Robot high-efficiency non-collision path planning method and system for non-standard welding

Abstract

The application discloses a kind of robot high-efficiency non-collision path planning method and system for non-standard welding piece, belong to robot welding technical field, comprising: S1, obtains the three-dimensional model data of non-standard welding piece and working environment, and extracts weld point cloud data;S2, feature extraction and down-sampling are carried out to weld point cloud data, generate a group of key path points;S3, global optimization is carried out to welding path using improved genetic algorithm, generate candidate path, and the welding sequence of multiple welds and the welding gun posture at each key path point are simultaneously optimized using improved genetic algorithm;S4, the candidate path generated by improved genetic algorithm is evaluated by a multi-objective fitness function including path length, posture stability and collision risk;S5, according to the evaluation result, output the optimal collision-free welding path, and convert it into smooth trajectory executable by robot.The application can automatically generate efficient, high-quality and collision-free welding gun movement trajectory.

Description

Technical Field

[0001] This invention belongs to the field of robotic welding technology, and in particular relates to a robot human-efficiency collision-free path planning method and system for non-standard welded parts. Background Technology

[0002] Modern manufacturing is moving towards small-batch, multi-variety, and customized production, leading to a continuous increase in the demand for welding non-standard structural components. Traditional robot welding programming mainly relies on two methods: "teach-and-playback" and "offline programming." Teach-and-playback requires experienced technicians to manually guide the robot to record path points. When dealing with complex welds, this method is not only time-consuming and labor-intensive but also struggles to ensure accuracy. While offline programming shifts the programming work from the production line to a computer, data interaction between CAD and offline programming software often suffers from gaps, requiring significant manual intervention to reset welding parameters and paths. This results in long production preparation cycles and poor flexibility.

[0003] Existing research indicates that these traditional methods, when dealing with unstructured environments and irregular workpieces, typically rely on predefined rules and templates, exhibiting insufficient flexibility and difficulty in adapting to changes in weld position and shape, thus affecting the consistency and reliability of weld quality. Furthermore, for automated path planning, the academic community has proposed various algorithms, such as the sampling-based Rapid Expanding Random Tree (RRT) and its variants (RRT*), and the graph search-based A* algorithm. However, these algorithms often face challenges when applied to welding scenarios: they typically focus on finding the geometrically shortest collision-free path, neglecting the specific requirements of the welding process for path quality, such as the smoothness of the welding torch posture, the continuity of angular velocity, and the dynamic adjustment of welding speed. Simply optimizing a single objective (such as path length) often reduces weld quality or fails to meet process constraints. Therefore, developing a multi-objective planning algorithm that can synergistically optimize efficiency, quality, and safety has become a pressing technical challenge in this field. Summary of the Invention

[0004] The purpose of this invention is to meet practical needs by providing a robot human-efficiency collision-free path planning method and system for non-standard welded parts. It integrates an improved genetic algorithm with a collaborative planning framework of mainstream robot motion planning libraries, and can automatically generate efficient, high-quality and collision-free welding torch motion trajectories for non-standard workpieces with irregular geometry and complex spatial welds.

[0005] The first objective of this invention is to provide a robot-based collision-free path planning method for non-standard welded parts, comprising:

[0006] S1. Obtain 3D model data of non-standard welded parts and working environment, and extract weld point cloud data;

[0007] S2. Perform feature extraction and downsampling on the weld seam point cloud data to generate a set of critical path points;

[0008] S3. An improved genetic algorithm is used to globally optimize the welding path and generate candidate paths. The improved genetic algorithm simultaneously optimizes the welding sequence of multiple weld seams and the welding torch posture at each critical path point.

[0009] S4. The candidate paths generated by the improved genetic algorithm are evaluated using a multi-objective fitness function that includes path length, attitude stability, and collision risk.

[0010] S5. Based on the evaluation results, output the optimal collision-free welding path and convert the optimal collision-free welding path into a smooth trajectory that can be executed by the robot.

[0011] Preferably, the feature extraction and downsampling includes: an adaptive method based on the local geometric curvature of the weld, performing sparse sampling in straight areas and dense sampling in corner areas.

[0012] Preferably, the improved genetic algorithm employs a hybrid encoded chromosome, which includes an integer arrangement portion representing the welding sequence of the weld seam and a real number or quaternion array portion representing the welding torch posture at each key point.

[0013] Preferably, during the initialization phase of the improved genetic algorithm, a sampling-based motion planning library is invoked to generate initial collision-free paths for some or all weld segments, and these paths are injected into the initial population as high-quality genes to accelerate algorithm convergence.

[0014] Preferably, the evaluation is based on modeling the environment using an Octomap octree map and performing collision detection on candidate paths.

[0015] The second objective of this invention is to provide a robot-based collision-free path planning system for non-standard welded parts, comprising:

[0016] The data acquisition module acquires 3D model data of non-standard welded parts and the working environment, and extracts weld point cloud data;

[0017] The data preprocessing module performs feature extraction and downsampling on the weld seam point cloud data to generate a set of critical path points;

[0018] The optimization module uses an improved genetic algorithm to globally optimize the welding path and generate candidate paths. The improved genetic algorithm simultaneously optimizes the welding sequence of multiple weld segments and the welding torch posture at each critical path point.

[0019] The evaluation module evaluates the candidate paths generated by the improved genetic algorithm using a multi-objective fitness function that includes path length, attitude stability, and collision risk.

[0020] The output module outputs the optimal collision-free welding path based on the evaluation results and converts the optimal collision-free welding path into a smooth trajectory that can be executed by the robot.

[0021] Preferably, it also includes a process coupling optimization model that adaptively adjusts the planned welding speed based on the local curvature of the path.

[0022] Preferably, the feature extraction and downsampling includes: an adaptive method based on the local geometric curvature of the weld, performing sparse sampling in straight areas and dense sampling in corner areas;

[0023] The improved genetic algorithm uses a hybrid encoded chromosome, which includes an integer arrangement part for representing the welding sequence of the weld seam and a real number or quaternion array part for representing the welding gun posture of each key point.

[0024] In the initialization phase of the improved genetic algorithm, a sampling-based motion planning library is called to generate initial collision-free paths for some or all weld segments, and these paths are injected into the initial population as high-quality genes to accelerate algorithm convergence.

[0025] The evaluation is based on Octomap octree map modeling of the environment and collision detection of candidate paths.

[0026] A third objective of this invention is to provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned robot-based collision-free path planning method for non-standard welded parts.

[0027] A fourth objective of this invention is to provide a computer program product, including a computer program that, when executed by a processor, implements the aforementioned efficient collision-free path planning method for robots targeting non-standard welded parts.

[0028] The advantages and positive effects of this application are:

[0029] 1) Efficiency Improvement: This invention demonstrates significant superiority in both "planning time" and "path length"—two key performance indicators. Specifically, compared to the traditional RRT algorithm, this invention, by introducing an advanced algorithm mechanism, can significantly reduce the average planning time and effectively shorten the path length. This significant efficiency improvement is mainly attributed to the global optimization capability of the Genetic Algorithm (GA) and its synergy with the open-source Motion Planning Library (OMPL). The combination of these two allows the algorithm to quickly converge to the optimal solution during the search process, thereby greatly improving the overall planning efficiency.

[0030] 2) Quality Improvement: Regarding posture smoothness, this invention ensures smooth and continuous posture changes of the robotic arm during welding through refined algorithm design and optimization, thereby effectively guaranteeing the stability and reliability of the welding process. Furthermore, this invention places special emphasis on avoiding collision risks during welding. Through precise path planning and posture control, it significantly improves the safety of the production process, reduces equipment damage and production accidents caused by collisions, and further enhances overall production quality. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a flowchart of a preferred embodiment of the present invention;

[0033] Figure 2 This is another flowchart of a preferred embodiment of the present invention;

[0034] Figure 3 This is a schematic diagram of the support leg model in a preferred embodiment of the present invention. Detailed Implementation

[0035] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] Please see Figure 1 and Figure 2

[0037] An efficient collision-free path planning method for robots designed for non-standard welded parts includes the following steps:

[0038] S1. Obtain 3D model data of non-standard welded parts and working environment, and extract weld point cloud data;

[0039] The purpose of this step is to obtain 3D model data of non-standard welded parts and their working environment. High-precision scanning technology is used to comprehensively capture the structural features of the welded parts and detailed information about the working environment. Based on this, point cloud data of the weld area is further extracted to ensure the accuracy and completeness of the point cloud data, providing reliable data support for subsequent welding process analysis and optimization.

[0040] S2. Perform feature extraction and downsampling on the weld seam point cloud data to generate a set of critical path points;

[0041] The purpose of this step is to perform a detailed and accurate feature extraction process on the weld point cloud data. By employing advanced algorithms and techniques, the geometric characteristics and structural information in the point cloud data are analyzed in depth to effectively identify the key features of the weld. Based on this, the extracted features are further subjected to efficient downsampling processing to reduce data redundancy and optimize data processing efficiency, ultimately generating a set of refined and representative critical path points. These critical path points not only accurately reflect the overall morphology and local details of the weld but also provide crucial data support for subsequent welding process optimization and quality control.

[0042] S3. An improved genetic algorithm is used to globally optimize the welding path and generate candidate paths. The improved genetic algorithm simultaneously optimizes the welding sequence of multiple weld seams and the welding torch posture at each critical path point.

[0043] In the global optimization of the welding path, this embodiment employs an improved genetic algorithm that generates a series of candidate welding paths. These candidate paths not only consider the overall optimization of the welding sequence but also finely adjust the welding sequence of multiple weld segments. Furthermore, the improved genetic algorithm can optimize the welding torch posture at each critical path point, ensuring that the welding torch operates at the most suitable angle and position during the welding process. Through such optimization, a highly efficient and high-quality welding path can be obtained, thereby improving both welding efficiency and welding quality.

[0044] S4. The candidate paths generated by the improved genetic algorithm are evaluated using a multi-objective fitness function that includes path length, attitude stability, and collision risk.

[0045] A comprehensive and detailed evaluation of candidate paths generated by the improved genetic algorithm is conducted using a multi-objective fitness function that considers multiple key factors such as path length, attitude stability, and collision risk. This fitness function aims to quantitatively analyze the merits of paths from multiple dimensions, ensuring that the selected path is not only optimal in length but also performs well in attitude maintenance and collision avoidance, thereby providing a more reliable and efficient decision-making basis for path planning.

[0046] S5. Based on the evaluation results, output the optimal collision-free welding path and convert the optimal collision-free welding path into a smooth trajectory that can be executed by the robot.

[0047] To better understand the technical solution of the present invention, the following non-limiting description is provided:

[0048] The feature extraction and downsampling are based on an adaptive method of the local geometric curvature of the weld, which performs sparse sampling in flat areas with low curvature and dense sampling in corner areas with high curvature.

[0049] The described feature extraction and downsampling method is an adaptive technique that intelligently adjusts based on the local geometric curvature of the weld. Specifically, when a local area of ​​the weld exhibits low curvature, i.e., relatively flat sections, the method employs a sparse sampling strategy. This means that in these flat areas, the density of feature extraction and downsampling is relatively low because feature changes in flat areas are not significant, requiring fewer sampling points to capture details. Conversely, when a local area of ​​the weld has high curvature, i.e., corners or other complex shapes, the method automatically adjusts to a dense sampling strategy. In corner areas with high curvature, the density of feature extraction and downsampling increases to ensure that more detailed information is captured, thereby more accurately describing the weld's features. This adaptive feature extraction and downsampling method intelligently adjusts the sampling strategy according to the actual shape and structure of the weld, ensuring both the accuracy of feature extraction and improving sampling efficiency, making it a highly intelligent and efficient technique.

[0050] The improved genetic algorithm employs a hybrid encoded chromosome, which includes an integer permutation portion representing the welding sequence of the weld seam and a real number or quaternion array portion representing the welding torch posture at each key point.

[0051] The improved genetic algorithm in this embodiment employs a hybrid-encoded chromosome, which consists of two main parts. The first part is an integer permutation representing the welding sequence. This part precisely describes the welding order through the integer sequence, ensuring the rationality and efficiency of the welding process. The second part is a real-number or quaternion array representing the welding torch posture at each key point. This part records detailed information about the welding torch posture at each key point, including key parameters such as angle and direction, ensuring that the welding torch operates precisely according to the preset posture during welding, thus improving welding quality and accuracy. Through this hybrid encoding method, the improved genetic algorithm can more comprehensively and precisely handle complex problems in the welding process, optimize the welding sequence and welding torch posture, and further enhance the overall performance and effect of the welding process.

[0052] The improved genetic algorithm employs an adaptive mutation strategy related to weld curvature, which gives the posture genes corresponding to positions with higher curvature on the path a higher mutation probability.

[0053] In the initialization phase of the improved genetic algorithm, a sampling-based motion planning library (such as OMPL) is invoked to generate high-quality initial collision-free paths for some or all weld segments, and these paths are injected into the initial population as high-quality genes to accelerate algorithm convergence.

[0054] The collision risk assessment is based on modeling the environment using an Octomap octree map and performing human-efficient collision detection on candidate paths.

[0055] It also includes a process coupling optimization that adaptively adjusts the planned welding speed based on the local curvature of the path, reducing the speed in areas of high curvature to ensure welding quality.

[0056] A high-efficiency, collision-free path planning system for robots designed for non-standard welded parts includes:

[0057] The data acquisition module acquires 3D model data of non-standard welded parts and the working environment, and extracts weld point cloud data;

[0058] The data preprocessing module performs feature extraction and downsampling on the weld seam point cloud data to generate a set of critical path points;

[0059] The optimization module uses an improved genetic algorithm to globally optimize the welding path and generate candidate paths. The improved genetic algorithm simultaneously optimizes the welding sequence of multiple weld segments and the welding torch posture at each critical path point.

[0060] The evaluation module evaluates the candidate paths generated by the improved genetic algorithm using a multi-objective fitness function that includes path length, attitude stability, and collision risk.

[0061] The output module outputs the optimal collision-free welding path based on the evaluation results and converts the optimal collision-free welding path into a smooth trajectory that can be executed by the robot.

[0062] In one specific embodiment, a process coupling optimization model is also included, which adaptively adjusts the planned welding speed based on the local curvature of the path.

[0063] In one specific embodiment, the feature extraction and downsampling includes: an adaptive method based on the local geometric curvature of the weld, performing sparse sampling in straight areas and dense sampling in corner areas;

[0064] In one specific embodiment, the improved genetic algorithm employs a hybrid encoded chromosome, which includes an integer permutation portion for representing the welding sequence of the weld seam and a real number or quaternion array portion for representing the welding gun posture at each key point.

[0065] In one specific embodiment, during the initialization phase of the improved genetic algorithm, a sampling-based motion planning library is invoked to generate initial collision-free paths for some or all weld segments, and these paths are injected into the initial population as high-quality genes to accelerate algorithm convergence.

[0066] In one specific embodiment, the evaluation is based on Octomap octree map to model the environment and perform collision detection on candidate paths.

[0067] The specific implementation process of the above system includes:

[0068] Step 1: Environment and Task Initialization: The system first loads the workpiece model, weld point cloud data, and STL models of peripheral equipment from the CAD file. An initial 3D environment containing all static obstacles is constructed within the MoveIt planning scene. The weld point cloud is resolved into a series of target path segments to be welded.

[0069] Step 2: Weld feature extraction: Apply a feature extraction algorithm to the original weld point cloud, perform adaptive downsampling, and generate a set of sparse but accurate key path points that can describe the weld geometry.

[0070] Step 3: Hybrid Initialization Population Generation: To accelerate the convergence of the genetic algorithm, a hybrid initialization strategy is adopted. High-quality, collision-free initial paths are quickly generated for the weld segment using MoveIt's OMPL planning library, and then decoded into chromosome form and injected into the initial population.

[0071] Step 4: Iterative optimization using genetic algorithm: The algorithm continuously generates new solutions (chromosomes) through selection, crossover, and mutation operations. Each solution is evaluated using a comprehensive fitness function that quantifies path length, pose smoothness, and collision risk.

[0072] Step 5: Path Generation and Verification: After the genetic algorithm iterations, the chromosome with the highest fitness is output. This optimal solution (including the welding sequence and the welding torch posture at each key point) is converted into a smooth trajectory. This trajectory will undergo final, more refined collision verification in MoveIt and be visualized.

[0073] Step 6: Output and Robot Execution: After successful verification, standard robot trajectory instructions (such as JointTrajectory messages) are generated and published to the ROS network, where the robot driver subscribes to and controls the robot to execute them precisely.

[0074] Non-standard weld feature extraction and path point downsampling:

[0075] Adaptive downsampling based on local geometric features (especially curvature) specifically includes:

[0076] For straight segments (with local curvature below the threshold ε1), large-amplitude uniform sparse sampling is performed, retaining only the key points at the end that can characterize the direction of the straight segment.

[0077] For high curvature corner segments (local curvature exceeding the threshold ε2), all original points are preserved or slightly encrypted sampling is performed to ensure that the path accurately fits the abrupt geometric changes of the weld.

[0078] For the transition curve segment (curvature between ε1 and ε2), the curvature is proportionally sampled.

[0079] In the improved genetic algorithm (GA) hybrid initialization population generation method, the chromosome encoding method includes two parts: optimizing the welding sequence and the welding torch posture.

[0080] 1) Weld sequence coding (integer permutation). Assume there are N independent weld segments on the workpiece. This part is a permutation of an integer of length N (such as 1 to N).

[0081] 2) Keypoint pose encoding (real number / quaternion array). This part is a variable-length real number array that stores the welding torch pose of all weld key points. The pose is usually represented by a quaternion q(w,x,y,z). The initial pose is determined by the normal vector information of the weld point cloud, and the genetic algorithm will fine-tune and optimize it based on this.

[0082] This embodiment proposes an adaptive mutation strategy related to weld curvature. In straight weld segments, the mutation probability should be low to maintain superior genes; at corners with large curvature, a higher mutation probability is used to encourage the algorithm to explore more possibilities.

[0083] The mutation probability P_mutation(i) of the i-th posture gene is determined by the following formula:

[0084] P_mutation(i) = P_base + k*Curvature(i), where P_base is the basic mutation rate, k is the coefficient, and Curvature(i) is the normalized curvature.

[0085] This invention improves the fitness function of the genetic algorithm (GA) by proposing a multi-objective fitness function F_total that maximizes the fitness function, as follows:

[0086] Fitness function F_total = w1*F_length + w2*F_pose + w3*F_collision

[0087] Where w1, w2, and w3 are weighting coefficients used to balance the importance of different optimization objectives. The definitions of each term are as follows:

[0088] F_length (path length): Calculates the total distance traveled by the welding torch TCP (tool center point), including the length of all welding paths and the transition path length between welds.

[0089] F_pose (Attitude Stability): Measures the smoothness of the welding torch attitude throughout the welding process.

[0090] F_collision (collision cost): When evaluating each chromosome, the collision detection module is invoked to check the generated paths. If any part of the path collides with the environment, F_collision will be assigned a very large penalty value to ensure that any solution at risk of collision is eliminated during the selection process.

[0091] Please see Figure 3 The workpiece legs are the welding object, and the welding equipment is a 6-axis welding robot;

[0092] Step 1: Environment and Task Initialization. The system first loads, such as... Figure 2 The image shows the CAD model of the workpiece support leg and the STL model of its surrounding environment. Simultaneously, pre-extracted weld point cloud data is imported, with an example of its data structure shown below: [

[0094] {Weld marking: C169, Weld length: 4.0 mm}

[0095] {Weld marking: f698, weld length: 63.0 mm}

[0096] {Weld marking: d3fe, weld length: 100.0 mm} ]

[0098] The data is parsed in MoveIt! planning scenarios to construct an initial 3D environment containing static obstacles.

[0099] Step 2: Weld Feature Extraction and Adaptive Downsampling. Adaptive downsampling based on local geometric curvature is performed on the original weld point cloud. In this example, the low curvature threshold ε1 = 0.1 and the human curvature threshold ε2 = 0.8 (after normalization) are set.

[0100] For long straight welds on the workpiece legs, whose curvature is less than ε1, the algorithm performs sparse sampling and only retains the first and last key points;

[0101] For the weld at the L-shaped corner, whose curvature is greater than ε2, the algorithm retains most of the original points to accurately describe the geometric changes, thereby generating a sparse but critical set of path points.

[0102] Step 3: Improved Genetic Algorithm Optimization. An improved genetic algorithm (GA) is used to globally optimize the welding sequence and welding torch posture. The core parameters of the algorithm are set as follows:

[0103] Hybrid initialization: The OMPL planning library is called to quickly generate high-quality, collision-free initial paths for each weld segment. After decoding, these paths are injected into the initial population as high-quality genes to accelerate convergence.

[0104] Adaptive variation strategy: Adaptive variation related to weld curvature is adopted.

[0105] The mutation probability of the i-th posture gene is determined by the following formula:

[0106] Pmutation(i)=Pbase+k*Curvature(i)

[0107] In this example, the base variation rate P_base = 0.01 and the curvature influence coefficient k = 0.05 are set. This allows the welding torch posture at corners (high curvature) to have more opportunities for exploratory variation, while it remains stable in straight sections (low curvature).

[0108] Multi-objective fitness function: A weighted sum method is used to evaluate the merits of each candidate solution (chromosome). The fitness function is as follows:

[0109] Ftotal=w1*Flength+w2*Fpose+w3*Fcollision

[0110] Where F_length is the total path length, F_pose evaluates the pose smoothness, and F_collision is the collision cost. In this embodiment, to balance efficiency, quality, and safety, the weighting coefficients are set to w1 = 0.4, w2 = 0.3, and w3 = 0.3. When a collision is detected, F_collision is given a very large penalty value to ensure that the solution is eliminated in the evolution process.

[0111] Step 4: Path Generation and Output

[0112] After a set number of iterations, the genetic algorithm outputs the chromosome with the highest fitness. This optimal solution (containing the optimal welding sequence and the welding torch posture at each key point) is converted into a smooth robot trajectory and undergoes final collision verification in MoveIt!. Upon successful verification, standard robot control commands (such as ROS JointTrajectory messages) are generated and sent to the robot for execution.

[0113] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described efficient collision-free path planning method for robots targeting non-standard welded parts.

[0114] A computer program product includes a computer program that, when executed by a processor, implements the above-described efficient collision-free path planning method for robots targeting non-standard welded parts.

[0115] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented, in whole or in part, as a computer program product, the computer program product includes one or more computer instructions. When the computer program instructions are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0116] The above description is only a preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for efficient collision-free path planning of robots for non-standard welded parts, characterized in that, include: S1. Obtain 3D model data of non-standard welded parts and working environment, and extract weld point cloud data; S2. Perform feature extraction and downsampling on the weld point cloud data to generate a set of critical path points; the feature extraction and downsampling includes: an adaptive method based on the local geometric curvature of the weld, specifically: For long straight welds on the workpiece legs, whose curvature is lower than the low curvature threshold ε1, sparse sampling is performed, and only the first and last key points are retained. For the weld at the L-shaped corner, its curvature is higher than the high curvature threshold ε2. Most of the original points are retained to accurately describe the geometric changes, thus generating a sparse but critical set of path points. The low curvature threshold ε1 is less than the high curvature threshold ε2; S3. An improved genetic algorithm is used to globally optimize the welding path and generate candidate paths. The improved genetic algorithm simultaneously optimizes the welding sequence of multiple weld seams and the welding torch posture at each critical path point. The improved genetic algorithm employs a hybrid encoded chromosome, which includes an integer permutation portion representing the welding sequence of multiple weld segments and a real number or quaternion array portion representing the welding torch posture at each critical path point. The improved genetic algorithm adopts an adaptive mutation strategy related to weld curvature, where the mutation probability Pmutation(i) of the i-th posture gene satisfies: Pmutation(i) = Pbase + k × Curvature(i), where Pbase is the basic mutation rate, k is the curvature influence coefficient, and Curvature(i) is the normalized curvature, so that the posture genes corresponding to positions with higher curvature on the path have a higher mutation probability. S4. The candidate paths generated by the improved genetic algorithm are evaluated using a multi-objective fitness function that includes path length, pose stability, and collision risk. The multi-objective fitness function is: Ftotal = w1 × Flength + w2 × Fpose + w3 × Fcollision; where Flength is the total path length, Fpose evaluates pose smoothness, and Fcollision is the collision cost. When a collision is detected, Fcollision is given a very large penalty value to ensure that the solution is eliminated in the evolution. S5. Based on the evaluation results, output the optimal collision-free welding path, convert the optimal collision-free welding path into a smooth trajectory that can be executed by the robot, and adaptively adjust the welding speed according to the local curvature of the smooth trajectory, reducing the welding speed where the local curvature is large.

2. The efficient collision-free path planning method for robots targeting non-standard welded parts according to claim 1, characterized in that, In the initialization phase of the improved genetic algorithm, a sampling-based motion planning library is invoked to generate initial collision-free paths for some or all weld segments, and these paths are injected into the initial population as high-quality genes to accelerate algorithm convergence.

3. The efficient collision-free path planning method for robots targeting non-standard welded parts according to claim 1, characterized in that, The evaluation is based on Octomap octree map modeling of the environment and collision detection of candidate paths.

4. A high-efficiency, collision-free path planning system for robots designed for non-standard welded parts, characterized in that, The method for efficient collision-free path planning of robots for non-standard welded parts as described in any one of claims 1-3 includes: The data acquisition module acquires 3D model data of non-standard welded parts and the working environment, and extracts weld point cloud data; The data preprocessing module performs feature extraction and downsampling on the weld seam point cloud data to generate a set of critical path points; The optimization module uses an improved genetic algorithm to globally optimize the welding path and generate candidate paths. The improved genetic algorithm simultaneously optimizes the welding sequence of multiple weld segments and the welding torch posture at each critical path point. The evaluation module evaluates the candidate paths generated by the improved genetic algorithm using a multi-objective fitness function that includes path length, attitude stability, and collision risk. The output module outputs the optimal collision-free welding path based on the evaluation results and converts the optimal collision-free welding path into a smooth trajectory that can be executed by the robot.

5. The efficient collision-free path planning system for robots targeting non-standard welded parts according to claim 4, characterized in that, It also includes a process coupling optimization model that adaptively adjusts the planned welding speed based on the local curvature of the path.

6. The efficient collision-free path planning system for robots targeting non-standard welded parts according to claim 4, characterized in that: The feature extraction and downsampling include: an adaptive method based on the local geometric curvature of the weld, which performs sparse sampling in straight areas and dense sampling in corner areas; The improved genetic algorithm uses a hybrid encoded chromosome, which includes an integer arrangement part for representing the welding sequence of the weld seam and a real number or quaternion array part for representing the welding gun posture of each key point. In the initialization phase of the improved genetic algorithm, a sampling-based motion planning library is called to generate initial collision-free paths for some or all weld segments, and these paths are injected into the initial population as high-quality genes to accelerate algorithm convergence. The evaluation is based on Octomap octree map modeling of the environment and collision detection of candidate paths.

7. A computer-readable storage medium storing a computer program that, when executed by a processor, implements the efficient collision-free path planning method for robots targeting non-standard welded parts as described in any one of claims 1-3.

8. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the efficient collision-free path planning method for robots targeting non-standard welded parts as described in any one of claims 1-3.

Images