Robot-assisted sheet metal bending and unloading path planning method based on SVSDF

Abstract

The application discloses a robot-assisted sheet metal bending feeding and discharging path planning method based on SVSDF, which comprises the following steps: step 1, constructing a two-dimensional pixel coordinate system; step 2, obtaining an initial motion trajectory of sheet metal feeding and discharging; step 3, calculating path safety; step 4, calculating path cost; step 5, calculating the fitness value of the initial motion trajectory of sheet metal feeding and discharging; and step 6, obtaining an optimal motion trajectory of sheet metal feeding and discharging by optimizing path cost and path safety. The application solves the problems that the RRT-Conncet algorithm is easily attracted by low-value areas and unstable convergence speed by using a synchronous biasing greedy strategy, adaptive elliptical region sampling and adaptive node growth. In addition, the quantitative path safety criterion based on SVSDF and the multi-objective optimization of the water cycle algorithm WCA balance the path cost and the safety distance from obstacles, so that the application can be applied to workpieces with different complexities.

Description

Technical Field

[0001] This invention relates to the field of sheet metal bending technology, and in particular to a robot-assisted sheet metal bending loading and unloading path planning method based on SSVSDF. Background Technology

[0002] During the bending process of sheet metal, the shape of the sheet metal changes from simple to complex as the process progresses, often resulting in non-convex geometries. In robot-assisted bending, the narrow space between the upper and lower dies of the bending machine, coupled with the constantly changing non-convex geometry of the workpiece, increases the risk of collisions.

[0003] The challenge of robot-assisted bending loading and unloading tasks lies in the confined space and varied shapes of the workpieces. Efficiency and safety are crucial in this task, and current path planning methods primarily focus on reducing path length, with limited research on balancing the safety and efficiency of object movement. Existing bending loading and unloading path planning algorithms, such as the traditional Rapid Random Tree Exploration (RRT) algorithm, still have the following shortcomings in practical application:

[0004] 1) Low computational efficiency, slow convergence speed and insufficient path smoothness.

[0005] 2) Existing algorithms either oversimplify the shapes of objects and the environment or use discrete sampling to approximate continuous motion. Simplifying the shape reduces the available space, making it impossible to generate smooth and appropriate motion trajectories within the narrow channels of the limited upper and lower molds. At the same time, discrete sampling methods may lose collision detection capabilities, thus failing to ensure continuous collision-free motion.

[0006] 3) It focuses on path efficiency but does not fully address path safety issues and lacks specific metrics for measuring path safety. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to address the shortcomings of the prior art by providing a robot-assisted sheet metal bending loading and unloading path planning method based on SSVSDF. This SSVSDF-based robot-assisted sheet metal bending loading and unloading path planning method is based on the SSVSDF quantitative path safety criterion and uses the water circulation algorithm for multi-objective optimization to balance path cost and safe distance from obstacles, thereby balancing the safety and efficiency of object movement.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0009] A robot-assisted sheet metal bending loading and unloading path planning method based on SSVSDF includes the following steps.

[0010] Step 1: Construct a 2D pixel coordinate system: Based on the binarized side view image of the upper and lower molds of the sheet metal bending machine, a 2D map plane is constructed. The upper left corner of the 2D map plane is taken as the origin, and the two adjacent right-angled sides passing through the origin are the x-axis and y-axis, respectively, thus forming a 2D pixel coordinate system. The sheet metal part is regarded as a moving object, and the bending point of the sheet metal part is regarded as the moving point. Then the loading and unloading movement path of the sheet metal part includes several path points, each of which is represented as (x, y, θ). Among them, (x, y) are the position parameters of the sheet metal part in the 2D pixel coordinate system; θ is the attitude angle of the sheet metal part.

[0011] Step 2: Obtain the initial motion trajectory for unloading and loading sheet metal parts: Using the RRT-Connect algorithm, the initial motion path for unloading and loading sheet metal parts is planned to obtain an initial motion trajectory for unloading and loading sheet metal parts; the initial planning is repeated N times. pop Then, we get N. pop Initial movement trajectory of sheet metal parts during loading and unloading.

[0012] Step 3: Calculate path safety: Identify all pixels in the upper and lower dies of the sheet metal bending machine that are likely to collide with the bending motion of the sheet metal part or are within the safe distance, and mark them as obstacles of interest. Based on the obstacles of interest and the initial motion trajectory of the sheet metal part loading and unloading, and based on the swept volume signed distance field SVSDF, calculate the path safety value of each initial motion trajectory of the sheet metal part loading and unloading.

[0013] Step 4: Calculate Path Cost: Assuming any initial movement trajectory for loading and unloading sheet metal parts has N path points, the specific formula for calculating the path cost is as follows:

[0014]

[0015] In the formula, Δx i and Δy i Let i be the difference in horizontal and vertical position parameters between two adjacent path points i and i+1; where 1≤i≤N-1.

[0016] Δθ i The difference in the attitude angle of the sheet metal parts between two adjacent path points i and i+1.

[0017] β is the attitude angle weighting coefficient, which is a set value;

[0018] Step 5: Calculate the COST t Calculate the fitness value (COST) of the initial motion trajectory for loading and unloading the t-th sheet metal part. t Where 1≤t≤N pop Then COST t The calculation formula is:

[0019] COST t =γ1path cost+γ2path safety

[0020] In the formula, γ1 is the cost weighting coefficient, and 1≤γ1≤1.5 is a set value.

[0021] γ2 is the safety weight coefficient, and 1≤γ2≤1.5 is a set value.

[0022] Step 6: Obtain the optimal motion trajectory for unloading and loading sheet metal parts: Based on the Water Circulation Algorithm (WCA), continuously iterate and optimize to find N in Step 5. pop Cost t Find the minimum value and set the minimum COST. t The corresponding sheet metal part loading and unloading trajectory is taken as the optimal loading and unloading trajectory for sheet metal parts.

[0023] In step 2, the RRT-Connect algorithm is an improved synchronous bias greedy RRT-Connect algorithm. When initially planning the loading and unloading movement path of sheet metal parts, the improved synchronous bias greedy RRT-Connect algorithm uses an adaptive elliptical region to sample each random point and performs adaptive growth of each search tree node based on the influence of obstacles.

[0024] Step 2, which uses an adaptive elliptical region to sample each random point, includes the following steps:

[0025] Step 2A-1: Establish an adaptive elliptical region: Set the starting point S of the sheet metal part loading / unloading motion path. start (x start ,y start ,θ start ) and target point S goal (x goal ,y goal ,θ goal The two foci of the adaptive elliptical region are defined as follows: the length l from the upper die to the lower die of the bending machine is defined as the minor axis length c of the adaptive elliptical region. short This allows for the construction of an adaptive elliptical region that changes adaptively with the starting and target points.

[0026] Step 2A-2: Construct a unit circle: Using the origin of the two-dimensional pixel coordinate system in Step 1 as the center, construct a unit circle with a radius of 1 pixel.

[0027] Step 2A-3, Random Point Stretch: Randomly generate a point (x) within the unit circle. c ,y c ,θ rand ), and stretch it to a point (x, y, θ) within an adaptive elliptical region.rand ); where the formulas for calculating x and y are:

[0028]

[0029] In the formula, c long To adapt to the major axis length of the elliptical region, based on c short The focal length between the two focal points is calculated.

[0030] Step 2A-4, Random Point Rotation: Based on the two foci of the adaptive elliptical region, rotate the point (x, y, θ) rand If the point is rotated, the rotated point will be the random sampling point S. rand (x rand ,y rand ,θ rand ); where x rand and y rand The calculation formula is:

[0031]

[0032] Where R is the rotation matrix, its expression is:

[0033]

[0034] In the formula, α represents the point (x, y, θ). rand The rotation angle of ) is calculated using the following formula:

[0035] α=arctan(|x start -x goal | / |y start -y goal |).

[0036] Step 2, the adaptive growth method for each search tree node, includes the following steps:

[0037] Step 2B-1: Calculate the expansion direction v0 of the search tree node without the influence of obstacles. The specific calculation formula is as follows:

[0038]

[0039] In the formula, x nearest and y nearest For the closest random sampling point S rand Spanning tree node S nearest Position parameters.

[0040] Step 2B-2: Find the node S that influences the newly generated search tree. new The total number of obstacles N o : Using random sampling points S randCentered on a square with a side length of a set number of pixels 'a', construct a square. The total number of pixels where the square intersects with either the upper or lower die of the bending machine is denoted as the number of pixels affecting the newly generated search tree node S. new (x new ,y new ,θ new The total number of obstacles N) o .

[0041] Step 2B-3: Calculate obstacle pair S new Direction of influence v obs The specific calculation formula is as follows:

[0042]

[0043] In the formula, x k and y k Let N be the pixel coordinates of the k-th obstacle; where 1≤k≤N0.

[0044] Step 2B-4: Calculate the expansion direction v of the search tree node affected by obstacles. The specific calculation formula is as follows:

[0045] v=γ3v0+γ4v obs

[0046] in:

[0047] γ3+γ4=1

[0048] In the formula, γ3 and γ4 are the set weighting coefficients.

[0049] Step 2B-5, Calculate x new and y new : with distance S rand Recent S nearest To expand the starting point, a set equidistant step size r is adopted. p Extend in the direction of v; where x new and y new The specific calculation formula is as follows:

[0050]

[0051] Step 2B-5, Calculate θ new : with θ nearest Using the set constant angular step size r as the initial value. o To θ rand To expand upon this, the specific calculation formula is as follows:

[0052] θ new =r o (θ rand -θ nearest )+θ nearest

[0053] In the formula, θ nearest For the closest random sampling point S rand The generated search tree node S nearest The posture angle of the sheet metal parts.

[0054] Determine if a collision occurs during the expansion process. If a collision occurs, resample; otherwise, add a new node S. new Add to the search tree.

[0055] Step 3, the method for calculating path safety, specifically includes the following steps:

[0056] Step 3-1: Calculate the swept volume: Using the Shape function, calculate the volume from the starting point S for any initial motion trajectory of the sheet metal part during loading and unloading. start Move to target point S goal The total number of pixels occupied at that time is recorded as the sweep volume.

[0057] Step 3-2: Construct a sweep volume binary image: Use the Map function to input the sweep volume obtained in Step 3-1 into a blank two-dimensional map plane and mark it as black, thereby forming a sweep volume binary image.

[0058] Step 3-3: Construct the SVSDF matrix: Calculate the SVSDF value of the sweep volume for the coordinates of any pixel in the sweep volume binary image constructed in Step 3-2, thereby forming the SVSDF matrix.

[0059] Steps 3-4: Finding obstacles of interest: The upper and lower dies of the sheet metal bending machine are obstacles in the bending motion of the sheet metal parts. From the two-dimensional map plane constructed in Step 1, find all pixels in the upper and lower dies that are likely to collide with the bending motion of the sheet metal parts or are within a safe distance. Mark them as obstacles of interest and record the pixel coordinates of each obstacle of interest.

[0060] Step 3-5: Obtain the SSVSDF value of the obstacle point of interest: Based on the pixel coordinates of the obstacle point of interest in Step 3-4, find the corresponding SSVSDF value from the SSVSDF matrix constructed in Step 3-3, thereby obtaining the SSVSDF value of each obstacle point of interest.

[0061] Steps 3-6: Calculate the average SSVSDF value of the obstacle points of interest, average_sdf.

[0062] Steps 3-7: Use the normal distribution to determine the location of average_sdf and the probability density value pdf_value.

[0063] Step 3-8: Calculate path safety, specifically: path safety = K × pdf_value; where K is an order of magnitude adjustment factor.

[0064] In step 3-1, the initial motion trajectory of the t-th sheet metal part includes N path points, namely: starting point S start The second path point S2, the third path point S3, ..., the i-th path point S i The (N-2)th path point S N-2 and target point S goal Where 2≤i≤N-2; then the path point S i The swept volume is: Shape(S) i ).

[0065] In step 3-2, the swept volume binary image SV_map of the initial motion trajectory of the t-th sheet metal part during loading and unloading is:

[0066]

[0067] In step 3-3, the formula for calculating SSVSDF is:

[0068]

[0069] In the formula, d(SV_map) is the distance from any pixel in the swept volume binary image to the nearest pixel in the swept volume, in pixels.

[0070] It is the complement of d(SV_map).

[0071] In steps 3-4, before finding the obstacle of interest, the upper and lower dies of the sheet metal bending machine are divided into narrow and non-narrow regions. The narrow region refers to the area between the bottom protrusion of the upper die and the groove of the lower die; the non-narrow region refers to the other bending movement areas besides the narrow region. Therefore, the obstacle of interest includes the obstacle of interest in the narrow region and the obstacle of interest in the non-narrow region.

[0072] The obstacle points of interest in the narrow region include pixels located at the bottom outer edge and the outer ring of the upper mold within the narrow region, and pixels located at the top outer edge and the outer ring of the lower mold within the narrow region.

[0073] Obstacles of interest in non-narrow areas include pixels on the outer edge of the upper or lower mold that are located in non-narrow areas and are prone to collision with sheet metal parts.

[0074] In steps 3-6, average_sdf includes the average SVSDF value of obstacle points of interest in narrow regions (average_sdf1) and the average SVSDF value of obstacle points of interest in non-narrow regions (average_sdf2).

[0075] In steps 3-7, pdf_value includes the probability density value of the narrow region pdf_value_narrow and the probability density value of the non-narrow region pdf_value__nonnarrow. The specific calculation formula is as follows:

[0076]

[0077] In the formula, μ1 is the expected safe distance in the narrow area, and is a set value.

[0078] μ2 is the expected safe distance in non-narrow areas, which is a set value.

[0079] σ1 is the width adjustment value of the normal distribution curve corresponding to the SSVSDF value of the obstacle point of interest in the narrow region, which is a set value;

[0080] σ2 is the width adjustment value of the normal distribution curve corresponding to the SSVSDF value of the obstacle point of interest in the non-narrow region, which is a set value.

[0081] This invention offers the following advantages: First, it addresses the issues of the RRT-Connect algorithm being susceptible to low-value regions and exhibiting unstable convergence speed by employing a synchronous bias greedy strategy, adaptive elliptical region sampling, and adaptive node growth. This improvement reduces the number of iterations and search time by 51.91% and 67.6%, respectively. Second, it introduces a quantitative path safety criterion based on the workpiece motion sweep volume symbolic distance field (SVSDF). A water cycle algorithm (WCA) is used for multi-objective optimization, balancing path cost and safe distance from obstacles. Experimental results demonstrate that this method is applicable to workpieces of varying complexity. Finally, the effectiveness of the path is verified on a digital twin-based simulation platform and physical entities. Attached Figure Description

[0082] Figure 1 This is a schematic diagram illustrating the framework of the robot-assisted sheet metal bending loading and unloading path planning method based on SSVSDF of this invention.

[0083] Figure 2 This diagram illustrates the use of an adaptive elliptical region for random point sampling in the improved SBG-RRT-Connect algorithm of this invention.

[0084] Figure 3 This is a map showing the region affected by obstacles during adaptive node growth in the improved SBG-RRT-Connect algorithm of this invention.

[0085] Figure 4 This is a schematic diagram illustrating the adaptive node growth process in the improved SBG-RRT-Connect algorithm of this invention.

[0086] Figure 5 This is a schematic diagram of the path interpolation results of the present invention; wherein, (a) is a schematic diagram of position parameter interpolation; and (b) is a schematic diagram of sheet metal part attitude angle interpolation.

[0087] Figure 6 This is a schematic diagram of the sweep volume of the workpiece motion in this invention.

[0088] Figure 7 The image stored in this invention is an SVSDF image.

[0089] Figure 8 The diagrams are for path planning experiments. (a), (b), and (c) are experimental diagrams when the orientation of the workpiece remains unchanged at the starting and target positions, using the traditional RRT-Connect algorithm, the conventional SBG-RRT-Connect algorithm, and the improved SBG-RRT-Connect algorithm, respectively. (d), (e), and (f) are experimental diagrams when the orientation of the workpiece changes significantly between the starting and target positions, using the traditional RRT-Connect algorithm, the conventional SBG-RRT-Connect algorithm, and the improved SBG-RRT-Connect algorithm, respectively.

[0090] Figure 9 The figure shows a comparison of the repeatability test results when using three algorithms; where (a) is the path cost; (b) is the search time; (c) is the number of iterations; and (d) is a magnified view of a portion of the figure in (c).

[0091] Figure 10 A comparison chart showing the WCA algorithm before and after multi-objective path optimization.

[0092] Figure 11 This provides the simulation and physical verification path for the present invention.

[0093] Figure 12 This invention provides digital twin virtual simulation and physical verification. (a), (b), (c), (d), (e), and (f) show six states of the bending robot during a material unloading operation, where (b) and (c) are positions where the sheet metal part is close to the bending die and is prone to collision. Detailed Implementation

[0094] The present invention will now be described in further detail with reference to the accompanying drawings and specific preferred embodiments.

[0095] In the description of this invention, it should be understood that the terms "left side," "right side," "upper part," "lower part," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. "First," "second," etc., do not indicate the importance of the components, and therefore should not be construed as a limitation of this invention. The specific dimensions used in this embodiment are only for illustrating the technical solution and do not limit the scope of protection of this invention.

[0096] like Figure 1 As shown, the robot-assisted sheet metal bending loading and unloading path planning method based on SSVSDF includes the following steps.

[0097] Step 1: Construct a 2D pixel coordinate system: Based on the binarized side view image of the 3D model of the upper and lower dies of the sheet metal bending machine, construct a 2D map plane. Take the upper left corner of the 2D map plane as the origin, and the two adjacent right-angled sides passing through the origin as the x-axis and y-axis, thus forming a 2D pixel coordinate system. Consider the sheet metal part as a moving object, and the bending point of the sheet metal part as the moving point. Then the loading and unloading movement path of the sheet metal part includes several path points, each of which is represented as (x, y, θ). Wherein, (x, y) are the position parameters of the sheet metal part in the 2D pixel coordinate system; θ is the attitude angle of the sheet metal part; the y-axis is parallel to the vertical plane where the upper and lower dies of the sheet metal bending machine are located.

[0098] Step 2: Obtain the initial motion trajectory for unloading and loading sheet metal parts: Using the RRT-Connect algorithm, the initial motion path for unloading and loading sheet metal parts is planned to obtain an initial motion trajectory for unloading and loading sheet metal parts; the initial planning is repeated N times. pop Then, we get N. pop Initial movement trajectory of sheet metal parts during loading and unloading.

[0099] The RRT-Connect algorithm mentioned above is preferably the improved synchronous bias greedy RRT-Connect algorithm (also known as the improved SBG-RRT-Connect algorithm). When the improved synchronous bias greedy RRT-Connect algorithm initially plans the loading and unloading movement path of sheet metal parts, it uses an adaptive elliptical region to sample each random point and performs adaptive growth of each search tree node based on the influence of obstacles.

[0100] The advantage of RRT-Connect is that it is likely to greedily grow the tree towards the target location, resulting in better performance in terms of computation time. The disadvantage is that the greedy strategy often gets bogged down in searching regions far from the target. Meanwhile, the biased RRT algorithm has a probability of expanding towards the target, but the expansion only occurs once.

[0101] To address this issue, we propose an SBG-RRT-Connect algorithm that incorporates a synchronous bias greedy strategy to improve path planning performance in bending loading and unloading tasks.

[0102] The main improvement of the algorithm lies in the random tree T at the target point. b The greedy expansion strategy is triggered when a randomly selected sampling point is chosen as the relative target point. This means that the tree expands greedily in a straight line towards the target point with a step size of `step` until an obstacle is encountered or the two search trees are connected. Otherwise, the tree is expanded only once. The synchronously biased greedy strategy can be interpreted as gradient descent search with probability `p` and uniform exploration with probability `1-p`. The former aims to improve path quality by reducing search time, while the latter prevents the algorithm from getting trapped in local minima.

[0103] In the bending operation space, the distribution of obstacles is more fixed, mainly the upper and lower dies of the bending machine, thus limiting the space for workpiece movement. When using robot-assisted handling, path feasibility is further constrained. To address these issues, we propose an improved SBG-RRT-Connect algorithm with the following two improvements:

[0104] I. Adaptive Elliptical Region Sampling

[0105] The method described above for sampling each random point using an adaptive elliptical region preferably includes the following steps.

[0106] Step 2A-1: Establish an adaptive elliptical region

[0107] like Figure 2 As shown, the starting point S of the sheet metal part loading and unloading movement path is... start (x start ,y start ,θ start ) and target point S goal (x goal ,y goal ,θ goal ), which are the two foci of the adaptive elliptical region c. focal The length l from the upper die to the lower die of the bending machine is taken as the minor axis length c of the adaptive elliptical region. short This allows for the construction of an adaptive elliptical region that changes adaptively with the starting and target points.

[0108] Step 2A-2: Construct a unit circle: Using the origin of the two-dimensional pixel coordinate system in Step 1 as the center, construct a unit circle with a radius of 1 pixel.

[0109] Step 2A-3, Random Point Stretch: Randomly generate a point (x) within the unit circle. c,y c ,θ rand ), and stretch it to a point (x, y, θ) within an adaptive elliptical region. rand ); where the formulas for calculating x and y are:

[0110]

[0111] In the formula, c long To adapt to the major axis length of the elliptical region, based on c short The focal length between the two focal points is calculated.

[0112] Step 2A-4, Random Point Rotation: Based on the two foci of the adaptive elliptical region, rotate the point (x, y, θ) rand If the point is rotated, the rotated point will be the random sampling point S. rand (x rand ,y rand ,θ rand ); where x rand and y rand The calculation formula is:

[0113]

[0114] Where R is the rotation matrix, its expression is:

[0115]

[0116] In the formula, α represents the point (x, y, θ). rand The rotation angle of ) is calculated using the following formula:

[0117] α=arctan(|x start -x goal | / |y start -y goal |).

[0118] II. Adaptive Node Growth

[0119] like Figure 4 As shown, the adaptive growth method for each search tree node preferably includes the following steps.

[0120] Step 2B-1: Calculate the tree node expansion direction v0 without the influence of obstacles. The specific calculation formula is as follows:

[0121]

[0122] In the formula, x nearest and y nearest For the closest random point S rand Spanning tree node S nearest Position parameters.

[0123] In bending and blanking tasks, the starting point is usually close to the obstacle, so the influence of the upper and lower dies of the bending machine on the growth direction of the node must be considered.

[0124] Step 2B-2: Find the node S that influences the newly generated tree. new The total number of obstacles N o

[0125] like Figure 3 As shown, with random sampling point S rand Centered on a square with a side length of a set number of pixels a (preferably 5 pixels); the total number of pixels that intersect the square with the upper or lower die of the bending machine is denoted as the number of nodes affecting the newly generated tree node S. new (x new ,y new ,θ new The total number of obstacles N) o .

[0126] exist Figure 3 In the diagram, white represents open areas, and gray represents obstacles. The area susceptible to obstacles (green) spans a 5×5 square centered on the sampling point. Therefore, obstacles affecting the direction are represented by the yellow area.

[0127] Step 2B-3: Calculate obstacle pair S new Direction of influence v obs The specific calculation formula is as follows:

[0128]

[0129] In the formula, x k and y k Let N be the pixel coordinates of the k-th obstacle; where 1≤k≤N0.

[0130] Step 2B-4: Calculate the node expansion direction v affected by obstacles. The specific calculation formula is as follows:

[0131] v=γ3v0+γ4v obs

[0132] in:

[0133] γ3+γ4=1

[0134] In the formula, γ3 and γ4 are the set weighting coefficients.

[0135] Step 2B-5, Calculate x new and y new : with distance S rand Recent S nearest To expand the starting point, a set equidistant step size r is adopted.p Extend in the direction of v; where x new and y new The specific calculation formula is as follows:

[0136]

[0137] Step 2B-5, Calculate θ new : with θ nearest Using the set constant angular step size r as the initial value. o To θ rand To expand upon this, the specific calculation formula is as follows:

[0138] θ new =r o (θ rand -θ nearest )+θ nearest

[0139] In the formula, θ nearest For the closest random sampling node S rand Spanning tree node S nearest The posture angle of the sheet metal parts.

[0140] Determine if a collision occurs during the expansion process. If a collision occurs, resample; otherwise, add a new node S. new Add to the search tree.

[0141] If the starting point is the search tree T a Search tree T with target point b If the two search trees are connected, the initial path is obtained by backtracking. If they are not connected, the two search trees continue to alternately repeat the search and expansion process until they are connected.

[0142] In this embodiment, due to the random tree T of the target point b The synchronous bias greedy expansion strategy is used when the nodes S of the adaptive elliptical region are randomly sampled. rand (x rand ,y rand ,θ rand Not relative to the target point S start (x start ,y start ,θ start When ), the adaptive growth or expansion of each search tree node is performed only once; when the nodes S randomly sampled in the adaptive elliptical region rand (x rand ,y rand ,θ rand () represents the relative target point S start (x start ,y start ,θ startWhen this happens, each path node can be adaptively grown or expanded multiple times until it connects with another random tree T. a Until they connect or collide.

[0143] III. Path Cubic Spline Interpolation

[0144] The discreteness of the path derived from the sampling algorithm may lead to the omission of collision detection for some potential states. Path interpolation is performed using spline interpolation (a method that fits data points using piecewise cubic polynomials). This method also helps in the subsequent calculation of the sweep volume; the interpolated path is as follows: Figure 5 As shown.

[0145] Step 3: Calculate path safety: Identify all pixels in the upper and lower dies of the sheet metal bending machine that are likely to collide with the bending motion of the sheet metal part or are within the safe distance, and mark them as obstacles of interest. Based on the obstacles of interest and the initial motion trajectory of the sheet metal part loading and unloading, and based on the swept volume signed distance field SVSDF, calculate the path safety value of each initial motion trajectory of the sheet metal part loading and unloading.

[0146] The method for calculating path safety as described above includes the following steps.

[0147] Step 3-1: Calculate the swept volume: Using the Shape function, calculate the volume from the starting point S for any initial motion trajectory of the sheet metal part during loading and unloading. start Move to target point S goal The total number of pixels occupied at that time is recorded as the sweep volume, such as... Figure 6 As shown.

[0148] Let the initial motion trajectory of the t-th sheet metal part include N path points, namely: starting point S start The second path point S2, the third path point S3, ..., the i-th path point S i The (N-2)th path point S N-2 and target point S goal Where 2≤i≤N-2; then the sheet metal part at path point S i The number of pixels occupied by the time is: Shape(S) i ).

[0149] Step 3-2: Construct a sweep volume binary image: Use the Map function to input the sweep volume obtained in Step 3-1 into a blank two-dimensional map plane and mark it as black, thereby forming a sweep volume binary image.

[0150] The swept volume binary image SV_map of the initial motion trajectory of the t-th sheet metal part during loading and unloading is as follows:

[0151]

[0152] Step 3-3: Construct the SSVSDF matrix: For any pixel coordinate in the swept volume binary image constructed in Step 3-2, calculate the SSVSDF value up to the swept volume, thus forming a matrix as shown below. Figure 7 The SVSDF matrix shown is an L×W matrix containing SDF data, where L is the length of the binary image and W is the width of the binary image.

[0153] The preferred formula for calculating SSVSDF is as follows:

[0154]

[0155] In the formula, d(SV_map) is the distance from any pixel in the swept volume binary image to the nearest pixel in the swept volume, in pixels.

[0156] It is the complement of d(SV_map).

[0157] Steps 3-4: Locate obstacles of interest

[0158] Before searching for points of interest, the upper and lower dies of the sheet metal bending machine are divided into narrow and non-narrow areas. The narrow area refers to the area between the bottom protrusion of the upper die and the groove of the lower die; the non-narrow area refers to the other bending movement areas besides the narrow area.

[0159] The upper and lower dies of the sheet metal bending machine are obstacles in the bending motion of the sheet metal parts. From the two-dimensional map plane constructed in step 1, find all pixels in the upper and lower dies that are likely to collide with the bending motion of the sheet metal parts or are within a safe distance, mark them as obstacles of interest, and record the pixel coordinates of each obstacle of interest.

[0160] The aforementioned obstacle points of interest include obstacle points in narrow regions and obstacle points in non-narrow regions.

[0161] The obstacle points of interest in the narrow region include pixels located at the bottom outer edge and the outer ring of the upper mold within the narrow region, and pixels located at the top outer edge and the outer ring of the lower mold within the narrow region.

[0162] Obstacles of interest in non-narrow areas include pixels on the outer edge of the upper or lower mold that are located in non-narrow areas and are prone to collision with sheet metal parts.

[0163] Step 3-5: Obtain the SSVSDF value of the obstacle point of interest: Based on the pixel coordinates of the obstacle point of interest in Step 3-4, find the corresponding SSVSDF value from the SSVSDF matrix constructed in Step 3-3, thereby obtaining the SSVSDF value of each obstacle point of interest.

[0164] Steps 3-6: Calculate the average SSVSDF value of the obstacle points of interest, average_sdf.

[0165] The above-mentioned average_sdf preferably includes the average SVSDF value of the obstacle points of interest in the narrow region, average_sdf1, and the average SVSDF value of the obstacle points of interest in the non-narrow region, average_sdf2.

[0166] Steps 3-7: Use the normal distribution to determine the location of average_sdf and the probability density value pdf_value.

[0167] The aforementioned pdf_value preferably includes the probability density value pdf_value_narrow for narrow regions and the probability density value pdf_value__nonnarrow for non-narrow regions. The specific calculation formula is as follows:

[0168]

[0169] In the formula, μ1 is the expected safe distance in the narrow area, and is a set value.

[0170] μ2 is the expected safe distance in non-narrow areas, which is a set value.

[0171] σ1 is the width adjustment value of the normal distribution curve corresponding to the SSVSDF value of the obstacle point of interest in the narrow region, which is a set value.

[0172] σ2 is the width adjustment value of the normal distribution curve corresponding to the SSVSDF value of the obstacle point of interest in the non-narrow region, which is a set value.

[0173] Step 3-8: Calculate path safety, specifically: path safety = K × pdf_value; where K is an order of magnitude adjustment coefficient. By adjusting the value of K, the path safety and the path cost described below can be on the same order of magnitude. In this embodiment, K = 200 is preferred.

[0174] Step 4: Calculate path cost: Assuming any initial movement trajectory for loading and unloading sheet metal parts has N path points, the specific formula for calculating path cost is as follows:

[0175]

[0176] In the formula, Δx i and Δy i Let i be the difference in horizontal and vertical position parameters between two adjacent path points i and i+1; where 1≤i≤N-1.

[0177] Δθ i The difference in the attitude angle of the sheet metal parts between two adjacent path points i and i+1.

[0178] β is the attitude angle weighting coefficient, which is a set value.

[0179] Step 5: Calculate the COST t Calculate the fitness value (COST) of the initial motion trajectory for loading and unloading the t-th sheet metal part. t Where 1≤t≤N pop Then COST t The calculation formula is:

[0180] COST t =γ1path cost+γ2path safety

[0181] In the formula, γ1 is the cost weighting coefficient, and 1≤γ1≤1.5 is a set value. In this embodiment, the preferred value is 1.

[0182] γ2 is a safety weighting coefficient, and 1≤γ2≤1.5 is a set value. In this embodiment, the preferred value is 1.2.

[0183] Step 6: Obtain the optimal motion trajectory for unloading and loading sheet metal parts: Based on the Water Circulation Algorithm (WCA), continuously iterate and optimize to find N in Step 5. pop Cost t Find the minimum value and set the minimum COST. t The corresponding sheet metal part loading and unloading trajectory is taken as the optimal loading and unloading trajectory for sheet metal parts.

[0184] Step 7: Experimental verification.

[0185] like Figure 8 As shown, two sets of experiments were conducted. The task of the experiments was to plan an unloading path for the workpiece after a second bending operation. The map size was 1400×1400 pixels, and the angles were in radians. Two sets of experiments were designed. In the first set of experiments, the initial and target postures of the workpiece were set to be the same, i.e., S. start (916,730,0) and S goal (660,730,0). In the second group of experiments, the initial and target postures of the workpiece changed significantly. Setting S... start (916,730,0) and S goal(650, 900, 0.783). The displacement step size is 5 pixels, and the rotation step size is 0.01744.

[0186] 1) RRT-Connect planning results

[0187] Greedy strategies may increase the number of iterations and path costs of the algorithm, especially when the random sampling points are significantly deviated from the target location, such as... Figure 8 As shown in (a) and (d).

[0188] 2) SBG-RRT-Connect planning results

[0189] The introduced synchronous bias greedy strategy specifies the triggering condition for the greedy search, restricting it to the selection of randomly sampled points as target points. This optimization focuses the algorithm on the target region, accelerating convergence to an effective path. Figure 8 As shown in (b) and (e), the algorithm outperforms RRT-Connect.

[0190] Uncontrolled random sampling can lead to random points consistently being overlooked as target points, significantly reducing search efficiency. In such cases, the algorithm expands only one node at a time, rather than employing a greedy strategy. Furthermore, the unrestricted sampling area may result in points far from the target or close to obstacles. Heuristic adjustments to the node expansion direction are needed to reduce unnecessary exploration, which impacts path planning efficiency and quality in practical applications.

[0191] 3) Improved SBG-RRT-Connect planning results

[0192] from Figure 8 As can be seen in (b) and (d), when uncontrolled random sampling leads to many non-target points, the adaptive elliptical region sampling strategy introduced in the improved synchronous bias greedy RRT-Connect algorithm heuristically and effectively restricts the range of random points to a specific region. This strategy better controls the distribution of random sampling points. Simultaneously, when approaching obstacles, adjusting the expansion direction of tree nodes heuristically adjusts the current search state, significantly reducing the number of explorations in unnecessary directions, thereby further improving search efficiency. In the second set of experiments, as... Figure 8 As shown in (f), a single simulation requires only 114 iterations, compared to Figure 8 The 14160 iterations in (d) and Figure 8 This is a substantial improvement compared to the 846 iterations in (e). Figure 8 The computation time in (f) is significantly reduced to 1.1 seconds, compared to Figure 8 (d) and Figure 8 Compared to the timeframe in (e), this demonstrates higher planning efficiency. Furthermore, Figure 8 The path cost in (f) is 367.74, which, compared to the costs of 1920.74 and 989.451 in the previous figure, indicates that the path is shorter and smoother. The comparison results of the first group of experiments are the same.

[0193] The two sets of experiments were repeated 50 times, and the experimental results were as follows: Figure 9 As shown.

[0194] The improved algorithm has an average path cost of 410.08, which is 81.33% lower than the RRT-Connect algorithm and 21.14% lower than the SBG-RRT-Connect algorithm. Figure 9 As shown in (a), the blue line exhibits better stability in multiple experiments, and its path cost is significantly reduced compared to the black and red lines.

[0195] In terms of the number of iterations, the improved algorithm has an average of 189 iterations, which is 98.17% lower than the RRT-Connect algorithm and 51.91% lower than the SBG RRT Connect method. Figure 9 (c) and (d) show that the RRT-Connect algorithm (represented by the black line) fluctuates due to its random search nature and unrestricted greedy strategy, leading to unnecessary exploration. While the SBG-RRT-Connect algorithm (red line) reduces the number of iterations through a synchronous bias greedy strategy, it lacks stability. In contrast, the improved algorithm (blue line) achieves fewer iterations and exhibits more stable performance. Figure 9 (b) shows that the average search time of the improved algorithm is 3.7 seconds, which is 91.36% more efficient than the RRT-Connect algorithm and 67.6% more efficient than the SBG-RRT-Connect method.

[0196] The results show that the improved SBG-RRT-Connect performs better in terms of path cost, number of iterations, and path search time. The inheritance and improvement of this algorithm make it more stable in bending processes.

[0197] Comparison table of data from 50 repeated experiments of three algorithms

[0198] algorithm RRT-Connect SBG-RRT-Connect Improved SBG-RRT-Connect Path cost (min) 389.53 371.97 366.1 Path cost (max) 4199.47 1584.69 559.47 Path cost (avg) 2196.95 520.02 410.08 Time (min) / s 1.45 1.18 0.96 time (max) / s 136.44 50.15 11.55 Time (avg) / s 42.84 11.42 3.7 Number of iterations (min) 673 106 99 Number of iterations (max) 40046 2025 472 Number of iterations (avg) 10313 393 189

[0199] In the first set of experiments, we used a workpiece that already contained a bend. The path planning task involved feeding the workpiece to a designated bend position to await a second bend. Figure 10 (a) shows the initial path planned by the SBG-RRT-Connect algorithm, while Figure 10(d) shows the path after multi-objective optimization using WCA. The data shows that the path cost has been further reduced from 630 to 520. The average SDF value of obstacles of interest in narrow areas decreased from 9.16 to 8.04, with a target value of 8, preventing the path from being too conservative. In non-narrow areas, the average SDF value of obstacles of interest increased from 17.3 to 29.98, with a target value of 30, significantly improving road safety.

[0200] In the second set of experiments, we planned for the workpiece to undergo an unloading motion after completing the second bend. Figure 10 (b) and (e) show the initial and optimized paths, respectively. The results show that the path cost decreased from 694 to 580. The average SDF value of obstacles of interest in narrow areas increased from 2.71048 to 5.19, with a target value of 5. In non-narrow areas, the average SDF value of obstacles of interest increased from 12.41 to 18.08, with a target value of 18. Both path cost and safety were significantly improved.

[0201] The third set of experiments planned the unloading motion of the workpiece after its third bend. At this point, the workpiece has three segments on the right side of the bend point, making its shape more complex. Furthermore, the path must pass through the narrow channel between the upper and lower dies, making the initial path planned by the optimized SBG-RRT-Connect algorithm more tortuous, such as... Figure 10 As shown in (c), after optimization, the path cost decreased from 1047 to 878. The average SDF value of obstacles of interest in narrow regions increased from 1.88 to 4.88, with a target value of 5. In non-narrow regions, the average SDF value of obstacles of interest increased from 11.27 to 18.01, with a target value of 18, as shown in (c). Figure 10 As shown in (f), both path cost and security have been significantly improved.

[0202] By comprehensively considering path cost and path safety, a water-cycle algorithm is used for multi-objective path planning. Results show that this path optimization method significantly improves path quality regardless of the workpiece's shape, whether simple or complex. From the perspective of path cost and safety, firstly, path cost is further reduced, and the path becomes smoother. Furthermore, this method enables the path's safety indicators to reach the set expected values, ensuring path safety without making the path planning overly conservative.

[0203] like Figure 11 and 12 As shown in the figure, simulation verification demonstrates that the method of the present invention can be applied to sheet metal bending processes.

[0204] In summary, this invention combines path planning algorithms, intelligent algorithms, and computer graphics techniques to propose a hierarchical path planning method. This method plans the initial path using an improved SBG-RRT-Connect algorithm, then establishes a quantitative standard for evaluating path safety based on SSVSDF, and finally utilizes the WCA algorithm for multi-objective optimization of the initial path. Results show that this method can generate shorter and smoother paths while still meeting specific safety indicators. Finally, we validated the path on both virtual and physical platforms, demonstrating the effectiveness of the method. Theoretically, given a sufficiently high degree of freedom for the robot, this method can reasonably satisfy full-body path planning for workpieces of arbitrary shapes.

[0205] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details of the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A robot-assisted sheet metal bending loading and unloading path planning method based on SSVSDF, characterized in that: Includes the following steps: Step 1: Construct a 2D pixel coordinate system: Based on the binarized side view image of the upper and lower molds of the sheet metal bending machine, a 2D map plane is constructed. The upper left corner of the 2D map plane is taken as the origin, and the two adjacent right-angled sides passing through the origin are the x-axis and y-axis, thus forming a 2D pixel coordinate system. The sheet metal part is regarded as a moving object, and the bending point of the sheet metal part is regarded as the moving point. Then the loading and unloading movement path of the sheet metal part includes several path points, each of which is represented as (x, y, θ). Among them, (x, y) are the position parameters of the sheet metal part in the 2D pixel coordinate system; θ is the attitude angle of the sheet metal part. Step 2: Obtain the initial motion trajectory for unloading and loading sheet metal parts: Using the RRT-Connect algorithm, the initial motion path for unloading and loading sheet metal parts is planned to obtain an initial motion trajectory for unloading and loading sheet metal parts; the initial planning is repeated N times. pop Then, we get N. pop Initial movement trajectory of sheet metal parts during loading and unloading; Step 3: Calculate path safety: Identify all pixels in the upper and lower dies of the sheet metal bending machine that are likely to collide with the bending motion of the sheet metal part or are within the safe distance, and mark them as obstacles of interest; based on the obstacles of interest and the initial motion trajectory of the sheet metal part loading and unloading, and based on the swept volume signed distance field SVSDF, calculate the path safety value of each initial motion trajectory of the sheet metal part loading and unloading. Step 4: Calculate Path Cost: Assuming any initial movement trajectory for loading and unloading sheet metal parts has N path points, the specific formula for calculating the path cost is as follows: In the formula, Δx i and Δy i Let i be the difference in the horizontal and vertical position parameters between two adjacent path points i and i+1; where 1≤i≤N-1; Δθ i The difference in the attitude angle of the sheet metal parts between two adjacent path points i and i+1; β is the attitude angle weighting coefficient, which is a set value; Step 5: Calculate the COST t Calculate the fitness value (COST) of the initial motion trajectory for loading and unloading the t-th sheet metal part. t Where 1≤t≤N pop Then COST t The calculation formula is: COST t =γ1path cost+γ2path safety In the formula, γ1 is the cost weighting coefficient, and 1≤γ1≤1.5 is a set value; γ2 is the safety weight coefficient, and 1≤γ2≤1.5, which is a set value; Step 6: Obtain the optimal motion trajectory for unloading and loading sheet metal parts: Based on the Water Circulation Algorithm (WCA), continuously iterate and optimize to find N in Step 5. pop Cost t Find the minimum value and set the minimum COST. t The corresponding sheet metal part loading and unloading trajectory is taken as the optimal loading and unloading trajectory for sheet metal parts.

2. The robot-assisted sheet metal bending loading and unloading path planning method based on SSVSDF according to claim 1, characterized in that: In step 2, the RRT-Connect algorithm is an improved synchronous bias greedy RRT-Connect algorithm. When initially planning the loading and unloading movement path of sheet metal parts, the improved synchronous bias greedy RRT-Connect algorithm uses an adaptive elliptical region to sample random points and performs adaptive growth of each tree node based on the influence of obstacles.

3. The robot-assisted sheet metal bending loading and unloading path planning method based on SSVSDF according to claim 2, characterized in that: Step 2, which uses an adaptive elliptical region to sample each random point, includes the following steps: Step 2A-1: Establish an adaptive elliptical region: Set the starting point S of the sheet metal part loading / unloading motion path. start (x start ,y start ,θ start ) and target point S goal (x goal ,y goal ,θ goal The two foci of the adaptive elliptical region are defined as follows: the length l from the upper die to the lower die of the bending machine is defined as the minor axis length c of the adaptive elliptical region. short This allows for the construction of an adaptive elliptical region that changes adaptively with the starting and target points. Step 2A-2: Construct a unit circle: Using the origin of the two-dimensional pixel coordinate system in Step 1 as the center, construct a unit circle with a radius of 1 pixel; Step 2A-3, Random Point Stretch: Randomly generate a point (x) within the unit circle. c ,y c ,θ rand ), and stretch it to a point (x, y, θ) within an adaptive elliptical region. rand ); where the formulas for calculating x and y are: In the formula, c long To adapt to the major axis length of the elliptical region, based on c short The focal length between the two focal points is calculated. Step 2A-4, Random Point Rotation: Based on the two foci of the adaptive elliptical region, rotate the point (x, y, θ) rand If the point is rotated, the rotated point will be the random sampling point S. rand (x rand ,y rand ,θ rand ); where x rand and y rand The calculation formula is: Where R is the rotation matrix, its expression is: In the formula, α represents the point (x, y, θ). rand The rotation angle of ) is calculated using the following formula: α=arctan(|x start -x goal | / |and start -and goal |)。 4. The robot-assisted sheet metal bending loading and unloading path planning method based on SSVSDF according to claim 3, characterized in that: Step 2, the adaptive growth method for each search tree node, includes the following steps: Step 2B-1: Calculate the direction v0 of tree node expansion affected by unobstructed access. The specific calculation formula is as follows: In the formula, x nearest and y nearest For the closest random sampling point S rand Spanning tree node S nearest Position parameters; Step 2B-2: Find the node S that influences the newly generated tree. new The total number of obstacles N o : Using random sampling points S rand Centered on a square with a side length of a set number of pixels 'a', construct a square. The total number of pixels where the square intersects with either the upper or lower die of the bending machine is denoted as the number of pixels affecting the newly generated tree node S. new (x new ,y new ,θ new The total number of obstacles N) o ; Step 2B-3: Calculate obstacle pair S new Direction of influence v obs The specific calculation formula is as follows: In the formula, x k and y k Let N be the pixel coordinates of the k-th obstacle; where 1 ≤ k ≤ N0; Step 2B-4: Calculate the expansion direction v of the search tree node affected by obstacles. The specific calculation formula is as follows: v=γ3v0+γ4v obs in: γ3+γ4=1 In the formula, γ3 and γ4 are the set weighting coefficients; Step 2B-5, Calculate x new and y new : with distance S rand Recent S nearest To expand the starting point, a set equidistant step size r is adopted. p Extend in the direction of v; where x new and y new The specific calculation formula is as follows: Step 2B-5, Calculate θ new : with θ nearest Using the set constant angular step size r as the initial value. o To θ rand To expand upon this, the specific calculation formula is as follows: i new =r o (i rand -θ nearest )+θ nearest In the formula, θ nearest For the nearest random sampling node S rand Spanning tree node S nearest The posture angle of the sheet metal parts; Step 2B-6: Determine if a collision occurs during the tree node expansion process. If a collision occurs, resample; otherwise, add the new node S. new Add to the search tree.

5. The robot-assisted sheet metal bending loading and unloading path planning method based on SSVSDF according to claim 1, characterized in that: Step 3, the method for calculating path safety, specifically includes the following steps: Step 3-1: Calculate the swept volume: Using the Shape function, calculate the volume from the starting point S for any initial motion trajectory of the sheet metal part during loading and unloading. start Move to target point S goal The total number of pixels occupied at that time is recorded as the sweep volume; Step 3-2: Construct a sweep volume binary image: Use the Map function to input the sweep volume obtained in Step 3-1 into a blank two-dimensional map plane and mark it as black, thereby forming a sweep volume binary image; Step 3-3: Construct the SVSDF matrix: Calculate the SVSDF value of the sweep volume for the coordinates of any pixel in the sweep volume binary image constructed in Step 3-2, thereby forming the SVSDF matrix; Steps 3-4: Find obstacles of interest: The upper and lower dies of the sheet metal bending machine are obstacles in the bending motion of the sheet metal parts. From the two-dimensional map plane constructed in Step 1, find all pixels in the upper and lower dies that are likely to collide with the bending motion of the sheet metal parts or are within the safe distance, mark them as obstacles of interest, and record the pixel coordinates of each obstacle of interest. Step 3-5: Obtain the SSVSDF value of the obstacle point of interest: Based on the pixel coordinates of the obstacle point of interest in Step 3-4, find the corresponding SSVSDF value from the SSVSDF matrix constructed in Step 3-3, thereby obtaining the SSVSDF value of each obstacle point of interest; Steps 3-6: Calculate the average SSVSDF value of the obstacle points of interest, average_sdf; Steps 3-7: Using the normal distribution, determine the location of average_sdf and the probability density value pdf_value; Step 3-8: Calculate path safety, specifically: path safety = K × pdf_value; where K is an order of magnitude adjustment factor.

6. The robot-assisted sheet metal bending loading and unloading path planning method based on SSVSDF according to claim 5, characterized in that: In step 3-1, the initial motion trajectory of the t-th sheet metal part includes N path points, namely: starting point S start The second path point S2, the third path point S3, ..., the i-th path point S i The (N-2)th path point S N-2 and target point S goal ; Where 2≤i≤N-2; then the sheet metal part at path point S i The number of pixels occupied by the time is: Shape(S) i ).

7. The robot-assisted sheet metal bending loading and unloading path planning method based on SSVSDF according to claim 6, characterized in that: In step 3-2, the swept volume binary image SV_map of the initial motion trajectory of the t-th sheet metal part during loading and unloading is:

8. The robot-assisted sheet metal bending loading and unloading path planning method based on SSVSDF according to claim 7, characterized in that: In step 3-3, the formula for calculating SSVSDF is: In the formula, d(SV_map) is the distance from any pixel in the sweep volume binary image to the nearest pixel in the sweep volume, in pixels; It is the complement of d(SV_map).

9. The robot-assisted sheet metal bending loading and unloading path planning method based on SSVSDF according to claim 7, characterized in that: In steps 3-4, before finding the obstacle of interest, the upper and lower dies of the sheet metal bending machine are divided into narrow and non-narrow regions. The narrow region refers to the area between the bottom protrusion of the upper die and the groove of the lower die; the non-narrow region refers to the other bending movement areas besides the narrow region. Therefore, the obstacle of interest includes the obstacle of interest in the narrow region and the obstacle of interest in the non-narrow region. The obstacle points of interest in the narrow area include pixels located at the bottom outer edge and the outer ring of the upper mold within the narrow area, and pixels located at the top outer edge and the outer ring of the lower mold within the narrow area; Obstacles of interest in non-narrow areas include pixels on the outer edge of the upper or lower mold that are located in non-narrow areas and are prone to collision with sheet metal parts.

10. The robot-assisted sheet metal bending loading and unloading path planning method based on SSVSDF according to claim 9, characterized in that: In steps 3-6, average_sdf includes the average SVSDF value of obstacle points of interest in narrow regions, average_sdf1, and the average SVSDF value of obstacle points of interest in non-narrow regions, average_sdf2. In steps 3-7, pdf_value includes the probability density value of the narrow region pdf_value_narrow and the probability density value of the non-narrow region pdf_value__nonnarrow. The specific calculation formula is as follows: In the formula, μ1 is the desired safe distance in the narrow area, and is a set value; μ2 is the expected safe distance in non-narrow areas, which is a set value; σ1 is the width adjustment value of the normal distribution curve corresponding to the SSVSDF value of the obstacle point of interest in the narrow region, which is a set value; σ2 is the width adjustment value of the normal distribution curve corresponding to the SSVSDF value of the obstacle point of interest in the non-narrow region, which is a set value.

Images