A robot multi-order behavior planning method based on resampling posture analysis

Abstract

The present invention discloses a robot multi-order behavior planning method based on resampling posture analysis, which relates to the field of robot arm grasping. The method includes setting the camera height, evaluating the grasping range and grasping accuracy of the robot arm; constructing a target posture analysis framework, optimizing the robot arm, establishing a kinematic model, optimizing posture calibration, and reducing model dependence and environmental interference; performing multiple fine sampling detections on the target at the boundary of the field of view to obtain sample data, and performing fusion analysis on the sampled data to generate a resampling edge compensation strategy; accurately positioning the target according to the resampling edge compensation strategy combined with multi-order planning, and the precise positioning includes angle positioning and position positioning; using a dynamic multi-order coupling method to imitate human natural movements to complete the synchronization of arm extension and hand claw opening; conducting grasping experiments and drawing curve graphs, and evaluating. The present application overcomes the problems of limited field of view and grasping errors caused by the existing robot arm grasping relying on passive sensors.

Description

Technical Field

[0001] The present invention relates to the field of robot grasping, and in particular to a robot multi-order behavior planning method based on resampling posture analysis. Background Art

[0002] At present, robotic grasping technology mainly relies on passive sensors and deep learning-based methods. Passive sensors have distortion at the edge of the field of view, which leads to grasping errors, and the limited field of view makes action planning difficult. Although end-to-end learning based on deep learning may solve complex grasping tasks, it faces many difficulties when directly applied to beyond-field-of-view grasping, such as the increase in variables and dimensions caused by complex tasks, the requirement for precise downstream actions, and the need for a large amount of experimental data to train the model during deployment, which consumes a lot of resources and time, and is not conducive to the fast and accurate operation of lightweight desktop tasks.

[0003] In addition, existing research on grasping motion planning is mostly carried out from the perspectives of target complexity, task complexity, and bionic difficulty, such as the grasping quality measurement method proposed by Tan et al., the dataset and deep completion network of Fang et al., the multimodal neural network of Song et al., and the machine learning architecture of Shukla et al. However, fast and accurate grasping and wide-area coverage in unstructured scenes are still challenging, and the complex planning logic involved in precise grasping has not been effectively solved.

[0004] Therefore, a robot multi-order behavior planning method based on resampling posture analysis is urgently needed to solve the above problems. Summary of the invention

[0005] To solve the above problems, the present application provides a robot multi-order behavior planning method based on resampling posture analysis, which aims to solve technical problems related to robotic arm grasping, including overcoming the limited field of view and grasping error problems caused by the existing robotic arm grasping relying on passive sensors. Solve the problems faced by deep learning-based grasping methods in beyond-visual-range grasping, such as complex tasks, multiple variables, high dimensions, difficult deployment, and high resource consumption. Break through the bottleneck that robotic arms are difficult to achieve fast and accurate grasping and wide-area coverage in unstructured scenes. Improve the deficiencies in target complexity, task complexity, bionic difficulty, etc. Solve the problems of complex planning logic involved in precise grasping.

[0006] This paper proposes a robot multi-order behavior planning method based on resampling posture analysis. Through the target posture analysis framework, resampling edge compensation strategy, dynamic multi-order coupling strategy and wide-area grasping application, it avoids tedious learning model construction and lengthy data training, reduces technical complexity, improves grasping success rate, speed and range, and enhances actual operation efficiency and system stability. The specific contents are as follows:

[0007] In this application, a robot multi-order behavior planning method based on resampling posture analysis is provided, comprising the following steps:

[0008] S1. Set the camera height according to the effect of the sensor field of view on the grasping accuracy.

[0009] S2. Explore the impact of several workspaces on the working performance of the robot arm, and evaluate the grasping range and grasping accuracy of the robot arm;

[0010] S3, build a target posture analysis framework, optimize the robot arm, establish a kinematic model, optimize posture calibration, and reduce model dependence and environmental interference;

[0011] S4, performing multiple fine sampling detections on the target at the boundary of the visual area to obtain sample data, and performing fusion analysis on the sample data to generate a resampling edge compensation strategy;

[0012] S5. Accurately positioning the target according to the resampling edge compensation strategy combined with multi-order planning, wherein the accurate positioning includes angle positioning and position positioning;

[0013] S6, using dynamic multi-order coupling method to imitate human natural movements to achieve arm extension and gripper opening synchronization;

[0014] S7. Perform a grasping experiment on steps S3-S6 and draw a curve graph, and then evaluate.

[0015] Preferably, the target posture analysis framework constructed in S3 uses six-dimensional data for grasping detection, and obtains the central grasping coordinates of the object based on the information of the anchor frame.

[0016] Preferably, the specific contents of constructing the target posture analysis framework in S3 include:

[0017] The robot obtains image information through the visual system and uses the deep learning algorithm at the detection node to mark the approximate location of the target;

[0018] The posture correction algorithm of the transformation matrix is ​​used to accurately correct the target posture and obtain the final accurate grasping information.

[0019] Preferably, in S4, the robot arm performs multiple fine samplings on the target at the boundary of the field of view;

[0020] When the robot arm is at the edge of the field of view or the position of the grasped target is beyond the optimal field of view of the sensor, the key information of the blurred area is resampled;

[0021] The target at the boundary of the field of view is sampled multiple times in detail, and the target is gradually approached by locally iterating the data.

[0022] Preferably, the resampling edge compensation function RECS of the resampling edge compensation strategy in S4 is expressed as follows:

[0023]

[0024] Where N is the number of resampling times. i ) is for the i-th sampling, the height between the camera and the target is h i The clarity resolution value of . At a specific time node T and position P, an optimized grasping position P0 is obtained.

[0025] Preferably, in S2, the motion control of the robotic arm can move the camera to a position directly above the grasped target.

[0026] Preferably, the specific content of the dynamic multi-order coupling method in S6 is:

[0027] By iterating the target information and the multi-stage approximation of the robot arm, the operation area of ​​the robot arm is expanded, the error accumulation is controlled, the grasping action is decomposed into multiple stages, and the actions of these stages are dynamically coupled, and dynamic coupling is achieved using continuous visual feedback.

[0028] Preferably, the camera height in the grasping experiment of S7 is 250 mm.

[0029] In summary, the robot multi-order behavior planning method based on resampling posture analysis of the present invention has the following advantages over traditional technologies:

[0030] 1. The present invention optimizes the robot arm, establishes a kinematic model, optimizes posture calibration, reduces model dependence and environmental interference, and improves system robustness and reliability. In order to make the grasping more accurate, the grasping posture is further represented as 6-dimensional information.

[0031] 2. The present invention proposes a resampling edge compensation strategy. This strategy aims to make up for the potential limitations of single sampling, implement multiple fine sampling detections on targets at the boundary of the field of view, fuse multiple sampling data for analysis, and combine multi-order planning to accurately locate the target from different angles and positions.

[0032] 3. The present invention adopts multi-order planning and further studies on the resampling edge compensation strategy to achieve field of view expansion. By iterating target information and multi-stage approximation of the robot arm, the operation area of ​​the robot arm is expanded and the error accumulation is controlled.

[0033] 4. This application uses continuous visual feedback to achieve dynamic coupling, reduce redundant stagnation in action switching, streamline calculations and shorten response time. In addition, it imitates natural human movements to achieve synchronization between arm extension and gripper opening, optimizes the grasping process, and improves efficiency and accuracy.

[0034] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a schematic diagram of a wide-area precision grasping device of a robot multi-order behavior planning method based on resampling posture analysis of the present invention;

[0036] Figure 2 It is a process framework diagram of a robot multi-order behavior planning method based on resampling posture analysis of the present invention;

[0037] Figure 3 A schematic diagram of a target posture analysis framework capture of a robot multi-order behavior planning method based on resampling posture analysis according to the present invention;

[0038] Figure 4 It is a target exhaustive search verification and multi-order posture verification diagram in a simulation environment of a robot multi-order behavior planning method based on resampling posture analysis of the present invention;

[0039] Figure 5 A stereoscopic view structure diagram of a robot vision sensor of a robot multi-order behavior planning method based on resampling posture analysis of the present invention;

[0040] Figure 6 This is a test verification diagram of the edge grasping performance of each component of a robot multi-order behavior planning method based on resampling posture analysis of the present invention;

[0041] Figure 7 This is a schematic diagram of relevant information about the exploration field of view of a robot multi-order behavior planning method based on resampling posture analysis in the present invention. DETAILED DESCRIPTION

[0042] The technical solution of the present invention is further described below by means of the accompanying drawings and embodiments. It should be noted that unless otherwise specifically stated, the relative arrangement of components and steps, numerical expressions and numerical values ​​described in these embodiments do not limit the scope of the present application.

[0043] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the present application, its application, or uses.

[0044] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.

[0045] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0046] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.

[0047] like Figure 1 , Figure 2 and Figure 3 As shown, Figure 1 It is a wide-area precision grasping system, consisting of a main control computer, a collaborative robot arm, a visual sensor and a gripper. Figure 2 The resampling posture analysis method includes the target posture analysis framework, the resampling edge compensation algorithm and the dynamic multi-order coupling algorithm. After the target information is input, it is preprocessed to obtain the anchor frame, and then the control actuator is used to perform the posture correction algorithm to correct the TF information. Finally, the iterative recognition data is sent to the execution controller to achieve accurate grasping. Figure 3 It is the target posture analysis framework grasping parameter. The plane parameters of the grip and the target under the pre-sampling posture include length, width and angle information. In the target recognition box displayed below the sensor, the application uses yellow lines to represent the length of the target and red lines to represent the width of the target. The present invention proposes in this application that a robot multi-order behavior planning method based on resampling posture analysis is provided, comprising the following steps:

[0048] S1. Set the camera height according to the effect of the sensor field of view on the grasping accuracy.

[0049] S2. Explore the impact of several workspaces on the working performance of the robot arm, and evaluate the grasping range and grasping accuracy of the robot arm;

[0050] Furthermore, in S2, the motion control of the robotic arm can enable the camera to move to a position directly above the grasped target.

[0051] S3, build a target posture analysis framework, optimize the robot arm, establish a kinematic model, optimize posture calibration, and reduce model dependence and environmental interference;

[0052] Furthermore, the target posture analysis framework constructed in S3 uses six-dimensional data for grasping detection, and obtains the central grasping coordinates of the object based on the information of the anchor frame.

[0053] Furthermore, the specific contents of constructing the target posture parsing framework in S3 include:

[0054] The robot obtains image information through the visual system and uses the deep learning algorithm at the detection node to mark the approximate location of the target;

[0055] The posture correction algorithm of the transformation matrix is ​​used to accurately correct the target posture and obtain the final accurate grasping information.

[0056] The visual system acquires image information, the deep learning algorithm is used for preliminary positioning of the target, and the posture correction algorithm of the transformation matrix is ​​used for precise correction.

[0057] In deep learning algorithms, different neural network architectures can be selected, such as convolutional neural network (CNN) or recurrent neural network (RNN), to adapt to different scenario requirements.

[0058] The target posture analysis framework generally uses five-dimensional data, expressed as follows:

[0059] g={x,y,h,w,θ}

[0060] Where {x, y} is the coordinate of the center point of the grabbing rectangle. The center of the rectangle is also the grabbing point, which is also agreed by the present invention. {x, y} is the size of the target, and θ is the angle between the grabbing rectangle and the horizontal axis. In order to make the grabbing more accurate, the present application further expresses the grabbing posture as 6-dimensional information, and the expression is as follows:

[0061]

[0062] G D Indicates that what needs to be grasped is the spatial information of the object, where The center grasping coordinates of the object are obtained based on the information of the anchor frame, where H represents the width of the anchor frame and L represents the length of the anchor frame. Z is the depth information of the target grasping point, δ is the offset angle between the anchor frame and the baselink, which describes the axis offset of the target, Q is the grasping quality score of the target grasping point, and W D Indicates the width of the gripping area for items, which affects the opening and closing of the grip.

[0063] S4, performing multiple fine sampling detections on the target at the boundary of the visual area to obtain sample data, and performing fusion analysis on the sample data to generate a resampling edge compensation strategy;

[0064] In previous studies, visual grasping systems usually rely on single recognition of traditional target detection algorithms to determine the grasped target. However, this method has the problem of center bias error, especially when the target is not in the center of the visual frame. In view of this, this application innovatively proposes a resampling edge compensation strategy. The resampling edge compensation strategy aims to make up for the potential limitations of single sampling, implement multiple fine sampling detections on targets at the boundary of the field of view, and fuse multiple sampling data for analysis, while combining multi-order planning to accurately locate the target from different angles and positions. For example, the grasping field of view under the 1 / 4 field of view also clearly shows the positioning differences at different positions.

[0065] This strategy allows to collect visual information about the object from multiple viewpoints. Using the motion control of the robot arm, the camera can be moved to a position directly above the grasped object.

[0066] Furthermore, in S4, the robot arm performs multiple fine samplings of the targets at the boundary of the field of view;

[0067] When the robot arm is at the edge of the field of view or the position of the grasped target is beyond the optimal field of view of the sensor, the key information of the fuzzy area is resampled, and the frequency and amplitude of the resampling can be adjusted according to the actual situation;

[0068] The target at the boundary of the field of view is sampled multiple times in detail, and the target is gradually approached by locally iterating the data.

[0069] Furthermore, the resampling edge compensation function RECS of the resampling edge compensation strategy in S4 is expressed as follows:

[0070]

[0071] Where N is the number of resampling times. i ) is for the i-th sampling, the height between the camera and the target is h i The clarity resolution value of . At a specific time node T and position P, an optimized grasping position P0 is obtained.

[0072] S5. Accurately positioning the target according to the resampling edge compensation strategy combined with multi-order planning, wherein the accurate positioning includes angle positioning and position positioning;

[0073] Multi-order planning is a further study based on the resampling edge compensation strategy, which is used to achieve field of view expansion. It improves the spatial recognition method, expands the robot's operating area and controls error accumulation by iterating target information and multi-stage approximation of the robot. At the same time, dynamic coupling is achieved using continuous visual feedback, reducing redundant stagnation of action switching, streamlining calculations and shortening response time. In addition, it also imitates natural human movements to achieve synchronization between arm extension and gripper opening, optimizes the grasping process, and improves efficiency and accuracy. The optimization strategy aims to optimize grasping performance by minimizing the total time and maximizing the grasping success rate.

[0074] S6, using dynamic multi-order coupling method to imitate human natural movements to achieve arm extension and gripper opening synchronization;

[0075] Multi-order planning is a further study based on the resampling edge compensation strategy, which is used to achieve field of view expansion. It improves the spatial recognition method, expands the robot's operating area and controls error accumulation by iterating target information and multi-stage approximation of the robot. At the same time, dynamic coupling is achieved using continuous visual feedback, reducing redundant stagnation of action switching, streamlining calculations and shortening response time. In addition, it also imitates natural human movements to achieve synchronization between arm extension and gripper opening, optimizes the grasping process, and improves efficiency and accuracy. The optimization strategy aims to optimize grasping performance by minimizing the total time and maximizing the grasping success rate.

[0076] Furthermore, the specific content of the dynamic multi-order coupling method in S6 is:

[0077] By iterating the target information and the multi-stage approximation of the robot arm, the operation area of ​​the robot arm is expanded, the error accumulation is controlled, the grasping action is decomposed into multiple stages, and the actions of these stages are dynamically coupled, and dynamic coupling is achieved using continuous visual feedback.

[0078] Transform fast grasping and wide-area coverage grasping into efficient trajectory planning problems. By decomposing the grasping action into multiple stages and dynamically coupling the actions of these stages, the work accuracy and overall speed can be improved.

[0079] Different motion decomposition and coupling methods can be selected according to the performance and task requirements of the robot arm.

[0080] S7. Perform a grasping experiment on steps S3-S6 and draw a curve graph, and then evaluate.

[0081] Furthermore, the camera height in the grasping experiment of S7 is 250mm.

[0082] Wide-area grasping applications use multi-order planning to achieve exhaustive search of targets. The robot arm attempts to grasp objects exhaustively based on the full coverage of the grasping range.

[0083] The strategy and scope of the exhaustive enumeration can be adjusted according to the actual environment and task requirements.

[0084] Non-preferred implementation solutions and differences:

[0085] In the target pose parsing framework, traditional image processing methods are used instead of deep learning algorithms for preliminary target positioning, but their accuracy and efficiency may be low.

[0086] In the resampling edge compensation strategy, external data is used to supplement rather than local iterative data, but this may increase the difficulty and time cost of data acquisition.

[0087] In the dynamic multi-stage coupling strategy, the use of fixed action stages and coupling methods without dynamic adjustment may lead to poor adaptability and flexibility.

[0088] The present invention provides an innovative robot arm control method, which realizes more accurate, efficient and intelligent control of the robot arm by integrating advanced sensor technology, optimized control algorithm and deep learning model, and has significant beneficial effects. It greatly improves the operation accuracy of the robot arm, significantly reduces the errors of actions such as grabbing and placing, thereby ensuring the high quality of the product; significantly improves work efficiency, greatly shortens the operation time through optimized motion trajectory planning and fast response mechanism, and brings higher production capacity to production; enhances the adaptability to complex and changeable working environment and different operation objects, can flexibly respond to various working conditions, and broadens the application scenarios of the robot arm; effectively reduces energy consumption, and intelligent control strategy and path planning reduce energy consumption in the operation of the robot arm, saving costs; the operation becomes simpler, reduces the high requirements for the professional skills of the operator, and makes the operation process easy to master; avoids subjective errors and inconsistencies that may occur in manual operation, and ensures the stability and reliability of the operation; reduces the waste of raw materials and equipment damage caused by operating errors, and reduces production costs; reduces waste emissions and noise pollution in the production process, and optimizes the working environment. In summary, the robot arm control method of the present invention significantly improves economic benefits and production efficiency, reduces time costs and environmental impact, and has wide application value and important practical significance.

[0089] Exhaustive target search is an innovative application of the present invention, which aims to evaluate the performance of beyond-horizon target search. Figure 4 As shown, target exhaustive search verification and multi-stage posture verification in simulation environment, Figure 4 The exhaustive target grasping experiment conducted in a simulation environment was demonstrated, and the experimental results proved that the reachable range of the robotic arm completely covers the maximum working radius of 0-180 degrees. Figure 4 The lower part further shows the multi-stage posture solution process of the robot arm after searching for the target. The two sets of experiments show this process from different perspectives. The yellow and green boxes in the figure mark the starting positions of the two sets of experiments, while the red box indicates the ending position.

[0090] like Figure 5 As shown in the figure, the stereoscopic field of view structure of the robot vision sensor shows the installation method of two sets of cameras. The present invention tests the grasping performance of the extreme field of view in the environment of the specified target, and the test targets are randomly distributed on the working plane. In order to accurately measure the grasping performance under different environmental conditions, four core indicators are evaluated, namely edge grasping ability, grasping success rate, benchmark sampling time and grasping time, as shown in Table 1. Edge sampling time refers to the time taken from the sensor working to identifying the grasping information of the edge position target. This process only records the time without judging the accuracy.

[0091] Table 1 Comparison of grasping parameters between different methods at edge position 1

[0092]

[0093] Experiments show that in the grasping success rate test, Kaiming et al.'s Mask R-CNN method is 54.9%. Zeng et al.'s SSD method can only reach 55.3%. Kumra et al.'s DETR method is slightly improved to 60.8%, while their YOLOv4 method can exceed 75%. At the same time, the edge 1 sampling time of this method is 10.14s. However, among all the methods, only the strategy of the present invention can perform well in edge recognition. Thanks to RPAG, the present invention performs well in edge recognition, grasping success rate and time. The edge 1 grasping success rate is 88.1% and the edge 1 sampling time is 4.93s. In order to further verify all the methods proposed in the present invention. Set the components RECS, DC, DMP, DMCS, and RPAG, which represent the resampling edge compensation strategy, dynamic coupling, dynamic multi-order planning, dynamic multi-order coupling strategy, and resampling posture analysis grasping strategy, respectively.

[0094] Since the target posture analysis framework integrates the posture correction algorithm, and the dynamic multi-order coupling strategy integrates dynamic coupling and dynamic multi-order planning. In order to deeply explore the impact of each downstream component and algorithm on the upstream strategy, the present invention has carried out a detailed experimental design. In order to intuitively show the performance, the present invention draws the following Figure 6 The graph shown. Figure 6 The edge grasping performance test verification of each component, the baseline model is abbreviated as BL, the X-axis represents the model, and the Y-axis represents the edge grasping performance. The experimental results show that the grasping strategy equipped with RPAG significantly exceeds the success rate of other control groups while maintaining similar running time. The DMP multi-order planning alone will slightly increase the running time, which is 0.27s higher than the baseline model. However, the introduction of dynamic coupling significantly improves the speed, making the DMP with dynamic coupling 9.99s faster than the baseline model.

[0095] This paper explores the influence of the sensor field of view on the grasping accuracy, where ψ(ρ) is a fixed value. Figure 1 As shown, Figure 7 Explore relevant information about the field of vision. Figure 7 (a) shows the relationship between camera height and grasping accuracy. Figure 7 (b) is a test of the field of view and grasping success rate in two ranges. The experimental results shown in (a) of 7 show that in the 300 grasping experiments conducted, the present invention observed a higher pre-sampled posture P tcp Capable of capturing a wider field of view tcp, but this will sacrifice grasping accuracy. The opposite setting, that is, giving priority to clarity, will lead to a decrease in grasping accuracy. The experimental results show that the camera height h = 250mm is the most appropriate choice. In order to explore the impact of different sizes of workspace on the working performance of the robot arm, the present invention designed two sets of experiments. The grasping range and accuracy tests were carried out on the 1 / 2 and 1 / 4 fan-shaped desktops, respectively. Figure 7 As shown in (b). The fields of view of the two groups are consistent and only depend on h and S(h). Thanks to the multi-stage grasping that can adjust the grasping posture, its success rate exceeds the performance of the baseline model.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A robot multi-order behavior planning method based on resampling posture analysis, characterized in that: The following steps are involved: S1. Set the camera height according to the effect of the sensor field of view on the grasping accuracy. S2. Explore the impact of several workspaces on the working performance of the robot arm, and evaluate the grasping range and grasping accuracy of the robot arm; S3, build a target posture analysis framework, optimize the robot arm, establish a kinematic model, optimize posture calibration, and reduce model dependence and environmental interference; S4, performing multiple fine sampling detections on the target at the boundary of the visual area to obtain sample data, and performing fusion analysis on the sample data to generate a resampling edge compensation strategy; S5. Accurately positioning the target according to the resampling edge compensation strategy combined with multi-order planning, wherein the accurate positioning includes angle positioning and position positioning; S6, using dynamic multi-order coupling method to imitate human natural movements to achieve arm extension and gripper opening synchronization; S7. Perform a grasping experiment on steps S3-S6 and draw a curve graph, and then evaluate.

2. A robot multi-order behavior planning method based on resampling posture analysis according to claim 1, characterized in that: The target posture analysis framework constructed in S3 uses six-dimensional data for grasping detection and obtains the central grasping coordinates of the object based on the information of the anchor frame.

3. A robot multi-order behavior planning method based on resampling posture analysis according to claim 1, characterized in that: The specific contents of constructing the target posture parsing framework in S3 include: The robot obtains image information through the visual system and uses the deep learning algorithm at the detection node to mark the approximate location of the target; The posture correction algorithm of the transformation matrix is ​​used to accurately correct the target posture and obtain the final accurate grasping information.

4. A robot multi-order behavior planning method based on resampling posture analysis according to claim 1, characterized in that: In S4, the robot performs multiple fine sampling of the targets at the boundary of the field of view; When the robot arm is at the edge of the field of view or the position of the grasped target is beyond the optimal field of view of the sensor, the key information of the blurred area is resampled; The target at the boundary of the field of view is sampled multiple times in detail, and the target is gradually approached by locally iterating the data.

5. A robot multi-order behavior planning method based on resampling posture analysis according to claim 1, characterized in that: The resampling edge compensation function RECS of the resampling edge compensation strategy in S4 is expressed as follows: Where N represents the number of resampling times, R(h i ) is for the i-th sampling, the height between the camera and the target is h i The clarity resolution value is used to obtain an optimized grasping position P0 at a specific time node T and position P.

6. A robot multi-order behavior planning method based on resampling posture analysis according to claim 1, characterized in that: In S2, the motion control of the robot arm can move the camera to a position directly above the grasped object.

7. A robot multi-order behavior planning method based on resampling posture analysis according to claim 1, characterized in that: The specific contents of the dynamic multi-order coupling method in S6 are: By iterating the target information and the multi-stage approximation of the robot arm, the operation area of ​​the robot arm is expanded, the error accumulation is controlled, the grasping action is decomposed into multiple stages, and the actions of these stages are dynamically coupled, and dynamic coupling is achieved using continuous visual feedback.

8. The robot multi-order behavior planning method based on resampling posture analysis according to claim 1 is characterized in that: The camera height in the grasping experiment of S7 is 250mm.

Images