Industrial robot container cargo grabbing control method

Abstract

The invention discloses an industrial robot container cargo grabbing control method. The method comprises the following steps that S1, cargo data are collected, and an initial data set is generated; s2, adaptively fusing multi-modal data to generate a cargo scene model; s3, cargo features are extracted through a squirrel optimization algorithm, and a feature database is generated; s4, hierarchical topology modeling is carried out to construct a grabbing sequence; s5, dynamically analyzing the topology model to adjust a grabbing sequence and a path; s6, the form and grabbing parameters of the flexible clamp are adjusted; s7, the game theory drives the multiple mechanical arms to collaboratively optimize task allocation; and S8, performing causal reasoning self-repairing optimization on the grabbing task. According to the method, the high efficiency, the intelligence and the robustness of goods grabbing of the industrial robot in a complex stacking scene are realized through multi-modal sensing, a self-adaptive flexible clamp, hierarchical topology modeling, game theory optimization and a causal reasoning self-repairing technology.

Description

Technical Field

[0001] The present invention relates to the field of industrial robots, and in particular to a method for controlling an industrial robot to grasp container cargo. Background Art

[0002] With the rapid development of industrial automation, industrial robots, as one of the core equipment of modern industrial production, are widely used in logistics, manufacturing, warehousing and other fields. Especially in the task of grabbing container cargo, industrial robots have greatly improved the efficiency of cargo handling and reduced the complexity of manual operation through intelligent control and collaborative work. However, with the diversification of container cargo types and the complexity of stacking methods, existing technologies still face many challenges and technical bottlenecks when dealing with dynamic and complex stacking scenarios.

[0003] In the existing technology, industrial robots mostly rely on single-mode perception methods, such as using two-dimensional vision or laser radar to collect basic information about goods. Although these technologies perform well in a single scenario, the limitations of single-mode perception are apparent in complex environments such as occlusion, lighting changes, and identification of special-shaped goods, resulting in a significant decrease in the accuracy of goods identification. In addition, existing perception algorithms usually use fixed parameter fusion, which cannot dynamically adapt to the weight changes of multimodal data in different scenarios, thereby limiting the system's adaptability to complex scenarios.

[0004] In terms of path planning and task collaboration, existing industrial robots mostly use rule-based static planning methods, such as completing cargo grabbing with fixed paths and priority rules. However, the stacking structure of cargo in containers is usually irregular, and position changes and unstable stacking may occur during the grabbing process. Static planning methods cannot respond to these dynamic changes in real time, resulting in inefficient path planning and may even cause cargo collapse, grabbing failure and other problems. In addition, for multi-robot collaborative tasks, traditional technologies often rely on predefined task allocation strategies, such as simple allocation according to distance or task priority. This method is prone to task conflicts and path interference when multiple robotic arms work together, reducing the overall task execution efficiency of the system.

[0005] In terms of flexible grasping technology, existing industrial robots mostly use rigid clamps or simple flexible material designs, and the grasping shape and force control capabilities of such clamps are limited. When faced with goods of various shapes, especially cylindrical, special-shaped or fragile goods, rigid clamps are difficult to achieve stable grasping of the goods, which can easily cause damage or falling of the goods. Although some technologies have tried to use flexible materials to improve the performance of clamps, during the grasping process, the shape adjustment and force control of the flexible clamps lack a real-time optimization mechanism and cannot adapt to the dynamic changes in the material, weight and shape of the goods.

[0006] In addition, fault handling and system self-repair are another significant shortcoming in existing technologies. In practical applications, industrial robots often interrupt tasks due to problems such as grasping failure, path planning errors, or hardware failures. Existing technologies mostly rely on post-process manual intervention or simple task restart strategies to solve these problems. They lack a highly targeted and self-learning intelligent repair mechanism, which results in long task recovery time and serious waste of resources, making it difficult to meet the needs of efficient operations.

[0007] Therefore, how to provide an industrial robot container cargo grabbing control method is a problem that technical personnel in this field urgently need to solve. Summary of the invention

[0008] One purpose of the present invention is to propose an industrial robot container cargo grabbing control method, which makes full use of multimodal perception technology, adaptive flexible fixture design, hierarchical topology modeling, game theory driven multi-manipulator collaborative optimization, multi-agent reinforcement learning algorithm based on graph attention mechanism and causal reasoning self-repair optimization algorithm, and describes in detail the intelligent grabbing control process from cargo identification, path dynamic optimization, multi-manipulator collaboration to fault self-repair. The method has the advantages of high recognition accuracy, strong dynamic adjustment capability of path planning, high efficiency of multi-manipulator collaboration, strong adaptability of flexible grabbing and intelligent system fault repair.

[0009] According to an embodiment of the present invention, an industrial robot container cargo grabbing control method comprises the following steps:

[0010] S1. Collect point cloud data, RGB images and depth information of goods through LiDAR, 3D camera and depth sensor, and capture the features of goods under occlusion and complex lighting scenes by combining biological vision-inspired algorithms to generate an initial goods data set.

[0011] S2. Adopt an adaptive perception fusion algorithm to fuse the multimodal data in the initial cargo data set, dynamically adjust the weight of each modal data according to the light intensity, occlusion degree and noise level, and generate a unified cargo scene model;

[0012] S3. Optimize and select the features in the cargo scene model based on the kingfisher optimization algorithm, dynamically search for the optimal feature set in the multi-dimensional feature space of the cargo by simulating the jumping search and static ambush mechanism of the kingfisher when foraging, extract the type, shape, size and spatial position features of the cargo, and store the extracted feature data as a cargo feature database;

[0013] S4. Using the hierarchical topology modeling method, the global layout layer, adjacency relationship layer and grasping priority layer of the goods are constructed based on the goods feature database. The global layout layer analyzes the stacking structure of the goods, the adjacency relationship layer describes the spatial constraint relationship between the goods, and the grasping priority layer generates a grasping sequence based on the stability of the goods stacking and the grasping difficulty ranking.

[0014] S5. Combined with the dynamic environment analysis algorithm, the hierarchical topology model is updated in real time. According to the changes in the cargo location, adjacency relationship and grabbing priority, the grabbing sequence and path planning scheme are dynamically adjusted to optimize the cargo grabbing strategy.

[0015] S6. Design a flexible gripper based on shape memory composite materials, adjust the gripper shape through biomechanical flexibility control theory, calculate the grasping force based on the dynamic adaptive force field model, and use capacitive stress sensors to monitor the force distribution during the grasping process, and adjust the gripper shape and grasping parameters in real time;

[0016] S7. Through the multi-robot collaboration model driven by game theory, the cargo grabbing task is decomposed into multiple subtasks. The multi-agent reinforcement learning algorithm based on the graph attention mechanism is combined to optimize the robot task allocation and path planning strategy, and dynamically adjust the collaboration path.

[0017] S8. Use the self-repair optimization algorithm based on causal reasoning to build a causal graph model to analyze abnormal event nodes in the grasping task, combine reverse reasoning to generate a fault resolution path, and adjust the grasping parameters and path planning through intervention optimization technology to complete the repair and adjustment of the grasping process when a fault occurs.

[0018] Optionally, the S3 specifically includes:

[0019] S31, constructing an initial feature space of cargo features, extracting geometric shape, surface texture and size information of the cargo, representing it as a feature set, normalizing the feature set, and generating a standardized feature space;

[0020] S32, based on the standardized feature space, initialize the feature search space, use the Ban Kingfisher optimization algorithm to globally optimize the feature search space, assign feature weights and set initial parameters for feature screening, including search step size and search range;

[0021] S33. According to the optimization strategy, the initial feature set of the goods is screened, all feature dimensions in the feature space are prioritized according to the target feature weights, and a candidate feature set is determined;

[0022] S34, during the screening process, simulate the foraging behavior of the kingfisher, conduct a global search on the initial feature set by means of a jump search method, remove redundant features, and retain features that are highly relevant to the grasping task;

[0023]

[0024] in, is the position of the current optimal individual, ∈ is the random disturbance value, represents the update result of feature weights, represents the feature weight position of the i-th individual at the t-th iteration, and s represents the search step length;

[0025] S35, in the feature candidate set, simulating the ambush behavior of the kingfisher, locally optimizing the feature candidate set, and further converging the range of the feature candidate set to the optimal feature set;

[0026] S36. Verify the optimized feature candidate set, generate a final feature data set by matching it with the actual type, shape and position features of the goods, and store it in the goods feature database.

[0027] Optionally, the S4 specifically includes:

[0028] S41, constructing a global layout layer of goods, representing the goods stacking structure as a topological graph model, and generating a global stacking graph based on the geometric center and physical adjacency relationship of the nodes;

[0029] S42. On the basis of the global layout layer, an adjacency relationship layer is constructed to analyze the adjacency relationship of the cargo nodes in the topological structure, calibrate the adjacency weight according to the contact situation between the nodes, and form an adjacency relationship matrix;

[0030] S43. Based on the adjacency weight of the adjacency relationship layer and the physical properties of the goods, the stability value of each goods node is calculated and a stability distribution map is generated:

[0031]

[0032] Among them, H(v i ) represents the cargo node v i The stack height, Represents node v i and v j The adjacency weight, w h and w n is the weight coefficient, S(v i ) represents the stability score of each node, and n represents the total number of cargo nodes in the cargo stacking topology;

[0033] S44. Take the stability distribution map as input, build a grab priority layer, combine the stability value and grab difficulty of the goods, assign grab priorities to the goods nodes, and form a grab priority list:

[0034] P(v i)=S(v i )-α·D(v i );

[0035] Among them, D(v i ) represents node v i The crawling difficulty coefficient, α is the crawling difficulty weight coefficient, P(v i ) is the cargo grabbing priority;

[0036] S45. According to the grab priority list, sort the cargo nodes from high to low priority, generate a grab sequence, and output the sorted grab sequence to the path planning module.

[0037] Optionally, the S5 specifically includes:

[0038] S51, based on the cargo feature database and the hierarchical topology modeling results, extract the location, adjacency relationship and grab priority information of the cargo nodes, and initialize the cargo grabbing path as a path set;

[0039] S52, constructing a dynamic environment analysis model based on the grabbing order and the goods adjacency relationship in the initial path set, and modeling the change of the goods stacking structure as a time series topological structure;

[0040] S53, analyzing the real-time changes of the cargo node positions and adjacency relationships according to the time series topology, updating the grabbing order in the path set, and calibrating the adjusted grabbing order of the paths according to the topology changes to generate an intermediate path set;

[0041] S54, using the intermediate path set and the path optimization objective function, minimizing the total path time, and generating the intermediate optimized path set:

[0042]

[0043] Among them, d i represents the grabbing moving distance of the path, v represents the moving speed of the robot arm, τ i represents the grabbing and adjustment time of the path, and n represents the number of goods to be grabbed;

[0044] S55, inputting the intermediate optimized path set into the verification module, and verifying the validity of the intermediate optimized path set and the accuracy of the grabbing task according to the cargo grabbing priority and topological state;

[0045] S56: Take the verified intermediate optimization path set as input, generate the final optimized grasping path sequence, and output it to the grasping control module to execute the task.

[0046] Optionally, the S6 specifically includes:

[0047] S61. Based on the cargo feature database and the optimized grasping path sequence, the shape, material and force characteristics of each cargo node are extracted, and the morphological parameter set of the flexible clamp is initialized;

[0048] S62. Design a fixture shape adjustment model, match the shape characteristics of the goods with the fixture shape parameters, and use the dynamic adaptive force field model to calculate the force distribution of the fixture:

[0049]

[0050] Among them, F i is the total force of the clamp when grabbing the object i, w j is the weight of the jth contact point of the fixture, A j is the effective area of ​​the contact point, θ j is the angle between the contact point and the force direction of the fixture, k is the number of fixture contact points;

[0051] S63, dynamically adjusting the contact surface morphological parameters of the fixture according to the total force of the fixture and the material characteristics of the cargo, forming a fixture shape that adapts to the current shape of the cargo, and updating the morphological parameter set;

[0052] S64, using a capacitive stress sensor to monitor the force distribution of each contact point in the grasping process in real time, collecting a force data set, verifying the pressure distribution of the contact point, comparing the monitoring results with the morphological parameter set, identifying abnormal force points and generating a force correction data set;

[0053] S65, taking the generated force correction data set as input, optimizing the pressure distribution of the contact points of the fixture in combination with the dynamic adaptive force field model, redistributing the pressure of each contact point, and generating an optimized morphological parameter set and pressure distribution data;

[0054] S66: Input the optimized morphological parameter set into the flexible control module, execute the cargo grabbing operation, and store the feedback data after the grabbing is completed in the cargo grabbing record.

[0055] Optionally, the S7 specifically includes:

[0056] S71. Based on the optimized grasping path sequence and the initial position of each robotic arm in the grasping task, a multi-robotic arm collaboration model is constructed to decompose the grasping task into a task set;

[0057] S72. Construct a game theory task allocation model for the task set and the robot arm set, and define the objective function as minimizing the task execution time:

[0058]

[0059] Among them, T jrepresents the subset of tasks assigned to the robot arm, m j represents the robot arm, t i represents the task, d i is the grasping path distance of the task, v j is the moving speed of the robot arm, τ i The operation time of the task;

[0060] S73. Use the graph attention mechanism to dynamically model the task set and the robot arm set, extract the conflict relationship and collaboration potential between tasks, and generate a weight matrix.

[0061] S74. Combine the multi-agent reinforcement learning algorithm, use the weight matrix as input, optimize the task allocation strategy and path planning scheme, and generate a dynamically optimized task allocation set;

[0062] S75, verifying the dynamically adjusted task set, evaluating the rationality of task allocation, the conflict rate of path planning, and the real-time performance of task execution;

[0063] S76. Output the verified task set and path planning strategy to the collaborative control module, start the multi-robotic arm collaborative grasping task in combination with the dynamically adjusted collaborative path, monitor the task completion status in real time, and dynamically adjust task allocation and path planning according to abnormal situations.

[0064] Optionally, the S8 specifically includes:

[0065] S81. Use the self-repair optimization algorithm based on causal reasoning to build a causal graph model;

[0066] S82. Identify abnormal event nodes in the crawling task based on the causal graph model, and calculate the causal weight of each abnormal event node:

[0067]

[0068] Among them, δ ij Indicates the causal influence of the abnormal event node, β j and β k represents the severity score of the abnormal event node, w i Represents the causal weight of the abnormal event node, N and N a Represents a set of abnormal event nodes, abnormal event node N a is a subset selected from N, and k is the abnormal event node set N a The index variable of

[0069] S83. Analyze the cause-effect chain of the abnormal event node by combining reverse reasoning, and generate a fault resolution path for each abnormal event node;

[0070] S84, adjusting the grasping parameters by using intervention optimization technology, and updating the grasping parameters in combination with the intervention operation in the fault resolution path;

[0071] S85, based on the updated grasping parameters, adjusting the grasping path set and generating a path plan, wherein the path planning process is combined with dynamic adjustment of the grasping parameters;

[0072] S86. When a fault occurs, use the repaired path planning and grasping parameters to execute the grasping task, monitor the repair effect in real time, and store the fault resolution path and repair data in the task history record.

[0073] The beneficial effects of the present invention are:

[0074] The industrial robot container cargo grabbing control method based on multimodal perception and self-repair optimization proposed in the present invention effectively overcomes many defects of the prior art in complex stacking scenarios and achieves significant beneficial effects. First of all, the present invention adopts multimodal perception technology, combined with lidar, three-dimensional camera and depth sensor, which can achieve high-precision recognition of goods in environments with complex lighting, occlusion and special-shaped goods, and dynamically adjusts the weight of multimodal data through an adaptive fusion algorithm, significantly improving the robustness and adaptability of the system. Compared with traditional single-modal perception methods, the recognition accuracy and reliability of the present invention in dynamic environments have been greatly improved.

[0075] Secondly, the present invention introduces hierarchical topological modeling and dynamic environment analysis technology. By hierarchically modeling the global layout, adjacency and grabbing priority of the cargo stacking structure, it can dynamically capture stacking changes and adjust path planning in real time. By combining the multi-robot collaboration model driven by game theory and the multi-agent reinforcement learning algorithm based on the graph attention mechanism, the problems of task conflict and path interference in traditional multi-robot collaboration are solved, the dynamic optimization of task allocation and the efficiency of path planning are achieved, and the efficiency of multi-robot collaboration is significantly improved.

[0076] In terms of flexible grasping, the present invention uses shape memory composite materials to design flexible clamps, combined with dynamic adaptive force field models and real-time stress monitoring technology, to accurately adapt to the shape and material of the goods, adjust grasping parameters in real time, and effectively avoid damage to the goods or grasping failure. Compared with traditional rigid clamps or fixed flexible clamps, the flexible grasping technology of the present invention shows stronger adaptability and safety.

[0077] In addition, the present invention uses a causal reasoning self-repair optimization algorithm to build a causal graph model to analyze abnormal events in the grasping task, combines reverse reasoning to generate a fault resolution path, and dynamically adjusts the grasping parameters and path planning through intervention optimization technology, thereby achieving online repair of the grasping task. Compared with the existing fault handling method that relies on manual intervention, the present invention greatly shortens the system recovery time and improves the continuity and reliability of task execution.

[0078] To sum up, the present invention can not only realize the accurate identification and dynamic grasping of goods in complex stacking scenarios, but also improve the working efficiency through efficient path optimization and multi-robotic arm collaboration. At the same time, it has flexible grasping and intelligent fault repair capabilities, which significantly improves the adaptability, safety and intelligence level of industrial robots in container cargo grasping tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0080] Figure 1 This is a flow chart of the industrial robot container cargo grabbing control method proposed by the present invention;

[0081] Figure 2 This is a schematic diagram of the structure of the multimodal perception module of the industrial robot container cargo grabbing control method proposed in the present invention. DETAILED DESCRIPTION

[0082] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0083] refer to Figure 1-2 , an industrial robot container cargo grabbing control method, comprising the following steps:

[0084] S1. Collect point cloud data, RGB images and depth information of goods through LiDAR, 3D camera and depth sensor, and capture the features of goods under occlusion and complex lighting scenes by combining biological vision-inspired algorithms to generate an initial goods data set.

[0085] S2. Adopt an adaptive perception fusion algorithm to fuse the multimodal data in the initial cargo data set, dynamically adjust the weight of each modal data according to the light intensity, occlusion degree and noise level, and generate a unified cargo scene model;

[0086] S3. Optimize and select the features in the cargo scene model based on the kingfisher optimization algorithm, dynamically search for the optimal feature set in the multi-dimensional feature space of the cargo by simulating the jumping search and static ambush mechanism of the kingfisher when foraging, extract the type, shape, size and spatial position features of the cargo, and store the extracted feature data as a cargo feature database;

[0087] S4. Using the hierarchical topology modeling method, the global layout layer, adjacency relationship layer and grasping priority layer of the goods are constructed based on the goods feature database. The global layout layer analyzes the stacking structure of the goods, the adjacency relationship layer describes the spatial constraint relationship between the goods, and the grasping priority layer generates a grasping sequence based on the stability of the goods stacking and the grasping difficulty ranking.

[0088] S5. Combined with the dynamic environment analysis algorithm, the hierarchical topology model is updated in real time. According to the changes in the cargo location, adjacency relationship and grabbing priority, the grabbing sequence and path planning scheme are dynamically adjusted to optimize the cargo grabbing strategy.

[0089] S6. Design a flexible gripper based on shape memory composite materials, adjust the gripper shape through biomechanical flexibility control theory, calculate the grasping force based on the dynamic adaptive force field model, and use capacitive stress sensors to monitor the force distribution during the grasping process, and adjust the gripper shape and grasping parameters in real time;

[0090] S7. Through the multi-robot collaboration model driven by game theory, the cargo grabbing task is decomposed into multiple subtasks. The multi-agent reinforcement learning algorithm based on the graph attention mechanism is combined to optimize the robot task allocation and path planning strategy, and dynamically adjust the collaboration path.

[0091] S8. Use the self-repair optimization algorithm based on causal reasoning to build a causal graph model to analyze abnormal event nodes in the grasping task, combine reverse reasoning to generate a fault resolution path, and adjust the grasping parameters and path planning through intervention optimization technology to complete the repair and adjustment of the grasping process when a fault occurs.

[0092] In this implementation, S3 specifically includes:

[0093] S31, constructing an initial feature space of cargo features, extracting geometric shape, surface texture and size information of the cargo, representing it as a feature set, normalizing the feature set, and generating a standardized feature space;

[0094] S32, based on the standardized feature space, initialize the feature search space, use the Ban Kingfisher optimization algorithm to globally optimize the feature search space, assign feature weights and set initial parameters for feature screening, including search step size and search range;

[0095] S33. According to the optimization strategy, the initial feature set of the goods is screened, all feature dimensions in the feature space are prioritized according to the target feature weights, and a candidate feature set is determined;

[0096] S34, during the screening process, simulate the foraging behavior of the kingfisher, conduct a global search on the initial feature set by means of a jump search method, remove redundant features, and retain features that are highly relevant to the grasping task;

[0097]

[0098] in, is the position of the current optimal individual, ∈ is the random disturbance value, represents the update result of feature weights, represents the feature weight position of the i-th individual at the t-th iteration, and s represents the search step length;

[0099] S35, in the feature candidate set, simulating the ambush behavior of the kingfisher, locally optimizing the feature candidate set, and further converging the range of the feature candidate set to the optimal feature set;

[0100] S36. Verify the optimized feature candidate set, generate a final feature data set by matching it with the actual type, shape and position features of the goods, and store it in the goods feature database.

[0101] In this implementation manner, the S4 specifically includes:

[0102] S41, constructing a global layout layer of goods, representing the goods stacking structure as a topological graph model, and generating a global stacking graph based on the geometric center and physical adjacency relationship of the nodes;

[0103] S42. On the basis of the global layout layer, an adjacency relationship layer is constructed to analyze the adjacency relationship of the cargo nodes in the topological structure, calibrate the adjacency weight according to the contact situation between the nodes, and form an adjacency relationship matrix;

[0104] S43. Based on the adjacency weight of the adjacency relationship layer and the physical properties of the goods, the stability value of each goods node is calculated and a stability distribution map is generated:

[0105]

[0106] Among them, H(v i ) represents the cargo node v i The stack height, Represents node v i and v j The adjacency weight, w h and w n is the weight coefficient, S(vi ) represents the stability score of each node, and n represents the total number of cargo nodes in the cargo stacking topology;

[0107] S44. Take the stability distribution map as input, build a grab priority layer, combine the stability value and grab difficulty of the goods, assign grab priorities to the goods nodes, and form a grab priority list:

[0108] P(v i )=S(v i )-α·D(v i );

[0109] Among them, D(v i ) represents node v i The crawling difficulty coefficient, α is the crawling difficulty weight coefficient, P(v i ) is the cargo grabbing priority;

[0110] S45. According to the grab priority list, sort the cargo nodes from high to low priority, generate a grab sequence, and output the sorted grab sequence to the path planning module.

[0111] In this implementation manner, S5 specifically includes:

[0112] S51, based on the cargo feature database and the hierarchical topology modeling results, extract the location, adjacency relationship and grab priority information of the cargo nodes, and initialize the cargo grabbing path as a path set;

[0113] S52, constructing a dynamic environment analysis model based on the grabbing order and the goods adjacency relationship in the initial path set, and modeling the change of the goods stacking structure as a time series topological structure;

[0114] S53, analyzing the real-time changes of the cargo node positions and adjacency relationships according to the time series topology, updating the grabbing order in the path set, and calibrating the adjusted grabbing order of the paths according to the topology changes to generate an intermediate path set;

[0115] S54, using the intermediate path set and the path optimization objective function, minimizing the total path time, and generating the intermediate optimized path set:

[0116]

[0117] Among them, d i represents the grabbing moving distance of the path, v represents the moving speed of the robot arm, τ i represents the grabbing and adjustment time of the path, and n represents the number of goods to be grabbed;

[0118] S55, inputting the intermediate optimized path set into the verification module, and verifying the validity of the intermediate optimized path set and the accuracy of the grabbing task according to the cargo grabbing priority and topological state;

[0119] S56: Take the verified intermediate optimization path set as input, generate the final optimized grasping path sequence, and output it to the grasping control module to execute the task.

[0120] In this implementation manner, S6 specifically includes:

[0121] S61. Based on the cargo feature database and the optimized grasping path sequence, the shape, material and force characteristics of each cargo node are extracted, and the morphological parameter set of the flexible clamp is initialized;

[0122] S62. Design a fixture shape adjustment model, match the shape characteristics of the goods with the fixture shape parameters, and use the dynamic adaptive force field model to calculate the force distribution of the fixture:

[0123]

[0124] Among them, F i is the total force of the clamp when grabbing the object i, w j is the weight of the jth contact point of the fixture, A j is the effective area of ​​the contact point, θ j is the angle between the contact point and the force direction of the fixture, k is the number of fixture contact points;

[0125] S63, dynamically adjusting the contact surface morphological parameters of the fixture according to the total force of the fixture and the material characteristics of the cargo, forming a fixture shape that adapts to the current shape of the cargo, and updating the morphological parameter set;

[0126] S64, using a capacitive stress sensor to monitor the force distribution of each contact point in the grasping process in real time, collecting a force data set, verifying the pressure distribution of the contact point, comparing the monitoring results with the morphological parameter set, identifying abnormal force points and generating a force correction data set;

[0127] S65, taking the generated force correction data set as input, optimizing the pressure distribution of the contact points of the fixture in combination with the dynamic adaptive force field model, redistributing the pressure of each contact point, and generating an optimized morphological parameter set and pressure distribution data;

[0128] S66: Input the optimized morphological parameter set into the flexible control module, execute the cargo grabbing operation, and store the feedback data after the grabbing is completed in the cargo grabbing record.

[0129] In this implementation manner, the S7 specifically includes:

[0130] S71. Based on the optimized grasping path sequence and the initial position of each robotic arm in the grasping task, a multi-robotic arm collaboration model is constructed to decompose the grasping task into a task set;

[0131] S72. Construct a game theory task allocation model for the task set and the robot arm set, and define the objective function as minimizing the task execution time:

[0132]

[0133] Among them, T j represents the subset of tasks assigned to the robot arm, m j represents the robot arm, t i represents the task, d i is the grasping path distance of the task, v j is the moving speed of the robot arm, τ i The operation time of the task;

[0134] S73. Use the graph attention mechanism to dynamically model the task set and the robot arm set, extract the conflict relationship and collaboration potential between tasks, and generate a weight matrix.

[0135] S74. Combine the multi-agent reinforcement learning algorithm, use the weight matrix as input, optimize the task allocation strategy and path planning scheme, and generate a dynamically optimized task allocation set;

[0136] S75, verifying the dynamically adjusted task set, evaluating the rationality of task allocation, the conflict rate of path planning, and the real-time performance of task execution;

[0137] S76. Output the verified task set and path planning strategy to the collaborative control module, start the multi-robotic arm collaborative grasping task in combination with the dynamically adjusted collaborative path, monitor the task completion status in real time, and dynamically adjust task allocation and path planning according to abnormal situations.

[0138] In this implementation manner, S8 specifically includes:

[0139] S81. Use the self-repair optimization algorithm based on causal reasoning to build a causal graph model;

[0140] S82. Identify abnormal event nodes in the crawling task based on the causal graph model, and calculate the causal weight of each abnormal event node:

[0141]

[0142] Among them, δ ij Indicates the causal influence of the abnormal event node, β j and β k represents the severity score of the abnormal event node, wi Represents the causal weight of the abnormal event node, N and N a Represents a set of abnormal event nodes, abnormal event node N a is a subset selected from N, and k is the abnormal event node set N a The index variable of

[0143] S83. Analyze the cause-effect chain of the abnormal event node by combining reverse reasoning, and generate a fault resolution path for each abnormal event node;

[0144] S84, adjusting the grasping parameters by using intervention optimization technology, and updating the grasping parameters in combination with the intervention operation in the fault resolution path;

[0145] S85, based on the updated grasping parameters, adjusting the grasping path set and generating a path plan, wherein the path planning process is combined with dynamic adjustment of the grasping parameters;

[0146] S86. When a fault occurs, use the repaired path planning and grasping parameters to execute the grasping task, monitor the repair effect in real time, and store the fault resolution path and repair data in the task history record.

[0147] Embodiment 1:

[0148] In order to verify the feasibility of the present invention in implementation, the present invention was applied to the loading and unloading of container cargo in an international logistics port. 300 pieces of cargo of different shapes, materials and weights were stored in the container, including rectangular, cylindrical and irregular shaped cargo. Some cargo was completely blocked and the stacking structure was complex. In this scenario, traditional industrial robots are unable to efficiently complete loading and unloading tasks due to insufficient cargo recognition accuracy, low path planning efficiency, frequent conflicts in multi-manipulator collaboration and poor flexible grasping adaptability. Frequent manual intervention seriously affects the operation efficiency.

[0149] In the implementation of the present invention, the container cargo grabbing task is performed by an industrial robot system equipped with a multimodal sensor. The system first starts the multimodal perception module, and collects the point cloud, RGB image and depth information of the cargo through the lidar, 3D camera and depth sensor. For complex scenes such as occlusion and lighting changes, the system uses an adaptive perception fusion algorithm to dynamically adjust the weight of multimodal data, accurately identify the shape, material and position of 300 pieces of cargo, and generate a standardized feature set. Compared with the traditional single sensor method, the present invention improves the accuracy of cargo identification in occluded areas by 15%, and shortens the global cargo identification time by 20%.

[0150] After the identification is completed, the system analyzes the stacking structure of the goods based on the hierarchical topological modeling technology, divides the cargo stack into the global layout layer, the adjacency relationship layer and the grasping priority layer, and allocates tasks in combination with the multi-robot collaboration model driven by game theory. Through the multi-agent reinforcement learning algorithm based on the graph attention mechanism, the system dynamically optimizes the path planning of the robot arm, significantly reducing task conflicts and path interference. For example, when grasping the goods on the 8th layer, robot arm 1 is responsible for grasping the goods, and robot arm 2 synchronously adjusts the lower layer of goods to ensure the stability of the stacking structure, and the path interference rate is reduced from 16% of the traditional method to 4%.

[0151] The flexible gripping module uses shape memory composite material fixtures, combined with a dynamic adaptive force field model to adapt the shape and material of the goods in real time. When grasping fragile items (such as glass products), the system monitors the force distribution during the grasping process through a capacitive stress sensor, controls the grasping force within 4N, and achieves a grasping success rate of 99.5%, which is 14% higher than the traditional clamp method. For irregularly shaped goods, the fixture shape automatically adjusts to fit the surface of the goods, achieving stable grasping of the goods.

[0152] During task execution, the system also monitors and handles abnormal events in real time through causal reasoning self-repair optimization algorithms. For example, when grabbing the 120th item, the robot arm path deviated due to the slipping of the item. The system analyzed the cause of the abnormality through the causal graph model, generated a fault resolution path and completed the repair of the grabbing process in just 1.8 seconds, which is significantly more efficient than the average 10 minutes of manual intervention time of traditional systems.

[0153] Table 1 Comparison of multimodal perception and cargo recognition performance

[0154]

[0155] Table 2 System performance and efficiency comparison

[0156]

[0157] Table 1 shows that the present invention has significantly improved the cargo recognition rate in occluded scenes through multimodal sensing technology, where the recognition accuracy of completely occluded cargo has increased from 70% of the traditional method to 89%, and the overall recognition time has been shortened by 20%. This shows that the present invention has stronger recognition ability and better adaptability in complex scenes.

[0158] Table 2 shows that the path interference rate is reduced from 16% of the traditional method to 4%, and the collaborative efficiency is improved by 75%. The average grasping time for each piece of cargo is shortened from the traditional 60 seconds to 32 seconds, and the efficiency is improved by 46.7%. At the same time, in flexible grasping, the present invention improves the grasping success rate of fragile and irregular cargo by 14.5% and 26.7% respectively, which fully demonstrates the adaptability and stability of the flexible clamp. In terms of fault repair, the present invention reduces the average fault repair time from 600 seconds to 1.8 seconds through the causal reasoning self-repair optimization algorithm, and improves the efficiency by 99.7%. This result highlights the intelligence and efficiency of the system in dynamic fault handling.

[0159] In summary, by analyzing the table data, it can be seen that the improvements in the recognition, planning, grasping and repair aspects of the present invention have greatly improved the performance of industrial robots in container cargo grasping tasks. Whether it is accurate recognition, dynamic path optimization or flexible grasping and fault handling, the present invention has shown significantly superior capabilities to traditional technologies, providing an efficient, safe and intelligent solution for the field of industrial automation.

[0160] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An industrial robot container cargo grabbing control method, characterized in that: The steps include: S1. Collect point cloud data, RGB images and depth information of goods through LiDAR, 3D camera and depth sensor, and capture the features of goods under occlusion and complex lighting scenes by combining biological vision-inspired algorithms to generate an initial goods data set. S2. Adopt an adaptive perception fusion algorithm to fuse the multimodal data in the initial cargo data set, dynamically adjust the weight of each modal data according to the light intensity, occlusion degree and noise level, and generate a unified cargo scene model; S3. Optimize and select the features in the cargo scene model based on the kingfisher optimization algorithm, dynamically search for the optimal feature set in the multi-dimensional feature space of the cargo by simulating the jumping search and static ambush mechanism of the kingfisher when foraging, extract the type, shape, size and spatial position features of the cargo, and store the extracted feature data as a cargo feature database; S4. Using the hierarchical topology modeling method, the global layout layer, adjacency relationship layer and grasping priority layer of the goods are constructed based on the goods feature database. The global layout layer analyzes the stacking structure of the goods, the adjacency relationship layer describes the spatial constraint relationship between the goods, and the grasping priority layer generates a grasping sequence based on the stability of the goods stacking and the grasping difficulty ranking. S5. Combined with the dynamic environment analysis algorithm, the hierarchical topology model is updated in real time. According to the changes in the cargo location, adjacency relationship and grabbing priority, the grabbing sequence and path planning scheme are dynamically adjusted to optimize the cargo grabbing strategy. S6. Design a flexible gripper based on shape memory composite materials, adjust the gripper shape through biomechanical flexibility control theory, calculate the grasping force based on the dynamic adaptive force field model, and use capacitive stress sensors to monitor the force distribution during the grasping process, and adjust the gripper shape and grasping parameters in real time; S7. Through the multi-robot collaboration model driven by game theory, the cargo grabbing task is decomposed into multiple subtasks. The multi-agent reinforcement learning algorithm based on the graph attention mechanism is combined to optimize the robot task allocation and path planning strategy, and dynamically adjust the collaboration path. S8. Use the self-repair optimization algorithm based on causal reasoning to build a causal graph model to analyze abnormal event nodes in the grasping task, combine reverse reasoning to generate a fault resolution path, and adjust the grasping parameters and path planning through intervention optimization technology to complete the repair and adjustment of the grasping process when a fault occurs.

2. The industrial robot container cargo grabbing control method according to claim 1 is characterized in that: The S3 specifically includes: S31, constructing an initial feature space of cargo features, extracting geometric shape, surface texture and size information of the cargo, representing it as a feature set, normalizing the feature set, and generating a standardized feature space; S32, based on the standardized feature space, initialize the feature search space, use the Ban Kingfisher optimization algorithm to globally optimize the feature search space, assign feature weights and set initial parameters for feature screening, including search step size and search range; S33. According to the optimization strategy, the initial feature set of the goods is screened, all feature dimensions in the feature space are prioritized according to the target feature weights, and a candidate feature set is determined; S34, during the screening process, simulate the foraging behavior of the kingfisher, conduct a global search on the initial feature set by means of a jumping search method, eliminate redundant features, and retain features that are highly relevant to the grasping task; in, is the position of the current optimal individual, ∈ is the random disturbance value, represents the update result of feature weights, represents the feature weight position of the i-th individual at the t-th iteration, and s represents the search step length; S35, in the feature candidate set, simulating the ambush behavior of the kingfisher, locally optimizing the feature candidate set, and further converging the range of the feature candidate set to the optimal feature set; S36. Verify the optimized feature candidate set, generate a final feature data set by matching it with the actual type, shape and position features of the goods, and store it in the goods feature database.

3. The industrial robot container cargo grabbing control method according to claim 1, characterized in that: The S4 specifically includes: S41, constructing a global layout layer of goods, representing the goods stacking structure as a topological graph model, and generating a global stacking graph based on the geometric center and physical adjacency relationship of the nodes; S42. On the basis of the global layout layer, an adjacency relationship layer is constructed to analyze the adjacency relationship of the cargo nodes in the topological structure, calibrate the adjacency weight according to the contact situation between the nodes, and form an adjacency relationship matrix; S43. Based on the adjacency weight of the adjacency relationship layer and the physical properties of the goods, the stability value of each goods node is calculated and a stability distribution map is generated: Among them, H(v i ) represents the cargo node v i The stack height, Represents node v i and v j The adjacency weight, w h and w n is the weight coefficient, S(v i ) represents the stability score of each node, and n represents the total number of cargo nodes in the cargo stacking topology; S44. Take the stability distribution map as input, build a grab priority layer, combine the stability value and grab difficulty of the goods, assign grab priorities to the goods nodes, and form a grab priority list: P(v i )=S(v i )-α·D(v i ); Among them, D(v i ) represents node v i The crawling difficulty coefficient, α is the crawling difficulty weight coefficient, P(v i ) is the cargo grabbing priority; S45. According to the grab priority list, sort the cargo nodes from high to low priority, generate a grab sequence, and output the sorted grab sequence to the path planning module.

4. The industrial robot container cargo grabbing control method according to claim 1, characterized in that: The S5 specifically includes: S51, based on the cargo feature database and the hierarchical topology modeling results, extract the location, adjacency relationship and grab priority information of the cargo nodes, and initialize the cargo grabbing path as a path set; S52, constructing a dynamic environment analysis model based on the grabbing order and the goods adjacency relationship in the initial path set, and modeling the change of the goods stacking structure as a time series topological structure; S53, analyzing the real-time changes of the cargo node positions and adjacency relationships according to the time series topology, updating the grabbing order in the path set, and calibrating the adjusted grabbing order of the paths according to the topology changes to generate an intermediate path set; S54, using the intermediate path set and the path optimization objective function, minimizing the total path time, and generating the intermediate optimized path set: Among them, d i represents the grabbing moving distance of the path, v represents the moving speed of the robot arm, τ i represents the grabbing and adjustment time of the path, and n represents the number of goods to be grabbed; S55, inputting the intermediate optimized path set into the verification module, and verifying the validity of the intermediate optimized path set and the accuracy of the grabbing task according to the cargo grabbing priority and topological state; S56: Take the verified intermediate optimization path set as input, generate the final optimized grasping path sequence, and output it to the grasping control module to execute the task.

5. The industrial robot container cargo grabbing control method according to claim 1, characterized in that: The S6 specifically includes: S61. Based on the cargo feature database and the optimized grasping path sequence, the shape, material and force characteristics of each cargo node are extracted, and the morphological parameter set of the flexible clamp is initialized; S62. Design a fixture shape adjustment model, match the shape characteristics of the goods with the fixture shape parameters, and use the dynamic adaptive force field model to calculate the force distribution of the fixture: Among them, F i is the total force of the clamp when grabbing the object i, w j is the weight of the jth contact point of the fixture, A j is the effective area of ​​the contact point, θ j is the angle between the contact point and the force direction of the fixture, k is the number of fixture contact points; S63, dynamically adjusting the contact surface morphological parameters of the fixture according to the total force of the fixture and the material characteristics of the cargo, forming a fixture shape that adapts to the current shape of the cargo, and updating the morphological parameter set; S64, using a capacitive stress sensor to monitor the force distribution of each contact point in the grasping process in real time, collecting a force data set, verifying the pressure distribution of the contact point, comparing the monitoring results with the morphological parameter set, identifying abnormal force points and generating a force correction data set; S65, taking the generated force correction data set as input, optimizing the pressure distribution of the contact points of the fixture in combination with the dynamic adaptive force field model, redistributing the pressure of each contact point, and generating an optimized morphological parameter set and pressure distribution data; S66: Input the optimized morphological parameter set into the flexible control module, execute the cargo grabbing operation, and store the feedback data after the grabbing is completed in the cargo grabbing record.

6. The industrial robot container cargo grabbing control method according to claim 1, characterized in that: The S7 specifically includes: S71. Based on the optimized grasping path sequence and the initial position of each robotic arm in the grasping task, a multi-robotic arm collaboration model is constructed to decompose the grasping task into a task set; S72. Construct a game theory task allocation model for the task set and the robot arm set, and define the objective function as minimizing the task execution time: Among them, T j represents the subset of tasks assigned to the robot arm, m j represents the robot arm, t i represents the task, d i is the grasping path distance of the task, v j is the moving speed of the robot arm, τ i The operation time of the task; S73. Use the graph attention mechanism to dynamically model the task set and the robot arm set, extract the conflict relationship and collaboration potential between tasks, and generate a weight matrix. S74. Combine the multi-agent reinforcement learning algorithm, use the weight matrix as input, optimize the task allocation strategy and path planning scheme, and generate a dynamically optimized task allocation set; S75, verifying the dynamically adjusted task set, evaluating the rationality of task allocation, the conflict rate of path planning, and the real-time performance of task execution; S76. Output the verified task set and path planning strategy to the collaborative control module, start the multi-robotic arm collaborative grasping task in combination with the dynamically adjusted collaborative path, monitor the task completion status in real time, and dynamically adjust task allocation and path planning according to abnormal situations.

7. The industrial robot container cargo grabbing control method according to claim 1, characterized in that: The S8 specifically includes: S81. Use the self-repair optimization algorithm based on causal reasoning to build a causal graph model; S82. Identify abnormal event nodes in the crawling task based on the causal graph model, and calculate the causal weight of each abnormal event node: Among them, δ ij Indicates the causal influence of the abnormal event node, β j and β k represents the severity score of the abnormal event node, w i Represents the causal weight of the abnormal event node, N and N a Represents a set of abnormal event nodes, abnormal event node N a is a subset selected from N, and k is the abnormal event node set N a The index variable of S83. Analyze the cause-effect chain of the abnormal event node by reverse reasoning, and generate a fault resolution path for each abnormal event node; S84, adjusting the grasping parameters by using intervention optimization technology, and updating the grasping parameters in combination with the intervention operation in the fault resolution path; S85, based on the updated grasping parameters, adjusting the grasping path set and generating a path plan, wherein the path planning process is combined with dynamic adjustment of the grasping parameters; S86. When a fault occurs, use the repaired path planning and grasping parameters to execute the grasping task, monitor the repair effect in real time, and store the fault resolution path and repair data in the task history record.

Images