Welding robot intelligent process redundancy driven obstacle avoidance motion planning method and system and computer equipment

Abstract

The invention discloses an obstacle avoidance motion planning method and system for redundant drive of an intelligent process of a welding robot and computer equipment. According to the method, firstly, a motion path of a robot is planned based on welding process redundancy, then a sampling strategy that the position and posture of a welding gun are separated is adopted, then the sampled posture is mapped to a planar two-dimensional space, and the relevance between an obstacle collision boundary and the obstacle avoidance posture of the welding gun is constructed; an incremental extreme gradient lifting model is adopted to carry out probability modeling on the collision boundary and obstacle avoidance attitude relevance, nearest neighbor search and new planning node expansion are carried out based on target cost to obtain new planning nodes, collision detection is carried out on the new planning nodes, and the new planning nodes passing the collision detection are used for incremental learning of an obstacle avoidance model; and finally, an optimization strategy of spherical linear interpolation and cosine slow motion time mapping is adopted to ensure continuity and smoothness of a motion planning path trace. The problem that in the prior art, it is difficult to quickly calculate and generate a collision-free welding track is solved.

Description

Technical Field

[0001] This invention relates to the field of robotic welding technology, specifically to an obstacle avoidance motion planning method, system, and computer equipment for intelligent process redundancy-driven welding robot. Background Technology

[0002] Industrial robots, with their flexible, efficient, and precise operation characteristics, are now widely used in arc welding to achieve high productivity and ensure welding quality. As applications become more widespread, there is a growing expectation that robots can perform arc welding operations in narrow and complex environments. However, limited working space and complex workpiece structures can easily lead to robot collisions. Therefore, how to plan obstacle avoidance paths for the robot's position and posture during the welding process, taking into account the characteristics of the welding process, is a challenge for welding robots. For welding scenarios prone to collisions, welding robots can use path planning methods to adjust the welding torch posture to avoid obstacles, generating a collision-free path from the initial pose to the target pose. In practical welding applications, the path planning of welding robots requires the welding torch position to strictly follow the weld seam trajectory and avoid collisions. However, welding tasks are performed in Cartesian space, and path planning faces computational complexity challenges due to pose coupling constraints, which intensify with increasing product complexity. Therefore, how to achieve fast, collision-free welding path planning in complex and collision-prone welding scenarios is the main problem of this research.

[0003] Existing research on collision-free path planning can be summarized into three types: classical path planning, sampling-based path planning, and heuristic-based path planning. These algorithms have shown significant effectiveness in both academic research and industrial applications, but the complex working environment of welding robots limits the performance of existing methods. Classical path planning algorithms can meet computational efficiency requirements and calculate collision-free paths in simple environments; however, in complex and collision-prone welding scenarios, these algorithms are prone to getting trapped in local minima and face high computational costs. While sampling-based path planning algorithms can effectively calculate and generate collision-free paths, they have limitations, only handling point-to-point path planning and calculating only the collision-free path from the initial pose to the target pose, without considering intermediate states. When strict constraints on weld seam positions are required in continuous welding path planning, sampling-based path planning algorithms become unsuitable for direct application.

[0004] The problem to be solved is how to quickly calculate and generate collision-free welding trajectories in complex and collision-prone continuous welding path planning applications. Summary of the Invention

[0005] The primary objective of this invention is to overcome the shortcomings of the existing technology and provide an obstacle avoidance motion planning method for welding robots driven by intelligent process redundancy. This obstacle avoidance motion planning method for welding robots driven by intelligent process redundancy can solve the problem of difficulty in quickly calculating and generating collision-free welding trajectories in complex and collision-prone continuous welding path planning applications, and achieve near-optimal collision-free continuous welding path generation in a single planning operation.

[0006] The second objective of this invention is to provide an obstacle avoidance motion planning system for intelligent process redundancy drive of welding robots.

[0007] A third objective of this invention is to provide a computer device.

[0008] The first objective of this invention is achieved through the following technical solution: The obstacle avoidance motion planning method for intelligent process redundancy-driven welding robot includes the following steps:

[0009] S1. Obtain the weld trajectory to be planned, and perform motion planning path for the robot based on welding process redundancy;

[0010] S2. A sampling strategy that separates the position and attitude of the welding torch is adopted, and the position and attitude of the welding torch are sampled separately.

[0011] S3. Map the sampled attitude to a two-dimensional planar space and construct the correlation between the obstacle collision boundary and the welding torch obstacle avoidance attitude;

[0012] S4. An incremental extreme gradient boosting model is used to probabilistically model the correlation between the collision boundary and the obstacle avoidance posture, resulting in a nonlinear mapping model.

[0013] S5. The position and orientation of the welding gun are sampled to obtain candidate nodes. Based on the target cost, the nearest neighbor search and new planning node expansion are performed to obtain new planning nodes.

[0014] S6. Perform collision detection on the new planning nodes and use the new planning nodes that pass the collision detection for incremental learning of the nonlinear mapping model.

[0015] S7. An optimization strategy using spherical linear interpolation and cosine easing time mapping is adopted to ensure the continuity and smoothness of the motion planning trajectory.

[0016] Step S1 includes the following specific steps:

[0017] S11. Let the motion planning path of the welding robot be described in Cartesian space, then the welding torch has x in the coordinate system. tcp y tcp z tcp Three translation variables and rx about the x, y, and z axes respectively. tcp rytcp rz tcp Three rotational variables, the weld trajectory is formed by the discrete weld point set T = {t1, t2, ..., t n} constitutes, each weld point t i =(p i ,r i The Cartesian space pose of a given position includes a position component p. i =(x i ,y i ,z i ) and attitude components r i =(rx i ,ry i ,rz i ), where i = 0, 1, ..., n, and n is a natural number;

[0018] S12. The position of the welding wire tip of the welding torch must strictly follow the position of the weld point, that is... The set of rotational transformations consists of multiple sets of attitude angle combinations. The j-th set of welding gun attitude angle combinations is:

[0019]

[0020] in, The welding torch attitude angle, α l ,β l ,γ l α is the lower limit of the welding torch attitude angle redundancy angle. u ,β u ,γ u This is the upper limit of the redundant angle of the welding torch attitude angle.

[0021] Step S2 includes the following specific steps:

[0022] S21. Position sampling uses a three-dimensional Gaussian distribution sampling strategy that integrates the correlation between adjacent weld positions. The sampling region of this three-dimensional Gaussian distribution sampling strategy is an ellipsoid centered on the current weld point position, with its major axis along the previous weld point t. i-1 Up to the current weld point t i The direction of the connection; let the position of the previous weld point be p. i-1 = [x1, y1, z1], the current weld point position is p i = [x2, y2, z2], then the 3D Gaussian sampling of the current node is:

[0023] p tcp-rand ~N(μ,σ,S(p) i-1 ,p i ))

[0024] Where, ptcp-rand Here, N(μ,σ,S) represents the sampling point of the current node, where S is the three-dimensional Gaussian sampling, μ is the mean, and μ = p. i σ is the standard deviation, σ = ||p i -p i-1 ||, S(·) is the covariance matrix;

[0025] S22. Attitude sampling utilizes the tolerance allowed by the redundancy of the welding process to achieve obstacle avoidance by minimizing the cost of adjusting the welding gun attitude.

[0026] Step S3 includes the following specific steps:

[0027] S31. The welding process involves 5 degrees of freedom motion, with 3 degrees of freedom used for translation and 2 degrees of freedom used for rotation to determine the welding torch posture; posture sampling is performed on the 2 degrees of freedom used for rotation to form a two-dimensional sampling space.

[0028] S32. Attitude sampling considers the distribution characteristics of obstacles and the collision patterns between objects. Collisions occur when there is point-to-point contact between objects. As the welding torch moves continuously, the trajectory of the collision point is fitted into a spatial curve, thus forming the collision boundary.

[0029] S33. When the collision boundary is oblique or circular, the collision force is projected onto the xy plane of the welding torch coordinate system in the third or fourth quadrant. When the collision boundary is a special type, the boundary extension method is used to add a transition area on the basis of the original collision boundary to correct it into oblique or circular shape.

[0030] Step S4 specifically includes the following steps:

[0031] S41. Based on the welding obstacle avoidance task, construct a dataset;

[0032] S42. Determine the initial value range of the core parameters of the incremental extreme gradient boosting model;

[0033] S43. Using the Bayesian optimization method, the optimal parameter combination is obtained within the value range, and based on the training dataset, a nonlinear mapping model between obstacle avoidance angle and collision force is constructed using the optimal parameter configuration.

[0034] Step S5 includes the following specific steps:

[0035] S51. The objective cost is to minimize the cost between adjacent search nodes, where the cost of adjacent nodes includes position cost, angle change cost, and attitude deviation cost. Let the attitude deviation be defined as the rotation angle between the current welding torch orientation and the corresponding weld seam reference orientation. The cost of adjacent nodes is... but:

[0036]

[0037] in, It is a node arrive Total cost It is the location cost between adjacent nodes. The angular cost between adjacent nodes It is the attitude deviation between the welding torch and the weld point at the previous node;

[0038] S52, after sampling, the candidate node t of the current weld point is obtained. tcp-rand Then, determine its nearest neighbor node, t. tcp-near For the current sampling point and the previous planned node The node with the minimum objective cost is expressed as:

[0039]

[0040] in, For the j-th node in the redundant attitude set of the previous weld point, From to t tcp-rand The total target cost;

[0041] S53. After determining the nearest neighbor node, a new planned node t for the current weld point is generated using a path turning strategy. tcp-new ,but:

[0042] t tcp-new =[p tcp-new ,r tcp-near +r tcp-rand ]

[0043] Where, p tc For the new node position, r tcp-near For the pose of the nearest neighbor node, r tcp-rand The obstacle avoidance posture of the welding torch is predicted by a nonlinear mapping model.

[0044] Step S6 includes the following steps:

[0045] S61. Based on the CAD model, perform collision detection and incremental adjustment to perform collision detection on the newly planned nodes;

[0046] S62. When a new planning node passes the collision detection, add this new planning node to the motion planning path and update the sequence number of the weld point to obtain a new motion planning path.

[0047] Step S7 includes the following steps:

[0048] S71. The new motion planning path is optimized by combining the path optimization method of quaternion spherical linear interpolation and cosine easing time mapping.

[0049] S72. Perform collision detection on each interpolation point in the optimized new motion planning path to ensure the continuity and smoothness of the motion planning trajectory.

[0050] An obstacle avoidance motion planning system driven by intelligent process redundancy for welding robots includes:

[0051] The acquisition module is used to acquire the weld trajectory to be planned;

[0052] The modeling module is used to probabilistically model the correlation between obstacle collision boundaries and welding torch obstacle avoidance posture using an incremental extreme gradient boosting model.

[0053] The first calculation module calculates candidate planning nodes based on weld seam trajectory points;

[0054] The second calculation module determines the nearest neighbor node of the candidate planning node based on the target cost, and then calculates the planning node of the current weld point.

[0055] The third calculation module is used for collision detection of newly planned nodes, and then the obstacle avoidance data of the new nodes is used for incremental learning of the nonlinear mapping model.

[0056] The fourth calculation module, based on an optimization strategy of spherical linear interpolation and cosine easing time mapping, ensures the continuity and smoothness of the planning process trajectory.

[0057] A computer device, including a processor and a memory for storing processor-executable programs, wherein when the processor executes the program stored in the memory, it implements the obstacle avoidance motion planning method for intelligent process redundancy drive of welding robots as described in any one of claims 1-8.

[0058] The present invention has the following advantages over the prior art:

[0059] 1. This invention proposes a process redundancy-driven obstacle avoidance planning method for welding robots, which is aimed at the automatic generation of welding trajectories for welding robots in complex and collision-prone scenarios. Based on the welding torch position constraints in the welding process, it utilizes the degree of freedom redundancy of the allowable deviations of the welding robot and welding torch postures, and realizes obstacle avoidance motion planning for the welding robot by dynamically optimizing the deviations of the welding torch postures.

[0060] 2. This invention maps the attitude adjustment values ​​of the welding process to a two-dimensional planar space, and constructs a correlation model between the obstacle collision boundary, the direction of the collision force and the obstacle avoidance angle in the welding process. It uses the incremental learning extreme gradient boosting model (XGBoost) to perform redundant two-dimensional space probability modeling, so as to realize the prediction of the collision boundary evolution trend and the optimal obstacle avoidance angle.

[0061] 3. This invention employs a motion planning method that separates position and attitude sampling and produces smooth trajectories, enabling the generation of near-optimal collision-free continuous welding paths in a single planning iteration. To address the inefficiency of pose-coupled planning in Cartesian space, position and attitude are sampled separately. Position sampling is based on a three-dimensional Gaussian distribution, while attitude sampling is performed using incremental learning XGBoost to predict the optimal obstacle avoidance angle. A composite optimization method considering position cost, angle cost, and attitude deviation, along with a spherical linear interpolation method based on cosine easing time mapping, is used to ensure the continuity of the welding trajectory and welding quality.

[0062] 4. This invention solves the problem that existing motion planning algorithms have low stability, poor applicability, and high computational cost in complex and collision-prone continuous welding motion planning applications, making it difficult to quickly calculate and generate collision-free welding trajectories. By using a motion planning method based on an incremental learning extreme gradient boosting model to predict the optimal obstacle avoidance angle, a near-optimal collision-free welding path can be generated in a single planning operation. Attached Figure Description

[0063] Figure 1 This is a flowchart of the present invention provided as an exemplary embodiment of the present invention;

[0064] Figure 2 This is a robotic welding system provided as an exemplary embodiment of the present invention;

[0065] Figure 3 This is a set of redundant attitude angles provided in an exemplary embodiment of the present invention;

[0066] Figure 4 This is a collision boundary type provided in an exemplary embodiment of the present invention;

[0067] Figure 5 This is an exemplary embodiment of the present invention, showing the projection of the collision force onto the xy plane.

[0068] Figure 6 This is a special collision boundary type provided in an exemplary embodiment of the present invention;

[0069] Figure 7 This is an obstacle model provided in an exemplary embodiment of the present invention;

[0070] Figure 8 This is a search space structure provided in an exemplary embodiment of the present invention;

[0071] Figure 9 This is an exemplary embodiment of the present invention, which provides the selection of nearest neighbor points and the expansion of new planning nodes;

[0072] Figure 10This is a robot collision detection workflow provided by an exemplary embodiment of the present invention. Detailed Implementation

[0073] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0074] like Figure 1 As shown, the obstacle avoidance motion planning method of intelligent process redundancy drive for this welding robot includes the following steps:

[0075] S1. Obtain the weld trajectory to be planned, and perform motion planning path for the robot based on welding process redundancy;

[0076] Step S1 includes the following specific steps:

[0077] S11, such as Figure 2 As shown, if the motion planning path of the welding robot is described in Cartesian space, then the welding torch has x in the coordinate system. tcp y tcp z tcp Three translation variables and rx about the x, y, and z axes respectively. tcp ry tcp rz tcp Three rotational variables, the weld trajectory is formed by the discrete weld point set T = {t1, t2, ..., t n} constitutes, each weld point t i =(p i ,r i The Cartesian space pose of a given position includes a position component p. i =(x i ,y i ,z i ) and attitude components r i =(rx i ,ry i ,rz i ), where i = 0, 1, ..., n, and n is a natural number; in the coordinate system, the weld trajectory posture is defined as the welding forward direction representing the x-axis and the weld normal direction representing the z-axis.

[0078] S12. According to the welding process requirements, the position of the welding wire tip of the welding torch must strictly track the weld point position, i.e., p tcpi =p i Meanwhile, the welding torch posture allows for specific angular deviations, such as... Figure 3 As shown, the attitude solution that satisfies the process requirements is not unique, but rather a set of rotational transformations consisting of multiple combinations of attitude angles. The j-th combination of welding gun attitude angles is:

[0079]

[0080] in, The welding torch attitude angle, α l ,β l ,γ l α is the lower limit of the welding torch attitude angle redundancy angle. u ,β u ,γ u This is the upper limit of the redundant angle of the welding torch attitude angle.

[0081] By determining the welding torch posture in the above manner, welding quality can be guaranteed while the flexibility of the welding path can be improved by adjusting the combination of posture angles.

[0082] S2. A sampling strategy that separates the position and attitude of the welding torch is adopted, and the position and attitude of the welding torch are sampled separately.

[0083] Step S2 includes the following specific steps:

[0084] S21. Position sampling uses a three-dimensional Gaussian distribution sampling strategy that integrates the correlation between adjacent weld positions. The sampling region of this three-dimensional Gaussian distribution sampling strategy is an ellipsoid centered on the current weld point position, with its major axis along the previous weld point t. i-1 Up to the current weld point t i The direction of the connection; let the position of the previous weld point be p. i-1 = [x1, y1, z1], the current weld point position is p i = [x2, y2, z2], then the 3D Gaussian sampling of the current node is:

[0085] p tcp-rand ~N(μ,σ,S(p) i-1 ,p i ))

[0086] Where, p tcp-rand Here, N(μ,σ,S) represents the sampling point of the current node, where S is the three-dimensional Gaussian sampling, μ is the mean, and μ = p. i σ is the standard deviation, σ = ||p i -p i-1 ||, S(·) is the covariance matrix;

[0087] The covariance matrix is ​​calculated as follows:

[0088]

[0089] Where Var(·) is the variance; Cov(·) is the covariance. Taking Var(x) as an example, the variance is calculated as follows:

[0090]

[0091] Where, xi This represents the coordinate value of the weld point in the x-direction. The average coordinate value in the x-direction of adjacent weld points.

[0092] Taking Cov(x,y) as an example, the covariance is calculated as follows:

[0093]

[0094] Among them, y i This represents the coordinate value of the weld point in the y-direction. This represents the average coordinate value in the y-direction of adjacent weld points.

[0095] The ellipsoid in this embodiment is more compact and closer to the effective search space.

[0096] S22. Attitude sampling utilizes the tolerance allowed by the redundancy of the welding process to achieve obstacle avoidance by minimizing the cost of adjusting the welding gun attitude.

[0097] This implementation employs a position and attitude separation sampling strategy, which can reduce the dimensionality of the search space and improve planning efficiency.

[0098] S3. Map the sampled attitude to a two-dimensional planar space and construct the correlation between the obstacle collision boundary and the welding torch obstacle avoidance attitude;

[0099] Step S3 includes the following specific steps:

[0100] S31. The welding process involves 5 degrees of freedom motion, with 3 degrees of freedom used for translation and 2 degrees of freedom used for rotation to determine the welding torch posture; posture sampling is performed on the 2 degrees of freedom used for rotation to form a two-dimensional sampling space.

[0101] S32. Attitude sampling considers the distribution characteristics of obstacles and the collision patterns between objects. Collisions occur as point-to-point contact between objects. As the welding torch moves continuously, the trajectory of the collision point is fitted into a spatial curve, thus forming a collision boundary. In the two-dimensional sampling space, the projection of this boundary is represented as a planar line segment. For example... Figure 4 As shown, based on the shape of the line segment and its interaction with the weld trajectory, the collision boundary can be further divided into oblique line type, circular arc type and special type.

[0102] S33. When the collision boundary is oblique or circular, the collision force is projected onto the xy plane of the welding torch coordinate system in the third or fourth quadrant; such as Figure 5 As shown, according to the obstacle avoidance direction and the right-hand rule, when the collision force is in the third quadrant, the welding torch should rotate in the negative x-axis and simultaneously rotate in the positive y-axis; when the collision force is in the fourth quadrant, it needs to rotate in the positive x-axis and simultaneously rotate in the positive y-axis.

[0103] When the collision boundary is of a special type, a boundary extension method is used to add a transition region to the original collision boundary to correct it into a diagonal or circular shape. For example... Figure 6 As shown, there is no direct correspondence between the collision force direction and the obstacle avoidance direction, making it difficult to use directly for attitude adjustment. This embodiment proposes a boundary extension method, which adds a transition region to the original collision boundary and modifies it into a diagonal or arc-shaped boundary to unify the collision boundary type.

[0104] The collision analysis described above reveals a significant correlation between the obstacle avoidance direction and the collision force direction. Specifically, the obstacle avoidance angle of the welding torch depends on the angle between the collision force and the x-axis and y-axis in the welding torch coordinate system: as the angle between the collision force and the x-axis increases, the rotation angle of the welding torch around the x-axis increases; as the angle between the collision force and the y-axis increases, the rotation angle of the welding torch around the y-axis exhibits a characteristic of first increasing and then decreasing.

[0105] S4. An incremental extreme gradient boosting model is used to probabilistically model the correlation between the collision boundary and the obstacle avoidance posture, resulting in a nonlinear mapping model, which is then used to predict the evolution trend of the collision boundary and the optimal obstacle avoidance posture of the welding torch.

[0106] Step S4 specifically includes the following steps:

[0107] S41. Based on the welding obstacle avoidance task, construct a dataset; specifically, based on a progressive obstacle avoidance difficulty enhancement strategy, by increasing the angle between the obstacle boundary and the weld trajectory with a fixed step size (e.g., an increment step size of 7°), construct 24 obstacle models with geometric differences. Figure 7 The four obstacle models shown are 1, 2, 3, and 4.

[0108] Based on this, a simulated welding environment of welding robot and obstacle was built, and collision force and corresponding obstacle avoidance angle data were collected during obstacle avoidance, including the angle θ between the collision force and the x-axis of the welding torch coordinate system. x The angle θ between the collision force and the y-axis of the welding torch coordinate system y The magnitude of the collision force F and the obstacle avoidance angle of the welding torch around the x and y axes. and The obstacle avoidance strategy is implemented using an angle incremental approach, that is, by using multiple small angles. The summation yields the minimum welding torch obstacle avoidance angle. The calculation strategy is as follows: when the angle between the collision force F and the y-axis is π / 2, Take the maximum value, and as the included angle approaches 0 or π, Decrease; when the angle between the collision force F and the y-axis is π / 4 or 3π / 4. Take the maximum value, and as the included angle approaches 0, π / 2, or π, Decrease. The corresponding calculation method is as follows:

[0109]

[0110] in, The increment of the welding torch rotation angle around the x-axis that minimizes obstacle avoidance at the node; The increment of the welding torch rotation angle around the x-axis is calculated during the i-th collision of the node; For a given angle increment; θ y Let F be the angle between the collision force F and the y-axis.

[0111]

[0112] r y =1-r x

[0113] in, The increment of the welding torch rotation angle around the y-axis that minimizes obstacle avoidance at the node. This is the increment of the welding torch rotation angle around the y-axis calculated during the i-th collision at the node.

[0114] S42. Determine the initial value range of the core parameters of the incremental extreme gradient boosting model. Specifically, XGBoost (Incremental Extreme Gradient Boosting Model) is a gradient boosting framework based on decision tree ensembles. It iteratively adds new trees to correct the residuals of the previous round, thereby gradually improving prediction accuracy. During model construction, the key parameters of the incremental extreme gradient boosting model include: n_estimators (number of decision trees), max_depth (maximum tree depth), learning_rate (learning rate), subsample (sample sampling ratio), and colsample_bytree (feature sampling ratio). The initial value range of each parameter is determined by referring to relevant literature and combining experience.

[0115] S43. Using the Bayesian optimization method, the optimal parameter combination is obtained within the value range. Based on the training dataset, a nonlinear mapping model between obstacle avoidance angle and collision force is constructed using the optimal parameter configuration. This model is then used to predict the evolution trend of the collision boundary and the optimal obstacle avoidance posture of the welding torch.

[0116] In the XGBoost hyperparameter tuning process, a set of hyperparameter combinations is first randomly initialized within a predetermined parameter range, and the model is trained on the training set. The corresponding objective function value is then calculated using the validation set. Next, assuming the objective function follows a Gaussian process, the posterior distribution can be calculated by combining the prior distribution and the observed data.

[0117]

[0118] Where t is the objective function; x is the hyperparameter combination; p(t) is the prior distribution; p(x) is the normalization factor; p(x|t) is the likelihood function; and p(t|x) is the posterior distribution.

[0119] Based on this, the Gaussian process parameters are updated to obtain a new surrogate model, which is then used to select the most promising hyperparameter combination for the next sampling and validation. By continuously iterating the above process, the model gradually approaches the global optimum until a preset stopping condition is met.

[0120] Through Bayesian optimization, the model can obtain the optimal hyperparameter configuration, thus possessing good learning ability. An incremental extreme gradient boosting model is trained using 80% of the dataset as the training set to establish a prediction model for obstacle avoidance angles and obstacle boundary features during the welding process. The model inputs are F and θ. x and θ y The outputs are ∇Rx and ∇Ry. After training, the remaining 20% ​​of the dataset is used as the test set for model validation.

[0121] S5. The position and orientation of the welding gun are sampled to obtain candidate nodes. Based on the target cost, the nearest neighbor search and new planning node expansion are performed to obtain new planning nodes.

[0122] Step S5 includes the following specific steps:

[0123] S51, such as Figure 8 As shown, the search space construction includes both position and attitude: position sampling guides the path search along the weld seam trajectory, while attitude sampling enables obstacle avoidance. In motion planning, the optimization objective is to minimize the cost between adjacent search nodes, where the cost of adjacent nodes includes position cost, angle change cost, and attitude deviation cost. Let the attitude deviation be defined as the rotation angle between the current welding torch orientation and the corresponding weld seam reference orientation. The cost of adjacent nodes is... but:

[0124]

[0125] in, It is a node arrive Total cost It is the location cost between adjacent nodes. It is the angular cost between adjacent nodes. It is the attitude deviation between the welding torch and the weld point at the previous node;

[0126] Location cost calculation

[0127]

[0128] Angular cost calculation,

[0129]

[0130] in, They are respectively The rotation matrix form.

[0131] Attitude deviation calculation

[0132]

[0133] in, n z They are respectively r i-1 Let r = (α, β, γ) be the yaw angle vector, then n is calculated as n = [sinβ - cosβsinα cosβcosα]. T .

[0134] S52, after sampling, the candidate node t of the current weld point is obtained. tcp-rand Then, determine its nearest neighbor node, such as Figure 9 As shown, the nearest neighbor node t tcp-near For the current sampling point and the previous planned node The node with the minimum objective cost is expressed as:

[0135]

[0136] in, For the j-th node in the redundant attitude set of the previous weld point, From to t tcp-rand The total target cost;

[0137] S53. After determining the nearest neighbor node, a new planned node t for the current weld point is generated using a path turning strategy. tcp-n ,but:

[0138] t tcp-new =[p tcp ,r tcp-near +r tcp-rand ]

[0139] Where, p tcp-n For the new node position, r tcp-near For the pose of the nearest neighbor node, r tcp-rand The obstacle avoidance posture of the welding torch is predicted by a nonlinear mapping model.

[0140] S6. Perform collision detection on the new planning nodes and use the new planning nodes that pass the collision detection for incremental learning of the nonlinear mapping model.

[0141] Step S6 includes the following steps:

[0142] S61. Collision detection and incremental adjustment are performed on newly planned nodes based on the CAD model. Specifically, this detection is implemented on the NVIDIA Isaac Sim platform and integrates contact force sensors to build a real-time detection system. The process is as follows: Figure 10 As shown. The specific steps include:

[0143] S611, System Initialization: Set the collision flag tag collission =false; initialize the iteration counter l = 1;

[0144] S612, Joint Space Mapping: Mapping new node t using robot inverse kinematics (IK). tcp-new Mapping from Cartesian space to robot joint space can be represented as: q = IK(t) tcp-new );

[0145] S613. Robot State Calculation: The robot's CAD state, described by the joint angle vector q, is calculated using the robot's forward kinematics (FK). This state is represented as: Robot config =FK(q);

[0146] S614. Collision detection and incremental adjustment based on CAD model: If a collision is detected, i.e., tag... collision If the value is true, then interference needs to be eliminated through obstacle avoidance angle prediction and welding torch posture adjustment. Specifically, there are two cases:

[0147] a) If the initially predicted obstacle avoidance angle is less than the expected angle, the contact force direction quadrant remains unchanged. At this point, based on the collision data F and θ... x and θ y Obstacle avoidance angle r predict Predict, then adjust the current welding torch angle r tcp-n With r predict Superimposed to obtain a new welding torch angle r tcp-n Collision detection is then performed. If collisions still exist, this incremental adjustment process continues until the collision detection is passed or the maximum number of iterations is reached. Max ;

[0148] b) If the initially predicted obstacle avoidance angle is greater than the expected angle, the contact force direction quadrant changes, potentially causing the welding torch to collide with an obstacle on the opposite side. In this case, an incremental adjustment process is also performed until collision detection is passed or the maximum number of iterations is reached. Max .

[0149] S62. When a new planning node passes the collision detection, add this new planning node to the motion planning path and update the sequence number of the weld point to obtain a new motion planning path.

[0150] S7. An optimization strategy using spherical linear interpolation and cosine easing time mapping is adopted to ensure the continuity and smoothness of the motion planning trajectory.

[0151] Step S7 includes the following steps:

[0152] S71. In actual welding, after obstacle avoidance, the welding torch needs to smoothly transition back to the preset weld trajectory to ensure the continuity of the welding process and the quality of the weld. Therefore, a path optimization method combining quaternion spherical linear interpolation and cosine easing time mapping is used to optimize the new motion planning path.

[0153] First, determine if N consecutive N num The attitude consistency of each extension point is verified to confirm that the welding torch has completely avoided obstacles. Subsequently, the Euler angle attitude of the first extension point of the continuous segment and the weld point attitude corresponding to the last extension point are converted into a unit quaternion Q. first and Q last This is to ensure the geometric correctness of the attitude interpolation.

[0154] Next, Slerp interpolation is performed on the quaternions at the beginning and end points, and the interpolation parameters are controlled by a cosine easing function to achieve a smooth change in angular velocity, first accelerating and then decelerating, thus generating a smooth interpolated attitude for the transition segment. The attitude interpolation calculation method is as follows:

[0155]

[0156] Where θ = arccos(Q) first Q lsst () is the angle between two quaternions, and u is the cosine easing function. i = 0, 1, ..., N num -1, N nim This represents the number of points extended from the continuous segment.

[0157] S72. Collision detection is performed on each interpolation point in the optimized new motion planning path to ensure the continuity and smoothness of the motion planning trajectory. Specifically, if all interpolation points pass the detection, the corresponding nodes in the planned path are updated; otherwise, the original planned nodes are retained to ensure path feasibility. Through the above method, the welding torch's attitude change after obstacle avoidance is smooth and continuous, and the combination of collision detection ensures path feasibility, thereby effectively improving the safety of welding operations and weld quality.

[0158] An obstacle avoidance motion planning system driven by intelligent process redundancy for welding robots includes:

[0159] The acquisition module is used to acquire the weld trajectory to be planned;

[0160] The modeling module is used to probabilistically model the correlation between obstacle collision boundaries and welding torch obstacle avoidance posture using an incremental extreme gradient boosting model.

[0161] The first calculation module calculates candidate planning nodes based on weld seam trajectory points;

[0162] The second calculation module determines the nearest neighbor node of the candidate planning node based on the target cost, and then calculates the planning node of the current weld point.

[0163] The third calculation module is used for collision detection of newly planned nodes, and then the obstacle avoidance data of the new nodes is used for incremental learning of the model.

[0164] The fourth calculation module, based on an optimization strategy of spherical linear interpolation and cosine easing time mapping, ensures the continuity and smoothness of the planning process trajectory.

[0165] A computer device, including a processor and a memory for storing processor-executable programs, wherein when the processor executes the program stored in the memory, it implements the obstacle avoidance motion planning method for intelligent process redundancy drive of welding robots as described in any one of claims 1-8.

[0166] The above-described specific embodiments are preferred embodiments of the present invention and are not intended to limit the present invention. Any other changes or equivalent substitutions made without departing from the technical solution of the present invention are included within the protection scope of the present invention.

Claims

1. A method for obstacle avoidance motion planning driven by intelligent process redundancy in welding robots, characterized in that, Includes the following steps: S1. Obtain the weld trajectory to be planned, and perform motion planning path for the robot based on welding process redundancy; S2. A sampling strategy that separates the position and attitude of the welding torch is adopted, and the position and attitude of the welding torch are sampled separately. S3. Map the sampled attitude to a two-dimensional planar space and construct the correlation between the obstacle collision boundary and the welding torch obstacle avoidance attitude; S4. An incremental extreme gradient boosting model is used to probabilistically model the correlation between the collision boundary and the obstacle avoidance posture, resulting in a nonlinear mapping model. S5. The position and orientation of the welding gun are sampled to obtain candidate nodes. Based on the target cost, the nearest neighbor search and new planning node expansion are performed to obtain new planning nodes. S6. Perform collision detection on the new planning nodes and use the new planning nodes that pass the collision detection for incremental learning of the nonlinear mapping model. S7. An optimization strategy using spherical linear interpolation and cosine easing time mapping is adopted to ensure the continuity and smoothness of the motion planning trajectory.

2. The obstacle avoidance motion planning method for intelligent process redundancy drive of welding robots according to claim 1, characterized in that, Step S1 includes the following specific steps: S11. Let the motion planning path of the welding robot be described in Cartesian space, then the welding torch has x in the coordinate system. tcp y tcp z tcp Three translation variables and rx about the x, y, and z axes respectively. tcp ry tcp rz tcp Three rotational variables, the weld trajectory is formed by the discrete weld point set T = {t1, t2, ..., t n } constitutes, each weld point t i =(p i ,r i The Cartesian space pose of a given position includes a position component p. i =(x i ,y i ,z i ) and attitude components r i =(rx i ,ry i ,rz i ), where i = 0, 1, ..., n, and n is a natural number; S12. The position of the welding wire tip of the welding torch must strictly follow the position of the weld point, that is... The set of rotational transformations consists of multiple sets of attitude angle combinations. The j-th set of welding gun attitude angle combinations is: in, The welding torch attitude angle, α l ,β l ,γ l α is the lower limit of the welding torch attitude angle redundancy angle. u ,β u ,γ u This is the upper limit of the redundant angle of the welding torch attitude angle.

3. The obstacle avoidance motion planning method for intelligent process redundancy drive of welding robots according to claim 1, characterized in that, Step S2 includes the following specific steps: S21. Position sampling uses a three-dimensional Gaussian distribution sampling strategy that integrates the correlation between adjacent weld positions. The sampling region of this three-dimensional Gaussian distribution sampling strategy is an ellipsoid centered on the current weld point position, with its major axis along the previous weld point t. i-1 Up to the current weld point t i The direction of the line connecting them; let The location of the previous weld point is p. i-1 = [x1, y1, z1], the current weld point position is p i = [x2, y2, z2], then the 3D Gaussian sampling of the current node is: p tcp-rand ~N(μ,σ,S(p i-1 ,p i )) Where, p tcp-rand Here, N(μ,σ,S) represents the sampling point of the current node, where S is the three-dimensional Gaussian sampling, μ is the mean, and μ = p. i σ is the standard deviation, σ = ||p i -p i-1 ||, S(·) is the covariance matrix; S22. Attitude sampling utilizes the tolerance allowed by the redundancy of the welding process to achieve obstacle avoidance by minimizing the cost of adjusting the welding gun attitude.

4. The obstacle avoidance motion planning method for intelligent process redundancy drive of welding robots according to claim 1, characterized in that, Step S3 includes the following specific steps: S31. The welding process involves 5 degrees of freedom motion, with 3 degrees of freedom used for translation and 2 degrees of freedom used for rotation to determine the welding torch posture; posture sampling is performed on the 2 degrees of freedom used for rotation to form a two-dimensional sampling space. S32. Attitude sampling considers the distribution characteristics of obstacles and the collision patterns between objects. Collisions occur when there is point-to-point contact between objects. As the welding torch moves continuously, the trajectory of the collision point is fitted into a spatial curve, thus forming the collision boundary. S33. When the collision boundary is oblique or circular, the collision force is projected onto the xy plane of the welding torch coordinate system in the third or fourth quadrant. When the collision boundary is a special type, the boundary extension method is used to add a transition area on the basis of the original collision boundary to correct it into oblique or circular shape.

5. The obstacle avoidance motion planning method for intelligent process redundancy drive of welding robots according to claim 1, characterized in that, Step S4 specifically includes the following steps: S41. Based on the welding obstacle avoidance task, construct a dataset; S42. Determine the initial value range of the core parameters of the incremental extreme gradient boosting model; S43. Using the Bayesian optimization method, the optimal parameter combination is obtained within the value range, and based on the training dataset, a nonlinear mapping model between obstacle avoidance angle and collision force is constructed using the optimal parameter configuration.

6. The obstacle avoidance motion planning method for intelligent process redundancy drive of welding robots according to claim 1, characterized in that, Step S5 includes the following specific steps: S51. The objective cost is to minimize the cost between adjacent search nodes, where the cost of adjacent nodes includes position cost, angle change cost, and attitude deviation cost. Let the attitude deviation be defined as the rotation angle between the current welding torch orientation and the corresponding weld seam reference orientation. The cost of adjacent nodes is... but: in, It is a node arrive Total cost It is the location cost between adjacent nodes. The angular cost between adjacent nodes It is the attitude deviation between the welding torch and the weld point at the previous node; S52, after sampling, the candidate node t of the current weld point is obtained. tcp-rand Then, determine its nearest neighbor node, t. tcp-near For the current sampling point and the previous planned node The node with the minimum objective cost is expressed as: in, For the j-th node in the redundant attitude set of the previous weld point, From to t tcp-rand The total target cost; S53. After determining the nearest neighbor node, a new planned node t for the current weld point is generated using a path turning strategy. tcp-new ,but: t tcp-n =[p tcp-n ,r tcp-near +r tcp-rand ] Where, p tcp-n For the new node position, r tcp-near For the pose of the nearest neighbor node, r tcp-ran The obstacle avoidance posture of the welding torch is predicted by a nonlinear mapping model.

7. The obstacle avoidance motion planning method for intelligent process redundancy drive of welding robots according to claim 1, characterized in that, Step S6 includes the following steps: S61. Based on the CAD model, perform collision detection and incremental adjustment to perform collision detection on the newly planned nodes; S62. When a new planning node passes the collision detection, add this new planning node to the motion planning path and update the sequence number of the weld point to obtain a new motion planning path.

8. The obstacle avoidance motion planning method for intelligent process redundancy drive of welding robots according to claim 1, characterized in that, Step S7 includes the following steps: S71. The new motion planning path is optimized by combining the path optimization method of quaternion spherical linear interpolation and cosine easing time mapping. S72. Perform collision detection on each interpolation point in the optimized new motion planning path to ensure the continuity and smoothness of the motion planning trajectory.

9. An obstacle avoidance motion planning system for intelligent process redundancy driven welding robots, characterized in that, include: The acquisition module is used to acquire the weld trajectory to be planned; The modeling module is used to probabilistically model the correlation between obstacle collision boundaries and welding torch obstacle avoidance posture using an incremental extreme gradient boosting model. The first calculation module calculates candidate planning nodes based on weld seam trajectory points; The second calculation module determines the nearest neighbor node of the candidate planning node based on the target cost, and then calculates the planning node of the current weld point. The third calculation module is used for collision detection of newly planned nodes, and then the obstacle avoidance data of the new nodes is used for incremental learning of the model. The fourth calculation module, based on an optimization strategy of spherical linear interpolation and cosine easing time mapping, ensures the continuity and smoothness of the planning process trajectory.

10. A computer device, characterized in that, It includes a processor and a memory for storing processor-executable programs. When the processor executes the program stored in the memory, it implements the obstacle avoidance motion planning method for intelligent process redundancy drive of welding robots as described in any one of claims 1-8.

Images