Industrial robot intelligent path planning method and system

Abstract

The invention relates to the technical field of industrial automation, and discloses an intelligent path planning method and system for an industrial robot, and the method comprises the steps: collecting environment initial information in a working region in real time; carrying out preprocessing and fusion processing on the collected environment initial information to obtain target environment data, and constructing a three-dimensional digital working scene model; receiving a production task instruction to analyze the task content to obtain a task analysis result; based on the three-dimensional digital working scene model and the operation target point of the industrial robot in each stage, an improved RRT algorithm is adopted to perform initial path search, and smoothing processing is performed on the initial path obtained through search to obtain a target path; real-time monitoring is carried out through a sensor, when abnormal conditions occur, dynamic adjustment is carried out on the basis of a target path through a D * algorithm, a re-planned target path is obtained, and the abnormal conditions at least comprise the occurrence of a new obstacle, the abnormal station machining progress and the deviation of the workpiece position; the production efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of industrial automation technology, and in particular to an industrial robot intelligent path planning method and system. Background Art

[0002] In modern industrial production, four-station flexible circulation lines and palletizing lines are widely used in the processing, handling, and palletizing of various products. As key execution units in these production lines, the rationality of their path planning directly affects many aspects, including production efficiency, energy consumption, and equipment lifespan. Traditional industrial robot path planning methods are often relatively simple, mostly based on fixed programming modes or preset trajectories, making them difficult to adapt to the complex and changing processing tasks required by four-station flexible circulation lines. For example, when switching between different products, due to differences in the processing sequence, workpiece shape and size, and palletizing layout of each station, a fixed path cannot guarantee that the robot can complete the task efficiently and safely, which can easily lead to collision risks, extend production cycle times, and reduce overall production efficiency. In addition, some existing path planning algorithms have high computational complexity and poor real-time performance when facing large-scale and complex scenarios, and cannot meet the requirements of rapid response in industrial sites. With the advancement of intelligent manufacturing, there is an urgent need for industrial robot control methods and systems that can intelligently adapt to various working conditions and quickly plan optimal paths. Summary of the Invention

[0003] The purpose of the present invention is to solve the above problems and to design an industrial robot intelligent path planning method and system.

[0004] A first aspect of the present invention provides an industrial robot intelligent path planning method, the industrial robot intelligent path planning method comprising the following steps:

[0005] Utilize multiple sensors installed around the four-station flexible circulation line and palletizing line to collect initial environmental information in the work area in real time. The sensors include at least visual sensors, laser radars, and proximity switches.

[0006] Preprocessing and fusing the collected initial environmental information to obtain target environmental data, and constructing a three-dimensional digital work scene model based on the target environmental data;

[0007] Receive production task instructions and parse the task content, determine the processing technology, workpiece flow sequence and final palletizing requirements of each of the four workstations, obtain task analysis results, and set the operation target points of the industrial robot at each stage, where the operation target points at least include the starting point for grabbing the workpiece, the processing position points of each workstation, the transition path points and the palletizing position points;

[0008] Based on the 3D digital work scene model and the target points of the industrial robot at each stage, an improved RRT algorithm is used to search for the initial path. The initial path obtained by the search is smoothed to obtain the target path.

[0009] During the process of an industrial robot executing a planned path, sensors are used to monitor the environment and the robot's own status in real time. When an abnormal situation occurs, the D* algorithm is used to dynamically adjust the target path to obtain a re-planned target path. The abnormal situation includes at least the appearance of new obstacles, abnormal processing progress of the workstation, and deviation of the workpiece position.

[0010] Optionally, in a first implementation of the first aspect of the present invention, preprocessing and fusing the collected initial environmental information to obtain target environmental data, and constructing a three-dimensional digital work scene model based on the target environmental data, includes:

[0011] Obtain initial environmental information. Use the median filter algorithm to remove salt and pepper noise from image data collected by the visual sensor, and use the Kalman filter technology to filter out distance noise from distance data collected by the lidar.

[0012] Based on the industrial robot base coordinate system, the image data coordinate system after noise filtering is converted to a three-dimensional coordinate system in the base coordinate system by obtaining the camera's intrinsic and extrinsic parameters through camera calibration technology.

[0013] The installation angle and height of the LiDAR and its relative position to the robot base are obtained based on the noise-filtered distance data. The polar coordinates of the LiDAR are converted to Cartesian coordinates, and the Cartesian coordinates are aligned with the robot base coordinate system through coordinate translation and rotation matrix operations.

[0014] After unifying the coordinates, the pre-processed initial environmental information is fused at the feature level to obtain the target environmental data;

[0015] The fused target environment data is organized according to spatial position relationships to construct a three-dimensional digital work scene model.

[0016] Optionally, in a second implementation of the first aspect of the present invention, after unifying the coordinates, performing feature-level fusion on the preprocessed initial environment information to obtain target environment data includes:

[0017] The image recognition algorithm is used to extract the workpiece contour and key geometric shape feature points from the pre-processed environmental initial information to obtain image features;

[0018] The three-dimensional geometric structure of obstacles and equipment is identified from the pre-processed initial environmental information through point cloud processing technology to obtain point cloud features;

[0019] The proximity switch data in the pre-processed initial environmental information is used to determine the position of the workpiece on the conveyor belt or the relative distance boundary between the robot end effector and the surrounding objects to obtain the distance feature;

[0020] A data fusion model is constructed, taking image features, point cloud features and distance features as input. The image features are processed using a convolutional neural network to extract the deep features of the image. The point cloud features are processed using a point cloud neural network to obtain the geometric features of the point cloud. The obtained features are concatenated with the distance features and input into a fully connected neural network for fusion and classification to obtain the target environment data.

[0021] Optionally, in a third implementation of the first aspect of the present invention, the initial path search is performed using an improved RRT algorithm based on the three-dimensional digital work scene model and the operation target points of the industrial robot at each stage, and the initial path obtained by the search is smoothed to obtain the target path, including:

[0022] The sampling step size is set to 0.1 meters, the maximum number of iterations is set to 5000, the target deviation threshold is set to 0.05 meters, and the neighboring node search radius is set to 0.2 meters;

[0023] The starting position of the industrial robot is used as the root node of the random tree. According to the three-dimensional digital work scene model, nodes are randomly sampled in the workspace. The first sampled node is connected to the root node to form the initial random tree edge.

[0024] Enter the loop. At the beginning of each loop, first determine whether the maximum number of iterations has been reached. If not, perform random sampling.

[0025] By traversing all nodes of the random tree, calculating the distance from each node to the random node, finding the node with the smallest distance as the nearest node, and extending a new node from the nearest node along the direction of the random node according to the preset sampling step size, during the extension process, the collision detection algorithm is used to determine whether the new node is feasible. If feasible, the new node is added to the random tree, and the nearest node and the new node are connected to form a new tree edge;

[0026] Determine whether the new node can connect to the target point. If it can be connected and the distance between the new node and the target point is less than the target deviation threshold, an initial path from the starting point to the target point is found and the search stops. Otherwise, continue to the next iteration until the maximum number of iterations is reached.

[0027] The initial path from the starting point to the target point is processed with new nodes and adjacent nodes, and the optimization process is performed, and the path generation and path smoothing process are completed to obtain the target path.

[0028] Optionally, in a fourth implementation of the first aspect of the present invention, performing new node neighboring node processing and optimization processing on the initial path from the starting point to the target point, and completing path generation and path smoothing processing to obtain the target path includes:

[0029] Get the initial path, search for adjacent nodes near the newly generated node on the initial path, and determine whether there are adjacent nodes. If so, use the adjacent node as a candidate to replace the parent node of the new node. If not, select the node with the shortest distance to the new node from the random tree.

[0030] Based on the calculated path cost from the neighboring nodes to the starting point, plus the path cost from the new node to each neighboring node, the node that minimizes the path cost of the new node is selected from the neighboring nodes as the best parent node;

[0031] Insert new nodes into the random tree and reconnect the new nodes with the best parent node to optimize the random tree structure;

[0032] When a path that can connect to the target point is found, a path is drawn according to the connection relationship of the random tree. If no path that can connect to the target point is found within the maximum number of iterations, the path is considered to have failed. If a path is successfully drawn, the path search ends.

[0033] According to the nodes on the initial path, a smooth curve is generated using the B-spline curve fitting algorithm. The smooth curve obtained by fitting is used as the new path, the path information is updated, and the target path is obtained.

[0034] Optionally, in a fifth implementation of the first aspect of the present invention, the step of calculating the path cost from the neighboring nodes to the starting point and adding the path cost from the new node to each neighboring node, and selecting a node from the neighboring nodes that minimizes the path cost of the new node as the best parent node, includes:

[0035] Assume that each node n in the random tree has an associated path cost C(n), which represents the path cost from the starting node n. start The path cost to the node n, for the starting node n start The path cost C(n start )=0;

[0036] When building a random tree, each time a new node n is added new And determine its relationship with parent node n parent After the connection, the path cost of the new node C(n new ) can be calculated by the following formula:

[0037] C(n new )=C(n parent )+d(n parent ,n new )

[0038] Among them, d(n parent ,n new ) represents the parent node n parent To the new node n new The Euclidean distance of

[0039] For a neighboring node n near , the path cost to the starting point is C(n near ) is obtained by gradually accumulating the Euclidean distance between each node and its parent node along the edge of the random tree starting from the starting node;

[0040] For the new node n new and a neighboring node n near , the path cost between them is obtained using the Euclidean distance;

[0041] For each neighboring node n near , the path cost C(n near ) and the path cost d(n new ,n near ) are added together to obtain the total path cost C of the new node connecting to the starting point through the neighboring node. total (n near );

[0042] Compare the total path costs C corresponding to all adjacent nodes total (n near ), select the neighboring node with the smallest total path cost as the best parent node of the new node.

[0043] Optionally, in a sixth implementation of the first aspect of the present invention, during the process of the industrial robot executing the planned path, the environment and the robot's own state are monitored in real time by sensors. When an abnormal situation occurs, a D* algorithm is used to dynamically adjust the target path based on the target path to obtain a re-planned target path, including:

[0044] When an abnormal situation is detected, the 3D digital work scene model is updated according to the latest sensor data, the newly appeared obstacle information is added to the 3D digital work scene model, and the position and posture information of the workpiece are updated at the same time;

[0045] Extract key information from the current target path, including node coordinates on the path, connection relationships between nodes, and cost information of each node, and use this information as the initial input of the D* algorithm;

[0046] Based on the updated sensor data, the cost of each node on the path is recalculated and the connection relationship between the nodes is updated;

[0047] Using the D* algorithm, starting from the current robot position, a path search and expansion is performed in the updated 3D digital work scene model. By continuously expanding nodes and comparing the costs of different paths, a new path from the current position to the target position is gradually found. The new searched path is smoothed using the B-spline curve fitting algorithm to obtain the re-planned target path.

[0048] The corresponding motion control instructions are generated according to the re-planned target path and sent to the industrial robot, driving the industrial robot to continue to perform the task according to the re-planned target path.

[0049] The second aspect of the present invention provides an industrial robot intelligent path planning system, which includes an information collection module, a model building module, a task analysis module, a path search module and a path adjustment module, wherein:

[0050] An information acquisition module is used to collect initial environmental information in the working area in real time using a variety of sensors installed around the four-station flexible circulation line and the palletizing line. The sensors include at least visual sensors, laser radars, and proximity switches.

[0051] The model building module is used to pre-process and fuse the collected initial environmental information to obtain target environmental data, and to build a three-dimensional digital work scene model based on the target environmental data;

[0052] The task parsing module is used to receive the production task instruction and parse the task content, determine the processing technology, workpiece flow sequence and final palletizing requirements of each of the four stations, obtain the task parsing results, and set the operation target points of the industrial robot at each stage, where the operation target points include at least the starting point for grasping the workpiece, the processing position points of each station, the transition path points and the palletizing position points;

[0053] The path search module is used to perform initial path search based on the 3D digital work scene model and the operation target points of the industrial robot at each stage, and smooth the initial path obtained to obtain the target path;

[0054] The path adjustment module is used to monitor the environment and the robot's own status in real time through sensors during the process of the industrial robot executing the planned path. When an abnormal situation occurs, the D* algorithm is used to dynamically adjust the target path based on the target path to obtain a re-planned target path. The abnormal situation includes at least the appearance of new obstacles, abnormal workstation processing progress and deviation of the workpiece position.

[0055] Optionally, in a first implementation of the second aspect of the present invention, the model building module includes an acquisition submodule, a filtering submodule, a conversion submodule, a fusion submodule and an organization submodule, wherein:

[0056] The acquisition submodule is used to obtain initial environmental information. It uses the median filter algorithm to remove salt and pepper noise from image data collected by the visual sensor, and uses the Kalman filter technology to filter out distance noise from distance data collected by the lidar.

[0057] The filtering submodule is used to convert the coordinate system of the image data after noise filtering based on the industrial robot base coordinate system. Through camera calibration technology, the intrinsic and extrinsic parameters of the camera are obtained to convert the two-dimensional coordinates in the image plane into three-dimensional coordinates in the base coordinate system.

[0058] The conversion submodule is used to obtain the installation angle, height, and relative position relationship of the lidar to the robot base based on the distance data after noise filtering, convert the lidar polar coordinates into Cartesian coordinates, and align the Cartesian coordinates with the robot base coordinate system through coordinate translation and rotation matrix operations;

[0059] The fusion submodule is used to perform feature-level fusion on the pre-processed initial environment information after unifying the coordinates to obtain the target environment data;

[0060] The organization submodule is used to organize the fused target environment data according to the spatial position relationship and construct a three-dimensional digital work scene model.

[0061] Optionally, in a second implementation of the second aspect of the present invention, the path adjustment module includes a judgment submodule, an extraction submodule, a calculation submodule, a smoothing processing submodule and a task execution submodule, wherein:

[0062] The judgment submodule is used to update the 3D digital work scene model based on the latest sensor data when an abnormal situation is detected, add the newly appeared obstacle information to the 3D digital work scene model, and update the position and posture information of the workpiece at the same time;

[0063] The extraction submodule is used to extract key information from the current target path, including the node coordinates on the path, the connection relationship between the nodes, and the cost information of each node, and use this information as the initial input of the D* algorithm;

[0064] The calculation submodule is used to recalculate the cost of each node on the path based on the updated sensor data and update the connection relationship between the nodes;

[0065] The smoothing submodule is used to use the D* algorithm to search and expand paths in the updated three-dimensional digital work scene model starting from the current robot position. By continuously expanding nodes and comparing the costs of different paths, a new path from the current position to the target position is gradually found. The new searched path is smoothed using the B-spline curve fitting algorithm to obtain the re-planned target path.

[0066] The task execution submodule is used to generate corresponding motion control instructions based on the re-planned target path and send them to the industrial robot to drive the industrial robot to continue to execute the task according to the re-planned target path.

[0067] In the technical solution provided by the present invention, by utilizing a variety of sensors installed around the four-station flexible circulation line and the palletizing line, the initial environmental information in the working area is collected in real time; the collected initial environmental information is pre-processed and fused to obtain target environmental data, and a three-dimensional digital work scene model is constructed based on the target environmental data; the production task instruction is received to parse the task content, the processing technology of each station on the four stations, the workpiece flow sequence and the final palletizing requirements are determined, the task parsing results are obtained, and the operation target points of the industrial robot at each stage are set; based on the three-dimensional digital work scene model and the operation target points of the industrial robot at each stage, an improved RRT algorithm is used to perform initial path search, and the initial path obtained by the search is smoothed. to obtain the target path; in the process of the industrial robot executing the planned path, the environment and the robot's own status are monitored in real time through sensors. When an abnormal situation occurs, the D* algorithm is used to perform dynamic adjustments on the basis of the target path to obtain a re-planned target path, wherein the abnormal situation at least includes the emergence of new obstacles, abnormal processing progress of the workstation and deviation of the workpiece position; the present invention enables the industrial robot to perceive complex and changeable working scenes in real time and accurately, effectively avoid collision risks, improve work safety, significantly shorten the path length, reduce the robot's movement time, improve production efficiency, reduce joint wear, and extend the service life of the equipment, greatly improving the overall intelligence level and flexible production capacity of the four-station flexible circulation line and palletizing line. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Various other advantages and benefits will become apparent to those skilled in the art by reading the following detailed description of the preferred embodiment.The accompanying drawings are only for the purpose of illustrating the preferred embodiment and are not to be considered as limiting the present invention.

[0069] Figure 1 A schematic diagram of an industrial robot intelligent path planning method provided by an embodiment of the present invention;

[0070] Figure 2 This is a schematic diagram of the structure of an industrial robot intelligent path planning system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0071] The terms "first," "second," "third," "fourth," and the like (if any) in the description and claims of the present invention and in the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, apparatus, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to these processes, methods, products, or apparatus.

[0072] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 A schematic diagram of an intelligent path planning method for an industrial robot provided by an embodiment of the present invention, the method specifically includes the following steps:

[0073] Step 101: Utilize various sensors installed around the four-station flexible circulation line and the palletizing line to collect initial environmental information in the working area in real time;

[0074] In this embodiment, the sensor includes at least a visual sensor, a laser radar, and a proximity switch;

[0075] Step 102: pre-process and fuse the collected initial environmental information to obtain target environmental data, and construct a three-dimensional digital work scene model based on the target environmental data;

[0076] In this embodiment, initial environmental information is obtained, and a median filtering algorithm is used to remove salt and pepper noise from image data collected by the visual sensor, and a Kalman filtering technique is used to filter out distance noise from distance data collected by the laser radar. With the industrial robot base coordinate system as a reference, the coordinate system of the image data after noise filtering is converted. Through the camera calibration technology, the intrinsic and extrinsic parameters of the camera are obtained to convert the two-dimensional coordinates in the image plane into three-dimensional coordinates in the base coordinate system. Based on the distance data after noise filtering, the installation angle, height and relative position relationship with the robot base of the laser radar are obtained, the polar coordinates of the laser radar are converted into Cartesian coordinates, and the Cartesian coordinates are aligned with the robot base coordinate system through coordinate translation and rotation matrix operations. After unifying the coordinates, the pre-processed initial environmental information is subjected to feature-level fusion to obtain target environmental data. The target environmental data obtained after fusion are organized according to the spatial position relationship to construct a three-dimensional digital work scene model.

[0077] In this embodiment, an image recognition algorithm is used to extract the workpiece contour and key geometric shape feature points from the preprocessed environmental initial information to obtain image features; point cloud processing technology is used to identify the three-dimensional geometric structure of obstacles and equipment from the preprocessed environmental initial information to obtain point cloud features; the position state of the workpiece on the conveyor belt or the relative distance boundary between the robot end effector and the surrounding objects is determined from the data of the proximity switch in the preprocessed environmental initial information to obtain distance features; a data fusion model is constructed, and image features, point cloud features and distance features are used as input. A convolutional neural network is used to process the image features to extract deep features of the image, and a point cloud neural network is used to process the point cloud features to obtain the geometric features of the point cloud. The obtained features and distance features are spliced and input into a fully connected neural network for fusion and classification to obtain target environmental data.

[0078] Step 103: Receive the production task instruction and parse the task content, determine the processing technology, workpiece flow sequence and final palletizing requirements of each of the four workstations, obtain the task analysis results, and set the operation target points of the industrial robot at each stage;

[0079] In this embodiment, the operation target points include at least the starting point for grabbing the workpiece, the processing position points of each workstation, the transition path points and the stacking position points;

[0080] Step 104: Based on the three-dimensional digital work scene model and the operation target points of the industrial robot at each stage, an improved RRT algorithm is used to perform an initial path search, and the searched initial path is smoothed to obtain a target path;

[0081] In this embodiment, the sampling step size is set to 0.1 meters, the maximum number of iterations is set to 5000 times, the target deviation threshold is set to 0.05 meters, and the neighboring node search radius is set to 0.2 meters; the starting position of the industrial robot is used as the root node of the random tree, and nodes are randomly sampled in the workspace according to the three-dimensional digital work scene model, and the first sampled node is connected to the root node to form an initial random tree edge; enter the loop, and at the beginning of each loop, first determine whether the maximum number of iterations has been reached. If not, random sampling is performed; by traversing all nodes of the random tree, the distance from each node to the random node is calculated, and the node with the smallest distance is found as the nearest node, and the nearest node is found from the nearest node along the random node. In the direction of the random node, a new node is extended according to the preset sampling step size. During the extension process, the collision detection algorithm is used to determine whether the new node is feasible. If feasible, the new node is added to the random tree, and the nearest node and the new node are connected to form a new tree edge; determine whether the new node can be connected to the target point. If it can be connected and the distance between the new node and the target point is less than the target deviation threshold, an initial path from the starting point to the target point is found and the search is stopped. Otherwise, the next iteration is continued until the maximum number of iterations is reached; the initial path from the starting point to the target point is processed and optimized with new node neighbors, and path generation and path smoothing are completed to obtain the target path.

[0082] In this embodiment, an initial path is obtained, and adjacent nodes are searched near the newly generated node on the initial path to determine whether adjacent nodes exist. If so, the adjacent nodes are used as candidates to replace the parent node of the new node. If not, the node with the shortest distance to the new node is selected from the random tree. Based on the calculated path cost from the adjacent nodes to the starting point and the path cost from the new node to each adjacent node, the node that minimizes the path cost of the new node is selected from the adjacent nodes as the best parent node. The new node is inserted into the random tree, and the best parent node is reconnected to the new node to optimize the random tree structure. When it is determined that a path that can connect to the target point is found, a path is drawn according to the connection relationship of the random tree. If a path that can connect to the target point is not found within the maximum number of iterations, the path is determined to have failed. If the path is successfully drawn, the path search is terminated. A smooth curve is generated based on the nodes on the initial path using a B-spline curve fitting algorithm. The smooth curve obtained by fitting is used as the new path, the path information is updated, and the target path is obtained.

[0083] In this embodiment, it is assumed that each node n in the random tree has an associated path cost C(n), which represents the path cost from the starting node n. start The path cost to the node n, for the starting node n start The path cost C(n start )=0;

[0084] When building a random tree, each time a new node n is addednew And determine its relationship with parent node n parent After the connection, the path cost of the new node C(n new ) can be calculated using the following formula:

[0085] C(n new )=C(n parent )+d(n parent ,n new )

[0086] Among them, d(n parent ,n new ) represents the parent node n parent To the new node n new The Euclidean distance of

[0087] For a neighboring node n near , the path cost to the starting point is C(n near ) is obtained by gradually accumulating the Euclidean distance between each node and its parent node along the edge of the random tree starting from the starting node;

[0088] For the new node n near ) and a neighboring node n near ), the path cost between them is obtained using the Euclidean distance;

[0089] For each neighboring node n near , the path cost C(n near ) and the path cost d(n new ,n near ) are added together to get the total path cost C of the new node connecting to the starting point through the neighboring node. total (n near );

[0090] Compare the total path costs C corresponding to all adjacent nodes total (n near ), select the neighboring node with the smallest total path cost as the best parent node of the new node.

[0091] Step 105: While the industrial robot is executing the planned path, the environment and the robot's own status are monitored in real time through sensors. When an abnormal situation occurs, the D* algorithm is used to dynamically adjust the target path to obtain a re-planned target path.

[0092] In this embodiment, the abnormal situation includes at least the appearance of a new obstacle, abnormal processing progress of the workstation, and deviation in the position of the workpiece.

[0093] In this embodiment, when an abnormal situation is determined to have occurred, the three-dimensional digital work scene model is updated based on the latest sensor data, the newly appeared obstacle information is added to the three-dimensional digital work scene model, and the position information and posture information of the workpiece are updated; key information is extracted from the current target path, including the node coordinates on the path, the connection relationship between the nodes, and the cost information of each node, and this information is used as the initial input of the D* algorithm; based on the updated sensor data, the cost of each node on the path is recalculated, and the connection relationship between the nodes is updated; the D* algorithm is used to search and expand the path in the updated three-dimensional digital work scene model starting from the current robot position, by continuously expanding the nodes and comparing the costs of different paths, a new path from the current position to the target position is gradually found, and the new path obtained by the search is smoothed using a B-spline curve fitting algorithm to obtain a re-planned target path; corresponding motion control instructions are generated based on the re-planned target path and sent to the industrial robot, driving the industrial robot to continue to perform the task according to the re-planned target path.

[0094] See also Figure 2 , a structural diagram of an industrial robot intelligent path planning system provided by an embodiment of the present invention, the system includes an information collection module, a model construction module, a task analysis module, a path search module and a path adjustment module, wherein,

[0095] The information collection module 201 is used to collect the initial environmental information of the working area in real time using a variety of sensors installed around the four-station flexible circulation line and the palletizing line, where the sensors include at least a visual sensor, a laser radar, and a proximity switch;

[0096] The model building module 202 is used to pre-process and fuse the collected initial environmental information to obtain target environmental data, and to build a three-dimensional digital work scene model based on the target environmental data;

[0097] The task parsing module 203 is used to receive the production task instruction, parse the task content, determine the processing technology, workpiece flow sequence and final palletizing requirements of each of the four workstations, obtain the task parsing results, and set the operation target points of the industrial robot at each stage, where the operation target points include at least the starting point for grasping the workpiece, the processing position points of each workstation, the transition path points and the palletizing position points;

[0098] The path search module 204 is configured to perform an initial path search using an improved RRT algorithm based on the three-dimensional digital work scene model and the operation target points of the industrial robot at each stage, and to smooth the initial path obtained to obtain a target path.

[0099] The path adjustment module 205 is used to monitor the environment and the robot's own status in real time through sensors during the process of the industrial robot executing the planned path. When an abnormal situation occurs, the D* algorithm is used to dynamically adjust the target path to obtain a re-planned target path. The abnormal situation includes at least the appearance of new obstacles, abnormal processing progress of the workstation, and deviation of the workpiece position.

[0100] In this embodiment, the model building module includes an acquisition submodule, a filtering submodule, a conversion submodule, a fusion submodule and an organization submodule, wherein:

[0101] The acquisition submodule is used to obtain initial environmental information. It uses the median filter algorithm to remove salt and pepper noise from image data collected by the visual sensor, and uses the Kalman filter technology to filter out distance noise from distance data collected by the lidar.

[0102] The filtering submodule is used to convert the coordinate system of the image data after noise filtering based on the industrial robot base coordinate system. Through camera calibration technology, the intrinsic and extrinsic parameters of the camera are obtained to convert the two-dimensional coordinates in the image plane into three-dimensional coordinates in the base coordinate system.

[0103] The conversion submodule is used to obtain the installation angle, height, and relative position relationship of the lidar to the robot base based on the distance data after noise filtering, convert the lidar polar coordinates into Cartesian coordinates, and align the Cartesian coordinates with the robot base coordinate system through coordinate translation and rotation matrix operations;

[0104] The fusion submodule is used to perform feature-level fusion on the pre-processed initial environment information after unifying the coordinates to obtain the target environment data;

[0105] The organization submodule is used to organize the fused target environment data according to the spatial position relationship and construct a three-dimensional digital work scene model.

[0106] In this embodiment, the path adjustment module includes a judgment submodule, an extraction submodule, a calculation submodule, a smoothing processing submodule and a task execution submodule, wherein:

[0107] The judgment submodule is used to update the 3D digital work scene model based on the latest sensor data when an abnormal situation is detected, add the newly appeared obstacle information to the 3D digital work scene model, and update the position and posture information of the workpiece at the same time;

[0108] The extraction submodule is used to extract key information from the current target path, including the node coordinates on the path, the connection relationship between the nodes, and the cost information of each node, and use this information as the initial input of the D* algorithm;

[0109] The calculation submodule is used to recalculate the cost of each node on the path based on the updated sensor data and update the connection relationship between the nodes;

[0110] The smoothing submodule is used to use the D* algorithm to search and expand paths in the updated three-dimensional digital work scene model starting from the current robot position. By continuously expanding nodes and comparing the costs of different paths, a new path from the current position to the target position is gradually found. The new searched path is smoothed using the B-spline curve fitting algorithm to obtain the re-planned target path.

[0111] The task execution submodule is used to generate corresponding motion control instructions based on the re-planned target path and send them to the industrial robot to drive the industrial robot to continue to execute the task according to the re-planned target path.

[0112] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. An industrial robot intelligent path planning method, characterized in that: The industrial robot intelligent path planning method comprises the following steps: Utilize multiple sensors installed around the four-station flexible circulation line and palletizing line to collect initial environmental information in the work area in real time. The sensors include at least visual sensors, laser radars, and proximity switches. Preprocessing and fusing the collected initial environmental information to obtain target environmental data, and constructing a three-dimensional digital work scene model based on the target environmental data; Receive production task instructions and parse the task content, determine the processing technology, workpiece flow sequence and final palletizing requirements of each of the four workstations, obtain task analysis results, and set the operation target points of the industrial robot at each stage, where the operation target points at least include the starting point for grabbing the workpiece, the processing position points of each workstation, the transition path points and the palletizing position points; Based on the 3D digital work scene model and the target points of the industrial robot at each stage, an improved RRT algorithm is used to search for the initial path. The initial path obtained by the search is smoothed to obtain the target path. During the process of an industrial robot executing a planned path, sensors are used to monitor the environment and the robot's own status in real time. When an abnormal situation occurs, the D* algorithm is used to dynamically adjust the target path to obtain a re-planned target path. The abnormal situation includes at least the appearance of new obstacles, abnormal processing progress of the workstation, and deviation of the workpiece position.

2. The intelligent path planning method for an industrial robot according to claim 1, wherein: The collected initial environmental information is pre-processed and fused to obtain target environmental data, and a three-dimensional digital work scene model is constructed based on the target environmental data, including: Obtain initial environmental information. Use the median filter algorithm to remove salt and pepper noise from image data collected by the visual sensor, and use the Kalman filter technology to filter out distance noise from distance data collected by the lidar. Based on the industrial robot base coordinate system, the image data coordinate system after noise filtering is converted to a three-dimensional coordinate system in the base coordinate system by obtaining the camera's intrinsic and extrinsic parameters through camera calibration technology. The installation angle and height of the LiDAR and its relative position to the robot base are obtained based on the noise-filtered distance data. The polar coordinates of the LiDAR are converted to Cartesian coordinates, and the Cartesian coordinates are aligned with the robot base coordinate system through coordinate translation and rotation matrix operations. After unifying the coordinates, the pre-processed initial environmental information is fused at the feature level to obtain the target environmental data; The fused target environment data is organized according to spatial position relationships to construct a three-dimensional digital work scene model.

3. The intelligent path planning method for an industrial robot according to claim 2, wherein: After the coordinates are unified, the pre-processed initial environment information is subjected to feature-level fusion to obtain target environment data, including: The image recognition algorithm is used to extract the workpiece contour and key geometric shape feature points from the pre-processed environmental initial information to obtain image features; The three-dimensional geometric structure of obstacles and equipment is identified from the pre-processed initial environmental information through point cloud processing technology to obtain point cloud features; The proximity switch data in the pre-processed initial environmental information is used to determine the position of the workpiece on the conveyor belt or the relative distance boundary between the robot end effector and the surrounding objects to obtain the distance feature; A data fusion model is constructed, taking image features, point cloud features and distance features as input. The image features are processed using a convolutional neural network to extract the deep features of the image. The point cloud features are processed using a point cloud neural network to obtain the geometric features of the point cloud. The obtained features are concatenated with the distance features and input into a fully connected neural network for fusion and classification to obtain the target environment data.

4. The intelligent path planning method for an industrial robot according to claim 1, wherein: The improved RRT algorithm is used to perform initial path search based on the three-dimensional digital work scene model and the operation target points of the industrial robot at each stage, and the initial path obtained by the search is smoothed to obtain the target path, including: The sampling step size is set to 0.1 meters, the maximum number of iterations is set to 5000, the target deviation threshold is set to 0.05 meters, and the neighboring node search radius is set to 0.2 meters; The starting position of the industrial robot is used as the root node of the random tree. According to the three-dimensional digital work scene model, nodes are randomly sampled in the workspace. The first sampled node is connected to the root node to form the initial random tree edge. Enter the loop. At the beginning of each loop, first determine whether the maximum number of iterations has been reached. If not, perform random sampling. By traversing all nodes of the random tree, calculating the distance from each node to the random node, finding the node with the smallest distance as the nearest node, and extending a new node from the nearest node along the direction of the random node according to the preset sampling step size, during the extension process, the collision detection algorithm is used to determine whether the new node is feasible. If feasible, the new node is added to the random tree, and the nearest node and the new node are connected to form a new tree edge; Determine whether the new node can connect to the target point. If it can be connected and the distance between the new node and the target point is less than the target deviation threshold, an initial path from the starting point to the target point is found and the search stops. Otherwise, continue to the next iteration until the maximum number of iterations is reached. The initial path from the starting point to the target point is processed with new nodes and adjacent nodes, and the optimization process is performed, and the path generation and path smoothing process are completed to obtain the target path.

5. The intelligent path planning method for an industrial robot according to claim 4, characterized in that: The initial path from the starting point to the target point is processed with new nodes and adjacent nodes, and optimized, and path generation and path smoothing are completed to obtain the target path, including: Get the initial path, search for adjacent nodes near the newly generated node on the initial path, and determine whether there are adjacent nodes. If so, use the adjacent node as a candidate to replace the parent node of the new node. If not, select the node with the shortest distance to the new node from the random tree. Based on the calculated path cost from the neighboring nodes to the starting point, plus the path cost from the new node to each neighboring node, the node that minimizes the path cost of the new node is selected from the neighboring nodes as the best parent node; Insert new nodes into the random tree and reconnect the new nodes with the best parent node to optimize the random tree structure; When a path that can connect to the target point is found, a path is drawn according to the connection relationship of the random tree. If no path that can connect to the target point is found within the maximum number of iterations, the path is considered to have failed. If a path is successfully drawn, the path search ends. According to the nodes on the initial path, a smooth curve is generated using the B-spline curve fitting algorithm. The smooth curve obtained by fitting is used as the new path, the path information is updated, and the target path is obtained.

6. The intelligent path planning method for an industrial robot according to claim 5, characterized in that: The method of calculating the path cost from the neighboring nodes to the starting point and adding the path cost from the new node to each neighboring node, and selecting a node from the neighboring nodes that minimizes the path cost of the new node as the best parent node, includes: Assume that each node n in the random tree has an associated path cost C(n), which represents the path cost from the starting node n. start The path cost to the node n, for the starting node n start The path cost C(n start )=0; When building a random tree, each time a new node n is added new And determine its relationship with parent node n parent After the connection, the path cost of the new node C(n new ) can be calculated using the following formula: C(n new )=C(n parent )+d(n parent ),n new ) Among them, d(n parent ,n new ) represents the parent node n parent To the new node n new The Euclidean distance of For a neighboring node n near , the path cost to the starting point is C(n ear ) is obtained by gradually accumulating the Euclidean distance between each node and its parent node along the edge of the random tree starting from the starting node; For the new node n new and a neighboring node n near , the path cost between them is obtained using the Euclidean distance; For each neighboring node n near , the path cost C(n near ) and the path cost d( new ,n near ) are added together to obtain the total path cost C of the new node connecting to the starting point through the neighboring node. total (n near ); Compare the total path costs C corresponding to all adjacent nodes total (n near ), select the neighboring node with the smallest total path cost as the best parent node of the new node.

7. The intelligent path planning method for an industrial robot according to claim 1, wherein: During the process of executing the planned path, the industrial robot monitors the environment and the robot's own status in real time through sensors. When an abnormal situation occurs, the D* algorithm is used to dynamically adjust the target path based on the target path to obtain a re-planned target path, including: When an abnormal situation is detected, the 3D digital work scene model is updated according to the latest sensor data, the newly appeared obstacle information is added to the 3D digital work scene model, and the position and posture information of the workpiece are updated at the same time; Extract key information from the current target path, including node coordinates on the path, connection relationships between nodes, and cost information of each node, and use this information as the initial input of the D* algorithm; Based on the updated sensor data, the cost of each node on the path is recalculated and the connection relationship between the nodes is updated; Using the D* algorithm, starting from the current robot position, a path search and expansion is performed in the updated 3D digital work scene model. By continuously expanding nodes and comparing the costs of different paths, a new path from the current position to the target position is gradually found. The new searched path is smoothed using the B-spline curve fitting algorithm to obtain the re-planned target path. The corresponding motion control instructions are generated according to the re-planned target path and sent to the industrial robot, driving the industrial robot to continue to perform the task according to the re-planned target path.

8. An industrial robot intelligent path planning system, characterized in that: The industrial robot intelligent path planning system includes an information collection module, a model building module, a task analysis module, a path search module and a path adjustment module, wherein: An information acquisition module is used to collect initial environmental information in the working area in real time using a variety of sensors installed around the four-station flexible circulation line and the palletizing line. The sensors include at least visual sensors, laser radars, and proximity switches. The model building module is used to pre-process and fuse the collected initial environmental information to obtain target environmental data, and to build a three-dimensional digital work scene model based on the target environmental data; The task parsing module is used to receive the production task instruction and parse the task content, determine the processing technology, workpiece flow sequence and final palletizing requirements of each of the four stations, obtain the task parsing results, and set the operation target points of the industrial robot at each stage, where the operation target points include at least the starting point for grasping the workpiece, the processing position points of each station, the transition path points and the palletizing position points; The path search module is used to perform initial path search based on the 3D digital work scene model and the operation target points of the industrial robot at each stage, and smooth the initial path obtained to obtain the target path; The path adjustment module is used to monitor the environment and the robot's own status in real time through sensors during the process of the industrial robot executing the planned path. When an abnormal situation occurs, the D* algorithm is used to dynamically adjust the target path based on the target path to obtain a re-planned target path. The abnormal situation includes at least the appearance of new obstacles, abnormal workstation processing progress and deviation of the workpiece position.

9. The intelligent path planning system for an industrial robot according to claim 8, characterized in that: The model building module includes an acquisition submodule, a filtering submodule, a conversion submodule, a fusion submodule and an organization submodule, wherein: The acquisition submodule is used to obtain initial environmental information. It uses the median filter algorithm to remove salt and pepper noise from image data collected by the visual sensor, and uses the Kalman filter technology to filter out distance noise from distance data collected by the lidar. The filtering submodule is used to convert the coordinate system of the image data after noise filtering based on the industrial robot base coordinate system. Through camera calibration technology, the intrinsic and extrinsic parameters of the camera are obtained to convert the two-dimensional coordinates in the image plane into three-dimensional coordinates in the base coordinate system. The conversion submodule is used to obtain the installation angle, height, and relative position relationship of the lidar to the robot base based on the distance data after noise filtering, convert the lidar polar coordinates into Cartesian coordinates, and align the Cartesian coordinates with the robot base coordinate system through coordinate translation and rotation matrix operations; The fusion submodule is used to perform feature-level fusion on the pre-processed initial environment information after unifying the coordinates to obtain the target environment data; The organization submodule is used to organize the fused target environment data according to the spatial position relationship and construct a three-dimensional digital work scene model.

10. The intelligent path planning system for an industrial robot according to claim 8, wherein: The path adjustment module includes a judgment submodule, an extraction submodule, a calculation submodule, a smoothing submodule and a task execution submodule, wherein: The judgment submodule is used to update the 3D digital work scene model based on the latest sensor data when an abnormal situation is detected, add the newly appeared obstacle information to the 3D digital work scene model, and update the position and posture information of the workpiece at the same time; The extraction submodule is used to extract key information from the current target path, including the node coordinates on the path, the connection relationship between the nodes, and the cost information of each node, and use this information as the initial input of the D* algorithm; The calculation submodule is used to recalculate the cost of each node on the path based on the updated sensor data and update the connection relationship between the nodes; The smoothing submodule is used to use the D* algorithm to search and expand paths in the updated three-dimensional digital work scene model starting from the current robot position. By continuously expanding nodes and comparing the costs of different paths, a new path from the current position to the target position is gradually found. The new searched path is smoothed using the B-spline curve fitting algorithm to obtain the re-planned target path. The task execution submodule is used to generate corresponding motion control instructions based on the re-planned target path and send them to the industrial robot to drive the industrial robot to continue to execute the task according to the re-planned target path.

Images