CN-122008259-A - Unordered grabbing method and unordered grabbing system for robots

CN122008259ACN 122008259 ACN122008259 ACN 122008259ACN-122008259-A

Abstract

The invention relates to the technical field of robot grabbing and discloses a method and a system for unordered grabbing of a robot, wherein the method comprises the steps of collecting three-dimensional data of a scene, identifying a target object, and generating grabbing pose and withdrawing direction; the method comprises the steps of grabbing a target object according to grabbing pose, executing verification lifting, collecting verification data, obtaining state judgment data according to the verification data, obtaining history data, building and training a state judgment model, calling the state judgment model, carrying out state judgment according to the verification data, generating countermeasures and updating grabbing pose and withdrawing direction when a state judgment result is in an unstable grabbing state, and completing the taking out of the target object according to the current withdrawing direction when the state judgment result is in a stable grabbing state. The method can improve the grabbing stability, the picking success rate and the continuous operation reliability of the robot in a disordered stacking scene, reduce the probability of slipping, rubbing and blocking of the target object in the picking process, and is beneficial to improving the overall grabbing efficiency.

Inventors

XU JUNFENG
LIU XUAN
TANG XIANG
ZOU YUXIN
GUO JIANQIANG
YAN QIUYAN
CHENG YU

Assignees

湘潭理工学院

Dates

Publication Date: 20260512
Application Date: 20260415

Claims (7)

1. A method for unordered grasping by a robot, comprising: Step S1, acquiring scene three-dimensional data of a working area, identifying a target object, and generating a grabbing pose and an evacuating direction; s2, controlling the robot to grasp the target object according to the grasping pose, executing verification lifting, collecting verification data, and obtaining state judgment data according to the verification data; Step S3, acquiring historical data, and building and training a state judgment model based on the historical data; S4, calling a state judgment model, carrying out state judgment according to the verification data, and generating countermeasures and updating the grabbing pose and the withdrawing direction when the state judgment result is in an unstable grabbing state; and S5, when the state judgment result is in a stable grabbing state, completing the taking out of the target object according to the current evacuation direction.
2. The unordered grabbing method of a robot according to claim 1, wherein the step S1 specifically includes: performing background separation on the three-dimensional scene data, removing a fixed scene area, and reserving entity data in a grabbing space; Dividing the reserved entity data to obtain a space contour corresponding to each target object; extracting a surface area exposed towards the grabbing space for each target object, calculating the center position and the surface orientation of the surface area, and generating the grabbing pose of the target object; Taking the reverse direction of the surface orientation as an initial evacuation direction, and detecting the occupancy condition in the three-dimensional data of the scene along the initial evacuation direction; when interference exists in the initial evacuation direction, the direction is adjusted and re-detected until the interference-free evacuation direction is obtained; Outputting the corresponding grabbing pose and the withdrawing direction of each target object.
3. The unordered grabbing method of a robot according to claim 2, wherein the step S2 specifically includes: setting an instruction lifting displacement, and carrying out short-range lifting along the evacuation direction after the robot finishes clamping the target object, wherein the short-range lifting stroke is the instruction lifting displacement; In the process of verifying lifting, collecting verification data, wherein the verification data comprises end effector feedback data, robot motion response data and local rechecking three-dimensional data; the end effector feedback data are clamping jaw opening amount, clamping force and driving current; The motion response data of the robot are command lifting displacement, actual lifting displacement and joint moment; the local rechecking of the three-dimensional data refers to local point cloud data acquired from a grabbing area in the verification lifting process; determining space occupation change data in the lifting height and the evacuation direction of the target object according to the local recheck three-dimensional data; The end effector feedback data, the robot motion response data, and the space occupation change data in the lifting height and the evacuation direction of the target object are used as the state determination data.
4. A method of unordered gripping of a robot according to claim 3, wherein the acquiring historical data, and the building and training of the state determination model based on the historical data comprises: The historical data comprises historical state judging data and corresponding labeling data, and the labeling data labels the state judging result of the historical state judging data; The state judgment result comprises a stable grabbing state and an unstable grabbing state, and the unstable grabbing state comprises an offset clamping state, a target failure state and a channel failure state; establishing a state judgment model, wherein the state judgment model comprises a first time convolution branch, a second time convolution branch, a third time convolution branch, a feature fusion layer, a first-stage judgment layer and a second-stage judgment layer; training a state judgment model in a hierarchical supervised training mode, wherein a primary judgment layer takes a stable grabbing state and an unstable grabbing state in marked data as a primary supervision tag; The second-level judging layer only carries out supervision training on historical state judging data marked as an unstable grabbing state, and takes an offset clamping state, a target failure state and a channel failure state in the marked data as a second-level supervision tag. Performing supervision training on the primary judgment layer by adopting binary focus loss, performing supervision training on the secondary judgment layer by adopting multi-class focus loss, and taking weighted summation of the binary focus loss and the multi-class focus loss as joint training loss; inputting historical state judgment data into a state judgment model, and updating model parameters by adopting a back propagation and adaptive moment estimation optimization algorithm according to the joint training loss; and stopping training when the joint training loss converges or reaches a preset training round, and obtaining a training-completed state judgment model.
5. The method of unordered grasping of a robot according to claim 4, wherein the state determination model includes: the first time convolution branch is used for extracting time sequence characteristics in the feedback data of the historical end effector, the second time convolution branch is used for extracting time sequence characteristics in the motion response data of the historical robot, and the third time convolution branch is used for extracting time sequence characteristics in the space occupation change data in the lifting height and the evacuation direction of the historical target object; The time convolution branches adopt multi-scale time convolution blocks formed by one-dimensional causal expansion convolution and residual error connection; performing first fusion on the time sequence characteristics output by the first time convolution branch and the time sequence characteristics output by the second time convolution branch to obtain deviation judging characteristics; Performing second fusion on the time sequence characteristics output by the first time convolution branch and the time sequence characteristics output by the third time convolution branch to obtain target failure judgment characteristics; thirdly, fusing the time sequence characteristics output by the second time convolution branch with the time sequence characteristics output by the third time convolution branch to obtain channel failure judgment characteristics; then comprehensively fusing the time sequence characteristics output by all the time convolution branches with the deviation judging characteristics, the target failure judging characteristics and the channel failure judging characteristics to obtain global judging characteristics; inputting the global judging characteristic into a first-stage judging layer, and outputting a first-stage judging result that the current grabbing result belongs to a stable grabbing state or an unstable grabbing state; inputting the deviation judging feature, the target failure judging feature and the channel failure judging feature into a secondary judging layer, and outputting a secondary judging result that the current grabbing result belongs to the deviation clamping state, the target failure state or the channel failure state.
6. A robotic unordered grasping method as in claim 5 wherein the countermeasures include: When the state judgment result is in a deviation clamping state, generating deviation countermeasures; Determining the offset direction and the offset of the current grabbing pose according to the feedback data of the end effector and the motion response data of the robot, controlling the robot to release a target object, collecting scene three-dimensional data again, identifying the target object according to the scene three-dimensional data collected again, correcting the grabbing pose generating process according to the offset direction and the offset, generating the grabbing pose and the withdrawing direction again, obtaining the updated grabbing pose and the updated withdrawing direction, controlling the robot to grab the target object again according to the updated grabbing pose and the updated withdrawing direction, and executing verification lifting again; when the state judgment result is the target failure state, generating a target failure countermeasure; the robot is controlled to release the target object, the scene three-dimensional data are collected again, the target object is identified according to the scene three-dimensional data which are collected again, the avoidance area is determined according to the contact area corresponding to the current grabbing pose on the surface of the target object, the grabbing pose and the withdrawing direction are regenerated, the updated grabbing pose and the updated withdrawing direction are obtained, the robot is controlled to grab the target object again according to the updated grabbing pose and the updated withdrawing direction, and verification lifting is executed again; when the state judgment result is a channel failure state, generating a channel failure countermeasure; The current grabbing pose is reserved, the current evacuation direction is adjusted according to the space occupation change data in the evacuation direction, the updated evacuation direction is obtained, the robot is controlled to keep clamping the target object, and the target object is taken out along the updated evacuation direction.
7. A robot unordered grabbing system is applied to the unordered grabbing method of the robot, which is characterized by comprising the following steps of, The acquisition module acquires scene three-dimensional data of the operation area and identifies a target object, and generates a grabbing pose and an evacuation direction; the verification module is used for controlling the robot to grasp the target object according to the grasping pose, executing verification lifting and collecting verification data, and obtaining state judgment data according to the verification data; The modeling module is used for acquiring historical data, and establishing and training a state judgment model based on the historical data; the judging and correcting module is used for calling a state judging model, judging the state according to the verification data, generating countermeasures and updating the grabbing pose and the withdrawing direction when the state judging result is in an unstable grabbing state; And the execution module is used for completing the taking out of the target object according to the current evacuation direction when the state judgment result is in the stable grabbing state.

Description

Unordered grabbing method and unordered grabbing system for robots Technical Field The invention relates to the technical field of robot grabbing, in particular to a method and a system for unordered grabbing of a robot. Background In robotic grasping scenarios, the target objects are often distributed in a disordered stack within a bin, turn-around bin, or table area. The robot needs to recognize the target object first, then determine the grabbing position and the taking-out direction, and then complete the clamping and the taking-out. In a disordered stacking scenario, there is typically shielding, pressing, bridging, and local interference between the target objects, and the exposed surfaces, accessible areas, and extraction channels of the target objects may change as the stacking state changes, thereby making the gripping process more prone to instability. The conventional disordered grabbing mode of the robot is mainly focused on grabbing pretreatment, such as three-dimensional visual recognition of a target object, generation of grabbing pose and determination of taking out direction, and then the robot is controlled to execute grabbing and taking out. Although the mode can finish target recognition and path planning before gripping, the method lacks specific judgment on the actual gripping state of the robot after gripping. That is, after the robot completes the clamping action, whether the target object has formed a stable bearing relationship and can be smoothly taken out along the current direction often cannot be reflected until the robot is formally taken out. Thus, once the gripping position is shifted, the gripping relationship is unstable, or the front passage is blocked, slipping, rubbing, jamming, or removal failure is liable to occur in the subsequent removal process. In addition, the conventional processing method is common to re-grab or re-plan after the grabbing failure, but the distinction between different failure sources is not available. In practice, the problems of the displacement of the gripping position, the failure of the gripping position and the blockage of the removal channel are of different types, and the corresponding adjustment objects are not the same, and some problems require the readjustment of the gripping position, some problems require the reselection of the contact area, and some problems only require the change of the removal direction. If the same rollback or retry mode is adopted for different situations, not only the invalid adjustment times are increased, but also the robot can be caused to repeatedly fail at the same position, and the unordered grabbing efficiency and grabbing stability are affected. Disclosure of Invention The present invention has been made in view of the above-described problems. The unordered grabbing method of the robot comprises the following steps of S1, collecting three-dimensional scene data of an operation area, identifying a target object, and generating grabbing pose and withdrawing direction; s2, controlling the robot to grasp the target object according to the grasping pose, executing verification lifting, collecting verification data, and obtaining state judgment data according to the verification data; Step S3, acquiring historical data, and building and training a state judgment model based on the historical data; S4, calling a state judgment model, carrying out state judgment according to the verification data, and generating countermeasures and updating the grabbing pose and the withdrawing direction when the state judgment result is in an unstable grabbing state; and S5, when the state judgment result is in a stable grabbing state, completing the taking out of the target object according to the current evacuation direction. As a preferable scheme of the unordered grabbing method of the robot, the method comprises the following steps of: performing background separation on the three-dimensional scene data, removing a fixed scene area, and reserving entity data in a grabbing space; Dividing the reserved entity data to obtain a space contour corresponding to each target object; extracting a surface area exposed towards the grabbing space for each target object, calculating the center position and the surface orientation of the surface area, and generating the grabbing pose of the target object; Taking the reverse direction of the surface orientation as an initial evacuation direction, and detecting the occupancy condition in the three-dimensional data of the scene along the initial evacuation direction; when interference exists in the initial evacuation direction, the direction is adjusted and re-detected until the interference-free evacuation direction is obtained; Outputting the corresponding grabbing pose and the withdrawing direction of each target object. As a preferable scheme of the unordered grabbing method of the robot, the method comprises the following step S2: setting an instruction lifting displacement, and c