Motion path planning system and method for crawler crane

Abstract

The invention discloses a motion path planning system and method of a crawler crane, and belongs to the field of intelligent driving of crawler cranes. Comprising a path planning module loaded into a workstation, a data acquisition module and an unmanned aerial vehicle. The method is easy to operate, the labor intensity of workers is effectively reduced, and the workers only need to use the unmanned aerial vehicle to shoot the environment, import automatic drawing software to generate a three-dimensional model and read the initial pose of the crane according to the GPS, so that the optimal path of the crane can be automatically calculated. And through a data acquisition module, an RTK module of the unmanned aerial vehicle, environment modeling software and the like, it is ensured that the generation result of the three-dimensional model is rapid and accurate. And based on the developed path algorithm, the optimal path is automatically generated, so that the efficiency, safety and accuracy of the hoisting process are improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent driving of crawler cranes, and in particular to a motion path planning system and method for crawler cranes. Background Art

[0002] At present, global industrialization is in a rapid development stage, and the lifting machinery industry has also developed well. The traditional lifting machinery industry is developing towards intelligence and informatization, and various manufacturing technologies are constantly being updated. People are beginning to pay more attention to the work efficiency, functional diversity and safety of cranes.

[0003] In traditional crawler crane hoisting operations, construction workers usually guide the crane driver to determine the hoisting path based on site conditions and operation objectives. The driver relies on experience to continuously adjust the crane's angle and posture to ultimately complete the hoisting task. However, during this process, the driver needs to constantly observe the surrounding environment and adjust the hanging position of the heavy object back and forth to find a path that can reach the target location. This method relies too much on the operator's experience, makes it difficult to accurately reach the target location, and is inefficient. Traditional hoisting solutions can no longer meet the increasingly stringent hoisting operation standards. Therefore, this paper proposes a more intelligent crawler crane autonomous hoisting method. By combining autonomous navigation technology with crawler crane hoisting operations, it can pre-design an optimized path before hoisting, build working environment information, and accurately locate installation points, thereby improving hoisting efficiency and safety, and realizing the automation and unmanned operation of hoisting operations. Summary of the Invention

[0004] The purpose of the present invention is to realize unmanned operation of a crane and propose an automatic path planning system and method for a crawler crane, thereby improving the work efficiency of the lifting operation.

[0005] A motion path planning system for a crawler crane includes a path planning module loaded into a workstation, a data acquisition module, and a drone.

[0006] The data acquisition module is used to collect the rotation center, the luffing center, the starting point of the hoisted object and the placement point of the hoisted object of the crane and transmit them to the workstation.

[0007] The drone is used to collect on-site environmental information and transmit it to the workstation.

[0008] The path planning module is used to generate the optimal path, which is achieved through the UAV-based environment reconstruction method and the improved RRT algorithm. The steps are as follows:

[0009] Step 1: Build a rigid body model of the crawler crane

[0010] According to the structure and working principle of crawler crane and the hoisted objects, the crane hoisting system is abstracted into five rigid bodies, including the turntable for getting off and getting on as a rectangular block; the boom and the hoisted objects are abstracted as cylindrical blocks; the wire rope is abstracted as a rigid straight line that can move up and down (such as Figure 1 shown).

[0011] Step 2: Construct the kinematic model of crawler crane

[0012] The kinematic model is established based on the rigid body model of the crawler crane. The motion relationship diagram is as follows: Figure 2 As shown. The crane's lowering local coordinate system {0} is placed at the center of the lowering bottom, the turntable local coordinate system {1} is set at the intersection of the upper part of the turntable and the rotation center, the main boom's luffing local coordinate system {2} is set at the luffing center, and the hook local coordinate system {6} is placed at the bottom center of the hoisted object. In actual crane structures, the wire rope is not connected at the center of the main boom head, but is connected by a pulley set offset from the main boom's central axis. Therefore, when establishing the kinematic model, the boom head local coordinate system {3} is set at the boom head axis, and the jib local coordinate system {4} is set at the jib head axis. If the crane only has a main boom, the coordinate system is set at the boom head axis, and the boom head's wire rope local coordinate system {5} is set at the center of the pulley set for lowering the wire rope. These three coordinate systems are used to simulate the actual connection position of the wire rope. Therefore, the position of the crane hoisting system can be expressed by P{θ1,θ2,θ3,θ4,θ5,d6}; where θ1 is the turntable rotation angle, θ2 is the boom luffing angle, θ3 is the angle between the boom extension line and the horizontal direction, θ4 is the angle between the boom normal direction and the horizontal direction, θ5 is the angle between the negative direction of the boom normal line and the horizontal direction, and d6 is the wire rope extension distance;

[0013] Let d1 be the total height of the crawler crane turntable and the lower car, a2 be the distance between the upper car's rotation axis and the boom's luffing axis, a3 be the main boom length, a5 be the pulley block offset distance, and d6 be the wire rope extension distance.

[0014] Depend on Figure 2 The kinematic model expresses the position of the end of the suspended object as:

[0015]

[0016] in, T1 is the pose transformation matrix from the local coordinate system {0} to the local coordinate system {1}. 2 is the pose transformation matrix from the local coordinate system {1} to the local coordinate system {2}, is the pose transformation matrix from the local coordinate system {2} to the local coordinate system {3}, is the pose transformation matrix from the local coordinate system {3} to the local coordinate system {4}, is the pose transformation matrix from the local coordinate system {4} to the local coordinate system {5}, is the pose transformation matrix from the local coordinate system {5} to the local coordinate system {6}, is the pose transformation matrix from the local coordinate system {0} to the local coordinate system {6}.

[0017] During the crane hoisting process, the wire rope always remains vertically downward, so the algebraic relationship between θ2, θ3, θ4 and θ5 can be obtained:

[0018]

[0019] Represent θ3, θ4, and θ5 by θ2, and substitute each transformation matrix into the formula You can get:

[0020]

[0021] When the coordinates of the end of the suspended object are known (P x ,P y ,P z ) can inversely calculate the data of each joint angle variable. Therefore, the posture transformation matrix of the suspended object can be set as Then we can inversely obtain θ1, θ2, and d6:

[0022]

[0023] Where t represents

[0024]

[0025] According to the obtained forward and inverse kinematics formulas, the position parameters of the crane can be obtained by inversely calculating the movement of the hoisted object.

[0026] Step 3: Based on the environment model obtained by UAV scanning and the kinematic relationship of the crane, the crawler crane path planning is implemented based on the improved RRT. This includes the following steps:

[0027] S1: A drone takes aerial photos of the environment, collecting a sufficient number of photos using a visible light camera. These photos are then fed into the environmental modeling software, which automatically generates a 3D model based on the captured photos. Obstacles in the 3D model's hoisting environment are enclosed by regular geometric shapes using a bounding box algorithm. Obstacles are expanded based on the size of the hoisted object, allowing the object to be represented as a point during planning. The bounding box algorithm extracts the 3D spatial information into a 2D grid array.

[0028] S2: Start path planning, initialize the search tree, set the information contained in each node in the search tree to be the three action numbers from the previous crane posture to the current crane posture, and record the previous node of the node, that is, the parent node. The action number refers to the numbering of the three actions of rotation, luffing, and lifting. Set the root node of the search tree as the starting point of lifting, and set the initial action of the root node according to the configuration information.

[0029] S3: Select a random point P in the planning space.

[0030] S4: Traverse all nodes in the search tree and select a node as the parent node of the random point. The selection rule is that the direction from the parent node to the random point P is close to the direction of the target point and the distance is the shortest.

[0031] S5: Based on the crane motion information contained in the parent node selected in step S4, select the action to reach the random point and set it to m. The selection rule is determined by K(m). K(m) = |V(m)·Vp| + F(m). Each action is assumed to perform only one motion with one motion number. Action m has three motion numbers, so three K(m) values can be generated. The action with the largest K(m) value is selected as the extended action. V(m) is the direction vector corresponding to the possible extended action m, Vp is the direction vector from the parent node to the random point P, and F(m) is the action selection correction parameter used to reduce the number of action switches during path generation and speed up the search. When action m is the same as the parent node action, F(m) = C1*Rand, where C1 is the preset motion retention parameter and Rand is a random number between 0 and 1. When action m is different from the parent node action, if the corresponding posture of the parent node action is already the same as the landing point, then F(m) = C1*Rand; otherwise, F(m) = 0.

[0032] S6: Follow the action selected in step S5 and intercept the next node with a step length of t. At this time, the conversion formula of θ1, θ2, and d6 calculated in step 2 can be substituted into the coordinates of the end of the suspended object at this node (P x ,P y ,P z ) and then calculate the three action number information of the crane posture of the node, and the step length t is the set movement distance parameter.

[0033] S7: Determine whether the new node is valid. The judgment conditions are that the corresponding crane posture is valid and there is no collision between the new node and the parent node. The valid crane posture means that the entire vehicle does not collide when the crane moves and each action number is within the set threshold range.

[0034] S8: Determine whether the end point has been reached. The judgment condition is that the distance from the node to the end point is less than a threshold. If the end point has been reached, the loop is exited and the process enters S10.

[0035] S9: Repeat steps S3-S8 until the number of repetitions reaches the set value or exit at step S8.

[0036] S10: A path is obtained based on the search tree nodes. The path includes all node positions in the search tree and action number information of each node.

[0037] S11: Repeat steps S2-S10 multiple times to obtain multiple paths. Since each search path is independent of each other, parallelization can be used to accelerate the process.

[0038] S12: Evaluate the quality of each path using E(p) = L(p) + Sw(p) * C2, and output the path with the lowest E(p). Here, p represents the path, L(p) represents the length of path p, Sw(p) represents the number of switching actions in path p, and C2 is the manually set action switching penalty coefficient.

[0039] Thus, the optimal path has been planned, and this path is imported into the industrial computer to automatically control the crane's motion system, thereby realizing automatic lifting of the crane.

[0040] The advantages of this invention are its simplicity and reduced labor intensity. Workers simply use a drone to capture the environment, import it into automated mapping software to generate a 3D model, and then automatically calculate the crane's optimal path based on GPS readings of the crane's initial position. The data acquisition module, the drone's RTK module, and environmental modeling software ensure rapid and accurate 3D model generation. Furthermore, based on a developed path algorithm, the optimal path is automatically generated, improving the efficiency, safety, and accuracy of the lifting process. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a diagram of the rigid body model of the crane of the present invention.

[0042] Figure 2 This is a diagram showing the crane motion relationship of the present invention.

[0043] Figure 3 It is a flow chart for implementing the present invention. DETAILED DESCRIPTION

[0044] The present invention will be further described in detail below with reference to the accompanying drawings and examples to facilitate understanding by those skilled in the art:

[0045] A motion path planning system for a crawler crane includes a path planning module loaded into a workstation, a data acquisition module, and a drone.

[0046] The data acquisition module is used to collect the rotation center, the luffing center, the starting point of the hoisted object and the placement point of the hoisted object of the crane and transmit them to the workstation.

[0047] The drone is used to collect on-site environmental information and transmit it to the workstation.

[0048] The path planning module is used to generate the optimal path, which is achieved through the UAV-based environment reconstruction method and the improved RRT algorithm. The steps are as follows:

[0049] Step 1: Build a rigid body model of the crawler crane

[0050] According to the structure and working principle of crawler crane and the hoisted objects, the crane hoisting system is abstracted into five rigid bodies, including the turntable for getting off and getting on as a rectangular block; the boom and the hoisted objects are abstracted as cylindrical blocks; the wire rope is abstracted as a rigid straight line that can move up and down (such as Figure 1 shown).

[0051] Step 2: Construct the kinematic model of crawler crane

[0052] The kinematic model is established based on the rigid body model of the crawler crane. The motion relationship diagram is as follows: Figure 2 As shown. The crane's lowering local coordinate system {0} is placed at the center of the lowering bottom, the turntable local coordinate system {1} is set at the intersection of the upper part of the turntable and the rotation center, the main boom's luffing local coordinate system {2} is set at the luffing center, and the hook local coordinate system {6} is placed at the bottom center of the hoisted object. In actual crane structures, the wire rope is not connected at the center of the main boom head, but is connected by a pulley set offset from the main boom's central axis. Therefore, when establishing the kinematic model, the boom head local coordinate system {3} is set at the boom head axis, and the jib local coordinate system {4} is set at the jib head axis. If the crane only has a main boom, the coordinate system is set at the boom head axis, and the boom head's wire rope local coordinate system {5} is set at the center of the pulley set for lowering the wire rope. These three coordinate systems are used to simulate the actual connection position of the wire rope. Therefore, the position of the crane hoisting system can be expressed by P{θ1,θ2,θ3,θ4,θ5,d6}; where θ1 is the turntable rotation angle, θ2 is the boom luffing angle, θ3 is the angle between the boom extension line and the horizontal direction, θ4 is the angle between the boom normal direction and the horizontal direction, θ5 is the angle between the negative direction of the boom normal line and the horizontal direction, and d6 is the wire rope extension distance;

[0053] Let d1 be the total height of the crawler crane turntable and the lower car, a2 be the distance between the upper car's rotation axis and the boom's luffing axis, a3 be the main boom length, a5 be the pulley block offset distance, and d6 be the wire rope extension distance.

[0054] Depend on Figure 2 The kinematic model expresses the position of the end of the suspended object as:

[0055]

[0056] in, T1 is the pose transformation matrix from the local coordinate system {0} to the local coordinate system {1}. 2 is the pose transformation matrix from the local coordinate system {1} to the local coordinate system {2}, is the pose transformation matrix from the local coordinate system {2} to the local coordinate system {3}, is the pose transformation matrix from the local coordinate system {3} to the local coordinate system {4}, is the pose transformation matrix from the local coordinate system {4} to the local coordinate system {5}, is the pose transformation matrix from the local coordinate system {5} to the local coordinate system {6}, is the pose transformation matrix from the local coordinate system {0} to the local coordinate system {6}.

[0057] During the crane hoisting process, the wire rope always remains vertically downward, so the algebraic relationship between θ2, θ3, θ4 and θ5 can be obtained:

[0058]

[0059] Represent θ3, θ4, and θ5 by θ2, and substitute each transformation matrix into the formula You can get:

[0060]

[0061] When the coordinates of the end of the suspended object are known (P x ,P y ,P z ) can inversely calculate the data of each joint angle variable. Therefore, the posture transformation matrix of the suspended object can be set as Then we can inversely obtain θ1, θ2, and d6:

[0062]

[0063] Where t represents

[0064]

[0065] According to the obtained forward and inverse kinematics formulas, the position parameters of the crane can be obtained by inversely calculating the movement of the hoisted object.

[0066] Step 3: Based on the environment model obtained by UAV scanning and the kinematic relationship of the crane, the crawler crane path planning is implemented based on the improved RRT. This includes the following steps:

[0067] S1: A drone takes aerial photos of the environment, collecting a sufficient number of photos using a visible light camera. These photos are then fed into the environmental modeling software, which automatically generates a 3D model based on the captured photos. Obstacles in the 3D model's hoisting environment are enclosed by regular geometric shapes using a bounding box algorithm. Obstacles are expanded based on the size of the hoisted object, allowing the object to be represented as a point during planning. The bounding box algorithm extracts the 3D spatial information into a 2D grid array.

[0068] S2: Start path planning, initialize the search tree, set the information contained in each node in the search tree to be the three action numbers from the previous crane posture to the current crane posture, and record the previous node of the node, that is, the parent node. The action number refers to the numbering of the three actions of rotation, luffing, and lifting. Set the root node of the search tree as the starting point of lifting, and set the initial action of the root node according to the configuration information.

[0069] S3: Select a random point P in the planning space.

[0070] S4: Traverse all nodes in the search tree and select a node as the parent node of the random point. The selection rule is that the direction from the parent node to the random point P is close to the direction of the target point and the distance is the shortest.

[0071] S5: Based on the crane motion information contained in the parent node selected in step S4, select the action to reach the random point and set it to m. The selection rule is determined by K(m). K(m) = |V(m)·Vp| + F(m). Each action is assumed to perform only one motion with one motion number. Action m has three motion numbers, so three K(m) values can be generated. The action with the largest K(m) value is selected as the extended action. V(m) is the direction vector corresponding to the possible extended action m, Vp is the direction vector from the parent node to the random point P, and F(m) is the action selection correction parameter used to reduce the number of action switches during path generation and speed up the search. When action m is the same as the parent node action, F(m) = C1*Rand, where C1 is the preset motion retention parameter and Rand is a random number between 0 and 1. When action m is different from the parent node action, if the corresponding posture of the parent node action is already the same as the landing point, then F(m) = C1*Rand; otherwise, F(m) = 0.

[0072] S6: Follow the action selected in step S5 and intercept the next node with a step length of t. At this time, the conversion formula of θ1, θ2, and d6 calculated in step 2 can be substituted into the coordinates of the end of the suspended object at this node (P x ,P y ,P z) and then calculate the three action number information of the crane posture of the node, and the step length t is the set movement distance parameter.

[0073] S7: Determine whether the new node is valid. The judgment conditions are that the corresponding crane posture is valid and there is no collision between the new node and the parent node. The valid crane posture means that the entire vehicle does not collide when the crane moves and each action number is within the set threshold range.

[0074] S8: Determine whether the end point has been reached. The judgment condition is that the distance from the node to the end point is less than a threshold. If the end point has been reached, the loop is exited and the process enters S10.

[0075] S9: Repeat steps S3-S8 until the number of repetitions reaches the set value or exit at step S8.

[0076] S10: A path is obtained based on the search tree nodes. The path includes all node positions in the search tree and action number information of each node.

[0077] S11: Repeat steps S2-S10 multiple times to obtain multiple paths. Since each search path is independent of each other, parallelization can be used to accelerate the process.

[0078] S12: Evaluate the quality of each path using E(p) = L(p) + Sw(p) * C2, and output the path with the lowest E(p). Here, p represents the path, L(p) represents the length of path p, Sw(p) represents the number of switching actions in path p, and C2 is the manually set action switching penalty coefficient.

[0079] Thus, the optimal path has been planned, and this path is imported into the industrial computer to automatically control the crane's motion system, thereby realizing automatic lifting of the crane.

[0080] A lifting experiment was conducted using an E-Zhong TGZC180 lattice-arm crawler crane. Based on common crane operating environments, two test scenarios were designed. The experiments involved lifting a 5t weight and setting different lifting positions. Multiple experiments were conducted. Scenario 1 simulated a complex lifting scenario, where the operator cannot see the position and must follow instructions to complete the lifting action. Obstacles were present around the target point. The experimental results are shown in Table 1 below.

[0081] Table 1 Scenario 1 (5t weight) 7 test data

[0082]

[0083] The eccentric distance within the positioning accuracy circle refers to the distance between the actual lifting positioning point and the positioning point set by the experiment.

[0084] It's easy to see that as the distance to the target increases, the lifting operation duration isn't significantly affected, remaining within the same order of magnitude. This suggests that the optimal path defined by path planning for intelligent lifting operations plays a significant role in lifting efficiency. While ensuring lifting accuracy and safety, it's reasonable to infer that the greater the distance between the target and starting point, and the heavier the load, the greater the advantages of the intelligent lifting solution.

[0085] In particular, in scenario 1, the 5-ton weight was hoisted using a combination of command coordination and manual lifting, as shown in the data for sequence 8 in Table 1, in 305 seconds. Under the same conditions, the automated lifting operation for sequence 7 took 212 seconds, a 30% improvement over manual lifting.

[0086] Scenario 2 simulates a complex, long-distance lifting scenario. Not only are obstacles placed around the target point, but multiple obstacles are also placed within the crane's path that must be crossed. The lifting load remains at 5 tons. The experimental results are shown in Table 2 below.

[0087] Table 2 Scenario 2 (5t weight) 6 test data

[0088]

[0089]

[0090] Table 2 shows the test data for lifting a 5-ton weight in Scenario 2. Compared to Scenario 1, Scenario 2's target point is farther away, requiring a more complex action sequence, with rotations accounting for a significantly higher proportion of the action sequence. Although the distance between the target point and the starting point in Scenario 2 is greater than in Scenario 1, the operation speed is actually faster and the efficiency is higher, demonstrating that automated operations over long distances can significantly improve efficiency.

Claims

1. A motion path planning system for a crawler crane, characterized in that: It includes the path planning module, data acquisition module and drone loaded into the workstation; The data acquisition module is used to collect the data of the crane's rotation center, luffing center, starting point of the hoisted object and the placement point of the hoisted object and transmit them to the workstation; The drone is used to collect on-site environmental information and transmit it to the workstation; The path planning module is used to generate the optimal path.

2. A method for planning a motion path of a crawler crane, applied to a motion path planning system for a crawler crane according to claim 1, characterized in that: Here are the steps: Step 1: Build a rigid body model of the crawler crane According to the structure and working principle of the crawler crane and the objects to be hoisted, the crane hoisting system is abstracted into five rigid bodies, including the lower and upper turntables which are abstracted into rectangular blocks. The boom and the suspended object are abstracted into cylindrical blocks; the wire rope is abstracted into a virtual rigid straight line that can move up and down; Step 2: Construct the kinematic model of crawler crane The kinematic model is established based on the rigid body model of the crawler crane; the crane's lower vehicle local coordinate system {0} is placed at the center of the lower vehicle bottom, the turntable local coordinate system {1} is set at the intersection of the upper part of the turntable and the rotation center, the main boom local coordinate system {2} is set at the center of the boom, and the hook local coordinate system {6} is placed at the bottom center of the hoisted object; in the actual crane structure, the wire rope is not connected at the center of the main boom head, but is connected by a pulley set offset from the main boom center axis. Therefore, when establishing the kinematic model, the boom head local coordinate system {3} is set at the boom head axis, and the auxiliary arm local coordinate system {4} is set at the center of the main boom head axis. {4} is set at the axis of the jib head. If the crane has only the main boom, the coordinate system is set at the axis of the boom head, and the local coordinate system of the wire rope at the boom head {5} is set at the center of the pulley group for lowering the wire rope. The three coordinate systems are used to simulate the actual connection position of the wire rope; therefore, the position and posture of the crane hoisting system can be expressed by P{θ1,θ2,θ3,θ4,θ5,d6}; among them, θ1 is the turntable rotation angle, θ2 is the boom amplitude change angle, θ3 is the angle between the boom extension line and the horizontal direction, θ4 is the angle between the boom normal direction and the horizontal direction, θ5 is the angle between the negative direction of the boom normal line and the horizontal direction, and d6 is the extension distance of the wire rope; Let d1 be the total height of the crawler crane turntable and the lower car, a2 be the distance between the upper car's rotation axis and the boom's luffing axis, a3 be the main boom length, a5 be the pulley block offset distance, and d6 be the wire rope extension distance; The kinematic model expresses the position of the end of the suspended object as follows: in, T1 is the pose transformation matrix from the local coordinate system {0} to the local coordinate system {1}. 2 is the pose transformation matrix from the local coordinate system {1} to the local coordinate system {2}, is the pose transformation matrix from the local coordinate system {2} to the local coordinate system {3}, is the pose transformation matrix from the local coordinate system {3} to the local coordinate system {4}, is the pose transformation matrix from the local coordinate system {4} to the local coordinate system {5}, is the pose transformation matrix from the local coordinate system {5} to the local coordinate system {6}, is the pose transformation matrix from the local coordinate system {0} to the local coordinate system {6}; During the crane hoisting process, the wire rope always remains vertically downward, so the algebraic relationship between θ2, θ3, θ4 and θ5 can be obtained: Represent θ3, θ4, and θ5 by θ2, and substitute each transformation matrix into the formula You can get: When the coordinates of the end of the suspended object are known (P x ,P y ,P z ) can inversely calculate the data of each joint angle variable; therefore, the posture transformation matrix of the suspended object can be set as Then we can inversely obtain θ1, θ2, and d6: Where t represents According to the obtained forward and inverse kinematics formulas, the position parameters of the crane can be obtained by inversely calculating the movement of the hoisted object. Step 3: Based on the kinematic relationship between the environment model obtained by the UAV scan and the crane, the crawler crane path planning is implemented based on the improved RRT. This includes the following steps: S1: A drone takes aerial photos of the environment, using a visible light camera to collect a sufficient number of photos. These photos are then input into the environmental modeling software, which automatically generates a 3D model based on the photos. Obstacles in the 3D model's hoisting environment are enclosed by regular geometric shapes using a bounding box algorithm. Obstacles in the environment are expanded accordingly based on the size of the hoisted object, allowing the hoisted object to be represented by points during planning. The bounding box algorithm extracts 3D spatial information into a 2D grid array. S2: Start path planning and initialize the search tree. Set each node in the search tree to contain the three action numbers from the previous crane position to the current crane position, and record the previous node of the node, that is, the parent node. The action number refers to the numbering of the three actions of rotation, luffing, and lifting. Set the root node of the search tree as the lifting starting point, and set the initial action of the root node according to the configuration information. S3: Select a random point P in the planning space; S4: traverse all nodes in the search tree and select a node as the parent node of the random point. The selection rule is that the direction from the parent node to the random point P is close to the direction of the target point and the distance is the shortest; S5: According to the crane motion information contained in the parent node selected in step S4, select the action to reach the random point, set the action to m, and the selection rule is determined according to K(m); wherein K(m) = |V(m)·Vp|+F(m), and it is set that each action only performs a movement with one action number. There are three action numbers for action m, so three K(m)s can be generated, and the action with the largest K(m) value is selected as the extended action; V(m) is the direction vector corresponding to the possible extended action m, Vp is the direction vector from the parent node to the random point P, and F(m) is the action selection correction parameter, which is used to reduce the number of action switches in the path and speed up the search when generating the path; when the action m is the same as the parent node action, F(m) = C1*Rand, where C1 is the preset action holding parameter and Rand is a random number between 0 and 1; when the action m is different from the parent node action, if the corresponding posture of the parent node action is already the same as the landing point, then F(m) = C1*Rand, otherwise F(m) = 0; S6: Follow the action selected in step S5 and intercept the next node with a step length of t. At this time, the conversion formula of θ1, θ2, and d6 calculated in step 2 can be substituted into the coordinates of the end of the suspended object at this node (P x ,P y ,P z ) and then calculate the three action number information of the crane posture of the node, and the step length t is the set movement distance parameter; S7: Determine whether the new node is valid. The judgment conditions are that the corresponding crane posture is valid and there is no collision between the new node and the parent node. The crane posture is valid when the crane moves without collision and each action number is within the set threshold range. S8: Determine whether the node has reached the end point. The judgment condition is that the distance from the node to the end point is less than a threshold. If the node has reached the end point, the loop is exited and the process goes to S10. S9: Repeat steps S3-S8 until the number of repetitions reaches the set value or exit at step S8; S10: Arrange the search tree nodes to obtain a path, which includes the positions of all nodes in the search tree and the action number information of each node; S11: Repeat steps S2-S10 multiple times to obtain multiple paths. Since each search path is independent of each other, parallelization can be used to accelerate the process. S12: Evaluate the quality of each path according to E(p) = L(p) + Sw(p) * C2, and output the path with the lowest E(p); where p represents the path, L(p) represents the length of path p, Sw(p) represents the number of switching actions in path p, and C2 is the manually set action switching penalty coefficient; Thus, the optimal path has been planned, and this path is imported into the industrial computer to automatically control the crane's motion system, thereby realizing automatic lifting of the crane.

Images