明的一些实施方式，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以从这些附图获得其他的附图。
图l是本发明实施例提供的一种抓取规划***及抓取***的结构示意图。
图2是本发明实施例提供的一种抓取质量检测的方法流程示意图。
图3是本发明实施例提供的一种抓取规划方法的流程示意图。
图4是本发明另一实施例提供的一种抓取规划方法的流程示意图。
图5是本发明实施例提供的一种抓取方法的流程示意图。
图6是本发明实施例提供的接触点分析示意图。
图7是根据本发明实施例实施的实验示意图。
下面将结合本发明实施方式中的附图，对本发明实施方式中的技术方案进行清楚、完整地描述。
本发明实施例提供一种抓取规划***10，包括末端执行器130，至少一个处理器150，数据存储包括由至少一个处理器150执行的指令，以使***具有可执行功能包括：凹面检测单元102，抓取质量检测单元104，抓取规划单元106。其中，凹面检测单元102，用于获取物体表面可执行抓取的接触点，该接触点包括至少一个凹点。抓取质量检测单元104，用于根据接触点的摩擦锥获得凸集，以及根据凸集判断是否符合抓取质量条件，若是，则接触点为符合抓取质量的可执行点。抓取规划单106，用于根据可执行点获取执行抓取的末端执行器的抓取方案。
以下实施方式中，末端执行器130将以三指机械手作为具体示例加以说明，但并不用于限定本发明方案，可以理解的是，末端执行器可以为夹具，相较夹具具有更多自由度的双手指机械手，以及其他形态多指机械手等。
本发明实施利提供的一种抓取规划***，提出了一种针对具有凹面物体的高质量抓取规划***，利用凹面物体的表面具有负曲率的局部特征，执行抓取。当采用多指机械手执行该抓取方法，具有负曲率的局部
特征有利于抓取抵抗外部干扰力，提升鲁棒性。
为充分阐述本发明方案的原理，以下提供的具体实施方式将结合三指机械手进行示例性说明。
如图1所示，为本实施例提供的一种抓取规划***10，包括机械手130，感测装置140，至少一个处理器150，以及数据存储包括由至少一个处理器150执行的指令，使得***具有可执行功能包括：凹面检测单元102，抓取质量检测单元104，抓取规划单元106，无碰撞检测单元108，物体匹配单元110。数据存储还包括可被处理器150存取的数据，包括：具有物体及与物体对应的抓取方案的数据库112。其中抓取方案为根据本发明提供的抓取规划方法，根据包括物体表面的凹点得到的一组接触点形成。具体的，该示例中，机械手130采用了三指具有9个自由度的方案。感测装置140，用于获取物体的可识别信息。即根据感测装置140获取欲抓取物体的数据，***可以识别该物体并执行对应该物体的后续执行步骤。为了更加清楚的阐述，以下将对可识别信息以“点云”(Point Cloud)加以具体示例说明。点云为通过感测装置140得到的物体外观表面的点数据集合。本发明的实验过程中，采用了
公司的RealSenseTM进行实施验证。数据库112，用于存储物体及与物体对应的抓取方案，包括物体信息，以及末端执行器130的基准点与该物体的相对位置。还可以包括抓取方案对应的抓取质量参数。其中，物体信息可以为与可识别信息对应的数据，包括直接对应或间接对应。其中，间接对应包括根据预设的规则转换得到的关联关系。以下将以机械手130来示例末端执行器130。其中，机械手130的基准点为定义的可代表机械手130位置的一致参考
信息。例如，可以为机械手130与机械臂连接的关节，或者对于复杂的机械手而言，包括机械手各个关节的角度值，等等。具体可以根据具体应用场景以及采用的末端执行器而灵活设定。物体信息包括物体标识、物体名称、物体编号等，可以匹配到具体物体的信息。
该***10包括的各单元彼此通信连接，包括直接或间接的通信方式。以下将结合图2至图4实施例提供的执行质量检测方法、抓取规划方法，示例性阐明该***10的执行原理。
***10通过感测装置140获取物体的点云(S402)。以及根据物体点云对物体表面进行重构。下述示例中，以三角网格重构的方法加以具体示例性阐述。
***10根据点云，构建物体表面的三角网格(S404)。可以理解的是，该步骤可以由凹面检测单元102实现，也可以为独立于凹面检测单元102、专用于重构的物体匹配单元110实现。凹面检测单元102根据三角网格，计算相邻三角网格共边的夹角，根据夹角判断获得凹线(S406)。一实施示例，假设两个三角网格相交于边E1，则存在不位于边E1所在的相交线上的两个定点V1及V2，分别与E1构成两个三角形，通过判断点(V1+V2)/2及E1上的任意点构成的向量，以及两个三角形法向量之和是否是同方向，即点积大于0，则得到为凹面夹角。如此，计算得到所有凹面夹角，根据所有具有凹面夹角的相邻三角网格的共边，从而获得所有凹线。
无碰撞检测单元108根据凹面检测单元102检测出的至少一个凹线，根据凹线获取无碰撞的接触点。包括，当仅有一个凹线，则执行对该凹
线的检测。当具有多个凹线，则分别对各凹线进行检测。为了便于说明，将从凹线上取样的点称为凹点。根据凹线包括的凹点进行无碰撞检测(S408)。具体检测方法包括，从凹线上取样获取凹点，模拟其中一只手指接触该凹点，具体的，可采用该手指的末端接触该凹点。例如，根据机械手130的模型，模拟该手指末端由当前状态运动到该凹点的过程中，是否会存在碰撞(S410)。若无碰撞，则获取该手指接触到该凹点时其他手指的接触点。如，模拟该手指末端接触到凹点，闭合其他手指，得到其他手指分别于物体表面的接触点。由于该示例采用三指机械手130例举，因此，该示例下，接触点包括该凹点以及基于该凹点下另外两只手指的接触点。为便于更加清楚的表述，将该凹点，以及基于该凹点获得的机械手130于物体表面的其他接触点作为一组接触点(S412)。一些实施方式下，由于末端执行器为机械手130时，具有较高自由度，因此在接触该凹点的手指末端不变的情况下，可以改变机械手130手指上关节的角度来改变手掌的位置，针对每一个手掌位置闭合其他的手指，则可能基于一个凹点存在多组接触点，进而可供抓取质量检测单元104选择高于一定阈值的几组接触点，或者，仅保留一组接触点。可以理解的是，无碰撞检测单元108将根据凹线包括的一凹点完成上述检测步骤(S408、S410)后，还可以包括判断是否有未检测的凹点的步骤(S426)，当判断还存在未检测的凹点时，获取下一个凹点，重复执行上述检测步骤(S408、S410)，直到所有凹线的凹点均完成无碰撞检测。可以理解的是，凹点可源于根据定义的取样规则得到。
进而，可根据获取的一组接触点，以及操作空间参数，判断机械手
130各手指到达各自的接触点过程中是否会发生碰撞。操作空间参数包括环境参数，障碍物参数，可以依据具体应用场景以及实施空间等灵活设置。若没有，则将该组接触点作为无碰撞的接触点，提供抓取质量检测单元104判断是否符合抓取质量的要求。
该示例中，采用机械手中间手指作为接触凹点的手指。
抓取质量检测单元104，根据无碰撞检测单元108获得的一组接触点，根据该一组接触点的摩擦锥计算得到凸集(S414)。具体的计算方法，可以参考论文《Fast grasp quality evaluation with partial convex hull computation》(基于部分凸包计算的快速抓取质量评估)中提供的一种计算方式，以及其他计算方式。根据凸集判断是否符合抓取质量条件，具体包括：根据凸集进行力闭合抓取质量分析(S416)。即手指对物体施加的接触力和物体所受到的外载荷，其矢量和为零。判断分析结果是否高于第一阈值(S418)。当分析结果高于第一阈值，则该一组接触点为符合抓取质量的可执行点(S420)。其中，第一阈值的设定可以为根据具体应用场景的需求而定，例如实验、操作过程中得到的经验值，也可以为理论值等。
抓取规划单元106，根据抓取质量检测单元104得到的符合抓取质量的可执行点，根据该可执行点获取执行抓取的末端执行器的抓取方案(S422)。具体的，获取机械手130接触对应可执行点时基于基准点的相对位置，将该相对位置保存至数据库112对应该物体的抓取方案。
一般而言，得到的符合抓取质量的可执行点为多组，因此，会得到多个抓取方案。
一种实施方式下，当根据抓取规划单元106获得多个抓取方案选取最终执行的抓取方案时，可根据设定的排序规则进行排序选取。例如，根据与世界坐标系Z轴的夹角，由小至大排序。根据该排序选出排序在前的几个方案，例如，排序前十的抓取方案。进而，根据该位列在前的方案，比对各方案的抓取质量。最终获取抓取质量最优的方案最为优先执行的抓取方案。一种实施方式下，数据库112中存储有物体以及与物体对应的抓取方案，其中，各物体可对应包括多个抓取方案。数据库112还可以存储有各抓取方案对应的抓取质量。可以理解的是，具体的排序方法可以依据具体应用场景而灵活变化，并不限定于上述具体实施示例。
本发明实施方式，提出了一种针对具有凹面物体的高质量抓取规划***，利用凹面物体的表面具有负曲率的局部特征，执行抓取。有利于抓取的抗外部干扰力，提升鲁棒性。并采用了一种利用摩擦锥定义的抓取质量，以进一步执行有效的抓取规划。
根据应用该方案进行实验得到的数据结果，有效验证了本发明实施方式提供的抓取方法，其抓取质量高于现有技术中未考虑负曲率特征的抓取方法。并且计算效率优于现有技术的抓取方法。
该方法克服了传统方法未将局部负曲率特征纳入选择接触点的考量要素。
为了更清楚的阐述本发明实施方案的原理，以下将例举一种利用负曲率定义抓取质量的方法。
如图6所示，为本发明实施例提供的接触点分析示意图，该图示以平面形式示意说明，需要说明的是，本发明方案可适用于三维场景的应
用。物体表面可以通过三角网格来重建。假设被抓取的物体的边界可以被分解为有限集合的面，接触点放置于可微点。即，假设于物体表面的接触点为p，p∈R3，那么将存在一个于物体表面临近p点的点x，且满足二次可微方程式g(x)＝0，其中g：R3→R，则为p点二次微分的适用函数。如图5，物体A的表面存在三个可接触的点a，b，c，其中，a点位于物体的局部凸表面，b点位于物体表面施加正交力的位置，c点位于物体的局部凹表面。
设v是在切平面Tp的单位向量，gv是根据v的方向定义的一维函数，h为给定的常数，据此定义梯度函数：
当给定力f，vf为在平面Tp上且具有与f2及f3相同方向的单位向量，则:
于是，定义扩大的摩擦锥FE(p)为：
当
且k为固定参数，k>0。上述定义的F(p)论证了图6示例的情况，即，沿负曲率的方向，摩擦锥F(p)扩大，且具有各向异性膨胀特性，即，具有更大负曲率方向上摩擦锥的角度越大。如图6的示例下，c点的抗外力(例如外部横向力)能力更优。因此，本发明实施方式采用物体局部负曲率作为抓取质量的考量要素，以及基于局部负曲
率进行抓取规划的方法，具有提高抓取质量的有益效果。
另一方面，与现有学术上作为基准判别标准的随机抓取方法行实验比对，具体采用了三个物体，如图7所示，包括标识模型、饮料瓶、玩具鸭，分别通过本发明实施方式采用负曲率作为抓取要素的抓取规划方法(见列表中的发明方法列)，以及前述随机抓取方法(见列表中的现有方法列)对三个物体各执行十次抓取，其中抓取成功率结果如下：
执行平均用时结果如下：
因此，本发明实施方式采用物体局部负曲率作为抓取质量的考量要素，以及基于局部负曲率进行抓取规划的方法，不仅具有更优的抓取质量，还具有提高运算效率的有益效果。
基于前述***10执行抓取规划方法构建了抓取方案存储至数据库112。***10还包括执行单元114。在实际操作中，当***10通过感测装置140获取到物体的点云，执行单元114根据物体的点云，判断该物体是否已储存于数据库112，若是，则可直接通过该物体对应的物体信息
获取数据库112对应的抓取方案。
可以理解的是，在一些应用场景下，需被抓取的物体可以均通过上述抓取规划方法完成获取可执行点，并得到对应的抓取方案存至数据库112。该实施方式下，本发明还提供一种抓取***20，参看图1，包括至少一个处理器150，机械手130，感测装置140，以及执行单元114，数据存储包括可被处理器150存取的数据，包括：具有物体及与物体对应的抓取方案的数据库112，抓取方案根据包括所述物体表面的凹点得到的一组接触点形成，具体可源于抓取规划***10的凹面检测单元102，抓取质量检测单元104，抓取规划单元106，无碰撞检测单元108，还可以包括物体匹配单元110，执行并构建。其中，至少一个处理器150可以读取存储数据包括数据库112的可存取的数据，以及可调取并执行的数据实现执行单元114。该示例下，抓取***20，通过感测装置140获取物体的点云(S502)。执行单元114可直接根据该物体的点云，于数据库130中索引对应该物体的抓取方案(S504)。
执行单元114，还用于根据物体的位姿，以及抓取方案包括的相对位置，获得在世界坐标系下对应该相对位置的基准点的位置。一实施方式下，当获取到抓取方案，以及根据感测装置140获取的物体的点云得到物体的位姿。其中，位姿包括位置参数x，y，z，以及姿态参数Pitch(俯仰角)，Yaw(偏航角)，Roll(横滚角)。根据物体的位姿得到物体在世界坐标系下的位置信息，进而根据抓取方案的相对位置，计算得到机械手130于世界坐标系下对应该物***姿与其相对位置对应的基准点位置。
假设当机械手130的基准点采用机械手130与机械臂连接的关节，则根据计算得到的世界坐标系下基准点位置，即得到机械臂的末端位置，进而根据逆运动学算法，得到机械臂于末端位置的位姿，即目标位姿。根据该目标位姿以及当前位姿，进行当前位姿到达目标位姿的运动规划。提供机械臂的驱动模块，完成执行。具体的，可以算出在世界坐标系里，以物体上方为抓取方向的机械手130和物体的相对位置进而获取抓取方案。
其他实施方式下，本发明实施例提供的抓取规划***，还包括，当抓取规划***20根据感测装置140获取物体的点云，若数据库112中不存在对应的物体数据，即不存在对应该物体的物体信息及抓取方案，则可包括抓取规划***10的凹面检测单元102，抓取质量检测单元104，抓取规划单元106，无碰撞检测单元108，还可以包括物体匹配单元110，通过处理器150调取并执行相应的功能，获得对该未存有记录的物体的抓取方案并存储至数据库112。
综上，本发明实施方式提供的抓取规划***具有多种可实现方式，以上仅为示例性原理说明，并不用于限制本发明，本领域技术人员基于该原理实现的未经创造性劳动而得到的变化方案仍属于本发明保护范围。
本发明还提供了一种抓取质量检测方法200，该方法200实施于一个或多个计算机***，包括执行步骤：
S210：根据物体表面可执行抓取的接触点的摩擦锥获得凸集，接触点包括凹点；
S212：根据凸集判断是否符合抓取质量条件；
S214：若是，则接触点为符合抓取质量的可执行点。
该实施方法，提供了一种将物体表面具有负曲率的凹面特征作为抓取质量检测的考量因素，有效结合。该抓取质量检测的方法有益于进行抓取规划的有效执行。
具体的，步骤S212根据凸集判断是否符合抓取质量条件包括：
S416：根据凸集进行力闭合抓取质量分析；
S418：根据力闭合抓取质量分析的结果判断是否高于第一阈值。
本发明还提供了一种应用上述抓取质量检测方法的抓取规划方法300，该方法300实施于一个或多个计算机***，包括执行步骤：
S310：获取物体表面可执行抓取的接触点，接触点包括凹点；
S312：根据接触点的摩擦锥获得凸集；
S314：根据凸集判断是否符合抓取质量条件；
S316：若是，则接触点为符合抓取质量的可执行点；
S318：根据可执行点获取执行抓取的末端执行器的抓取方案。
本发明实施方式，提出了一种针对具有凹面物体的高质量抓取规划方法，利用凹面物体的表面具有负曲率的局部特征，执行抓取。当采用多指机械手执行该抓取方法，具有负曲率的局部特征有利于抓取抵抗外部干扰力，提升鲁棒性。
该方法克服了传统方法未将局部负曲率特征纳入选择接触点的考量要素。
以下将结合图4提供的实施例加以具体说明，但可以理解的是，实
施例仅用于阐明本发明原理，并不用于限定本发明。
具体的，步骤S316获取物体表面可执行抓取的接触点包括：
S406：获取物体表面的凹线；
S408：根据凹线包括的凹点，进行无碰撞检测；
S412：若检测结果为无碰撞，则将该凹点、以及基于该凹点获得的末端执行器于物体表面的其他接触点作为一组接触点。
具体的，无碰撞检测包括：
根据一组接触点，以及物体所处的环境参数；
末端执行器到达该一组接触点是否存在碰撞；
若无，则检测结果为无碰撞。
具体的，获取物体表面的凹线包括：
S402：获取物体的点云；
S404：构建物体表面的三角网格；
S406：根据共边的相邻三角网格的夹角获得凹线。
具体的，根据凸集判断是否符合抓取质量条件包括：
S416：根据凸集进行力闭合抓取质量分析；
S418：根据力闭合抓取质量分析的结果判断是否高于第一阈值。
当被抓取物体及对应的抓取方案已存储于数据库112时，本发明还提供一种抓取方法500，该方法500实施于一个或多个计算机***，包括执行步骤：
S502：获取物体的可识别信息。具体的，可以通过传感器获取物体的点云，进而识别物体。
S504：根据可识别信息获取数据库中对应该物体的抓取方案。
抓取方案包括末端执行器的基准点与物体的相对位置。还可以包括对应抓取方案的抓取质量。
其他实施方式下，若数据库112中不存在对应的物体，如物体信息以及该物体对应的抓取方案，则可执行前述实施方式的方法300：
S310：获取物体表面可执行抓取的接触点，所述接触点包括凹点；
S312：根据所述接触点的摩擦锥获得凸集；
S314：根据所述凸集判断是否符合抓取质量条件；
S316：若是，则所述接触点为符合抓取质量的可执行点；
S318：根据所述可执行点获取执行抓取的末端执行器的抓取方案。
也可以实施上述实施方式提供的方法400以获得抓取方案。
由于上述方法步骤在前述抓取规划***的具体实施方式中，有应用实施示例，其运作原理类似，在此不再赘述。
以上是本发明的例举的实施方式，应当指出，对于本技术领域的普通技术人员来说，在不脱离本发明原理的前提下，还可以做出若干改进和润饰，这些改进和润饰也视为本发明的保护范围。需要说明的是，上述实施方式中包括的功能单元模块，包括存储于存储介质的程序代码，通过处理器读取并执行实现其功能性。可以理解的是，所述的各单元，并不限定于一个连续程序段或单个的有形物理单元，其物理存储的分布形式可以为多个程序段分离的存储，也可以为同一存储介质上非连续的程序段。该功能单元模块还可以通过硬件实现，包括单独的元器件实现，也可以通过多个元器件组合实现，或者与其他元器件联合实现。
计算机可读介质还包括非暂时性计算机可读介质，如计算机可读介质存储数据的时间很短，如内存，处理器的缓存，和随机存取存储器(RAM)。计算机可读介质也可能包括非暂时性的计算机可读介质长时间存储程序代码和/或数据，如继发性或持续性的长期储存，如只读存储器(ROM)，光盘或磁盘、光盘只读存储器(CD-ROM)，例如。计算机可读介质也可以是其它易失性或非易失性存储***。计算机可读介质可被认为是计算机可读存储介质，或有形的存储设备。
Claims (20)
- 一种抓取质量检测方法，其特征在于，所述方法实施于一个或多个计算机***，包括执行步骤：根据物体表面可执行抓取的接触点的摩擦锥获得凸集，所述接触点包括凹点；根据所述凸集判断是否符合抓取质量条件；若是，则所述接触点为符合抓取质量的可执行点。
- 如权利要求1所述的抓取质量检测方法，其特征在于，所述根据所述凸集判断是否符合抓取质量条件包括：根据所述凸集进行力闭合抓取质量分析；根据所述力闭合抓取质量分析的结果判断是否高于第一阈值。
- 一种抓取规划方法，其特征在于，所述方法实施于一个或多个计算机***，包括执行步骤：获取物体表面可执行抓取的接触点，所述接触点包括凹点；根据所述接触点的摩擦锥获得凸集；根据所述凸集判断是否符合抓取质量条件；若是，则所述接触点为符合抓取质量的可执行点；根据所述可执行点获取执行抓取的末端执行器的抓取方案。
- 如权利要求3所述的抓取规划方法，其特征在于，所述获取物体表面可执行抓取的接触点包括：获取物体表面的凹线；根据所述凹线包括的凹点，进行无碰撞检测；若检测结果为无碰撞，则将该凹点、以及基于该凹点获得的所述末端执行器于物体表面的其他接触点作为一组接触点。
- 如权利要求4所述的抓取规划方法，其特征在于，所述无碰撞检测包括：根据所述凹点，以及所述物体所处的环境参数；判断所述末端执行器到达该凹点是否存在碰撞；若无，则所述检测结果为无碰撞。
- 如权利要求4所述的抓取规划方法，其特征在于，所述获取物体表面的凹线包括：获取所述物体的点云；构建所述物体表面的三角网格；根据共边的相邻三角网格的夹角获得凹线。
- 如权利要求3所述的抓取规划方法，其特征在于，所述根据所述凸集判断是否符合抓取质量条件包括：根据所述凸集进行力闭合抓取质量分析；根据所述力闭合抓取质量分析的结果判断是否高于第一阈值。
- 如权利要求3所述的抓取规划方法，其特征在于，还包括将所述物体以及对应的所述抓取方案存储至数据库。
- 如权利要求3所述的抓取规划方法，其特征在于，所述抓取方案包括所述末端执行器的基准点与所述物体的相对位置。
- 一种抓取方法，其特征在于，所述方法实施于一个或多个计算机***，包括执行步骤：获取物体的可识别信息；根据所述可识别信息获取数据库中对应该物体的抓取方案；若数据库中不存在对应的物体，则：获取物体表面可执行抓取的接触点，所述接触点包括凹点；根据所述接触点的摩擦锥获得凸集；根据所述凸集判断是否符合抓取质量条件；若是，则所述接触点为符合抓取质量的可执行点；根据所述可执行点获取执行抓取的末端执行器的抓取方案。
- 如权利要求10所述的方法，其特征在于，所述抓取方案包括所述末 端执行器的基准点与所述物体的相对位置。
- 一种抓取规划***，其特征在于，包括：末端执行器；至少一个处理器；数据存储包括由至少一个处理器执行的指令，以使***具有可执行功能包括：凹面检测单元，用于：获取物体表面可执行抓取的接触点，所述接触点包括凹点；抓取质量检测单元，用于：根据所述接触点的摩擦锥获得凸集；根据所述凸集判断是否符合抓取质量条件；若是，则所述接触点为符合抓取质量的可执行点；抓取规划单元，用于：根据所述可执行点获取执行抓取的所述末端执行器的抓取方案。
- 如权利要求10所述的***，其特征在于，所述凹面检测单元还用于：获取物体表面的凹线以及凹线包括的凹点；所述***还包括无碰撞检测单元，用于：根据所述凹点进行无碰撞检测；若检测结果为无碰撞，则将该凹点、以及基于该凹点获得的所述末端执行器于物体表面的其他接触点作为一组接触点。
- 如权利要求11所述的***，其特征在于，所述无碰撞检测单元，还用于：根据所述凹点，以及所述物体所处的环境参数；判断所述末端执行器到达该凹点是否存在碰撞；若无，则所述检测结果为无碰撞。
- 如权利要求11所述的***，其特征在于，还包括物体匹配单元，用于：获取所述物体的点云；构建所述物体表面的三角网格；所述凹面检测单元，还用于：根据共边的相邻三角网格的夹角获得凹线。
- 如权利要求10所述的***，其特征在于，所述抓取质量检测单元，还用于：根据所述凸集进行力闭合抓取质量分析；根据所述力闭合抓取质量分析的结果判断是否高于第一阈值。
- 如权利要求10所述的***，其特征在于，所述抓取规划单元，还用于将所述物体以及对应的所述抓取方案存储至数据库。
- 如权利要求10所述的***，其特征在于，所述抓取方案包括所述末端执行器的基准点与所述物体的相对位置。
- 一种抓取***，其特征在于，包括：末端执行器；感测装置，用于获取被抓物体的点云；至少一个处理器；数据存储包括可被所述处理器存取的数据，包括：数据库，存储物体及与物体对应的抓取方案，所述抓取方案根据包括所述物体表面的凹点得到的一组接触点形成；执行单元，用于：根据所述感测装置获取的所述物体的可识别信息，于所述数据库中索引对应该物体的抓取方案；根据所述抓取方案控制所述末端执行器对所述物体执行抓取。
- 如权利要求19所述的***，其特征在于，所述执行单元还用于根据 所述物体的位姿，以及所述抓取方案包括的所述相对位置，获得在世界坐标系下对应所述相对位置的所述基准点的位置。
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201780022587.8A CN109153118A (zh) | 2017-03-09 | 2017-03-09 | 抓取质量检测方法及其应用的方法与*** |
| PCT/CN2017/076128 WO2018161305A1 (zh) | 2017-03-09 | 2017-03-09 | 抓取质量检测方法及其应用的方法与*** |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| PCT/CN2017/076128 WO2018161305A1 (zh) | 2017-03-09 | 2017-03-09 | 抓取质量检测方法及其应用的方法与*** |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2018161305A1 true WO2018161305A1 (zh) | 2018-09-13 |
Family
ID=63447351
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2017/076128 WO2018161305A1 (zh) | 2017-03-09 | 2017-03-09 | 抓取质量检测方法及其应用的方法与*** |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN109153118A (zh) |
| WO (1) | WO2018161305A1 (zh) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109986560A (zh) * | 2019-03-19 | 2019-07-09 | 埃夫特智能装备股份有限公司 | 一种面向多目标种类的机械臂自适应抓取方法 |
| CN113538459A (zh) * | 2021-07-07 | 2021-10-22 | 重庆大学 | 一种基于落点区域检测的多模式抓取避障检测优化方法 |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1623741A (zh) * | 2004-05-24 | 2005-06-08 | 熊勇刚 | 一种多机械臂机器人关节间的碰撞检测方法 |
| CN102126221A (zh) * | 2010-12-23 | 2011-07-20 | 中国科学院自动化研究所 | 一种基于图像信息的机械手抓取物体的方法 |
| CN105184019A (zh) * | 2015-10-12 | 2015-12-23 | 中国科学院自动化研究所 | 机器人抓取方法及*** |
| CN105598965A (zh) * | 2015-11-26 | 2016-05-25 | 哈尔滨工业大学 | 基于立体视觉的机器人欠驱动手自主抓取方法 |
| CN105856232A (zh) * | 2016-05-30 | 2016-08-17 | 先驱智能机械（深圳）有限公司 | 物体的抓取方法及抓取*** |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102565081B (zh) * | 2012-01-12 | 2014-01-22 | 北京化工大学 | 基于圆结构光视觉检测三维数据点的管道缺陷检测方法 |
| CN104809276B (zh) * | 2015-04-14 | 2017-09-12 | 长安大学 | 一种多手指机器人动力学解析模型及其建模方法 |
| CN104950692B (zh) * | 2015-05-20 | 2017-08-11 | 上海交通大学 | 一种适用于欠驱动机械手的抓取仿真*** |
| US9889564B2 (en) * | 2015-07-08 | 2018-02-13 | Empire Technology Development Llc | Stable grasp point selection for robotic grippers with machine vision and ultrasound beam forming |
-
2017
- 2017-03-09 CN CN201780022587.8A patent/CN109153118A/zh active Pending
- 2017-03-09 WO PCT/CN2017/076128 patent/WO2018161305A1/zh active Application Filing
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1623741A (zh) * | 2004-05-24 | 2005-06-08 | 熊勇刚 | 一种多机械臂机器人关节间的碰撞检测方法 |
| CN102126221A (zh) * | 2010-12-23 | 2011-07-20 | 中国科学院自动化研究所 | 一种基于图像信息的机械手抓取物体的方法 |
| CN105184019A (zh) * | 2015-10-12 | 2015-12-23 | 中国科学院自动化研究所 | 机器人抓取方法及*** |
| CN105598965A (zh) * | 2015-11-26 | 2016-05-25 | 哈尔滨工业大学 | 基于立体视觉的机器人欠驱动手自主抓取方法 |
| CN105856232A (zh) * | 2016-05-30 | 2016-08-17 | 先驱智能机械（深圳）有限公司 | 物体的抓取方法及抓取*** |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109986560A (zh) * | 2019-03-19 | 2019-07-09 | 埃夫特智能装备股份有限公司 | 一种面向多目标种类的机械臂自适应抓取方法 |
| CN113538459A (zh) * | 2021-07-07 | 2021-10-22 | 重庆大学 | 一种基于落点区域检测的多模式抓取避障检测优化方法 |
| CN113538459B (zh) * | 2021-07-07 | 2023-08-11 | 重庆大学 | 一种基于落点区域检测的多模式抓取避障检测优化方法 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109153118A (zh) | 2019-01-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liang et al. | Pointnetgpd: Detecting grasp configurations from point sets | |
| CN109015640B (zh) | 抓取方法、***、计算机装置及可读存储介质 | |
| Schmidt et al. | Grasping of unknown objects using deep convolutional neural networks based on depth images | |
| Kappler et al. | Leveraging big data for grasp planning | |
| CN108983978B (zh) | 虚拟手控制方法及装置 | |
| JP2019217632A (ja) | ロボットの把持のための深層機械学習方法および装置 | |
| Guo et al. | Robotic grasping using visual and tactile sensing | |
| CN110298886B (zh) | 一种基于四级卷积神经网络的灵巧手抓取规划方法 | |
| Zhang et al. | Robotic grasp detection based on image processing and random forest | |
| CN110785268A (zh) | 用于语义机器人抓取的机器学习方法和装置 | |
| Jiang et al. | Learning hardware agnostic grasps for a universal jamming gripper | |
| Tang et al. | Learning collaborative pushing and grasping policies in dense clutter | |
| Ottenhaus et al. | Visuo-haptic grasping of unknown objects based on gaussian process implicit surfaces and deep learning | |
| Aktaş et al. | Deep dexterous grasping of novel objects from a single view | |
| Kirby et al. | Comparing single touch to dynamic exploratory procedures for robotic tactile object recognition | |
| WO2018161305A1 (zh) | 抓取质量检测方法及其应用的方法与*** | |
| JP7051751B2 (ja) | 学習装置、学習方法、学習モデル、検出装置及び把持システム | |
| CN114310892B (zh) | 基于点云数据碰撞检测的物体抓取方法、装置和设备 | |
| CN109508707B (zh) | 基于单目视觉的机器人稳定抓取物体的抓取点获取方法 | |
| Yang et al. | Predict robot grasp outcomes based on multi-modal information | |
| CN113762159A (zh) | 一种基于有向箭头模型的目标抓取检测方法及*** | |
| Detry et al. | Unsupervised learning of predictive parts for cross-object grasp transfer | |
| De Magistris et al. | Experimental force-torque dataset for robot learning of multi-shape insertion | |
| CN114211490B (zh) | 一种基于Transformer模型的机械臂抓手位姿预测方法 | |
| Chen et al. | Robotic grasp control policy with target pre-detection based on deep Q-learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 17899652 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 32PN | Ep: public notification in the ep bulletin as address of the adressee cannot be established |
Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205 DATED 06/02/2020) |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 17899652 Country of ref document: EP Kind code of ref document: A1 |