Path planning method of pulmonary vessel targeting robot based on improved RRT* algorithm

Abstract

The invention discloses a pulmonary vessel targeted robot path planning method based on an improved RRT * algorithm, and belongs to the technical field of robot path planning. The method comprises the following steps: firstly, loading a pulmonary blood vessel STL model and initializing a space boundary; then, candidate sampling points are generated by adopting a self-adaptive strategy; then carrying out accurate in-vivo collision detection to ensure that all nodes and path sections are strictly located in a blood vessel cavity; carrying out node expansion and progressive path optimization on the detected sampling points by utilizing neighborhood search of an RRT * algorithm and an optimal father node selection and reconnection optimization mechanism; and finally, performing post-processing on the initial path through a greedy redundant node pruning strategy, deleting unnecessary turning points, and further shortening the length of the path. The method effectively solves the problems of low sampling efficiency, slow convergence and poor path quality during intravascular planning of a traditional RRT * algorithm, can quickly generate a safe and smooth intravascular navigation path with an approximately optimal length, and is suitable for preoperative planning of a lung interventional operation.

Description

Technical Field

[0001] This invention belongs to the field of robot path planning technology, specifically relating to an improved RRT* algorithm for pulmonary vascular targeting robot path planning. Background Technology

[0002] Intravascular targeted therapy is an important method in modern medicine for treating vascular diseases such as those of the heart, brain, and lungs, especially lung diseases like pulmonary thrombosis. In these procedures, surgeons guide medications or catheters to the lesion site via a vascular pathway. Precise pathway planning before surgery is crucial for improving surgical success rates, reducing surgical time, and lowering costs and risks.

[0003] The Rapid Random Tree (RRT) algorithm is a commonly used path planning algorithm, widely applied in robotics due to its probabilistic completeness and asymptotic optimality. However, traditional RRT algorithms face several challenges when applied to intravascular path planning: 1. Low sampling efficiency: Traditional RRT randomly and uniformly samples in a vast free space, while the blood vessel cavity only occupies a small part of the entire space. Therefore, a large number of sampling points fall in invalid areas, resulting in serious waste of computational resources.

[0004] 2. Slow convergence speed: Due to blind sampling, the algorithm needs a lot of iterations to find a feasible path, and finding the optimal path takes even longer, which makes it difficult to meet the real-time requirements of clinical practice.

[0005] 3. Lack of specific adaptation to anatomical structures: Traditional collision detection methods have difficulty accurately handling complex, unstructured biomedical anatomical structures such as STL models, and are prone to planning invalid paths that "pass through walls".

[0006] 4. Unstable path quality: In complex and narrow vascular networks, the algorithm is prone to getting trapped in local optima, resulting in detours or uneven paths, which do not meet the actual needs of clinical operation.

[0007] While the traditional RRT* algorithm ensures asymptotic optimality through parent node reselection and reconnection, it requires frequent local reconstruction of the tree structure, increasing computational costs and making it prone to getting trapped in local optima. Therefore, there is an urgent need in this field for a new method that can perform efficient and accurate path planning within a three-dimensional pulmonary vascular model. Summary of the Invention

[0008] [Technical Issues] The technical problem to be solved by this invention is: how to overcome the shortcomings of the existing RRT algorithm, such as blind sampling and low efficiency, as well as the slow convergence speed, high computational cost and easy getting trapped in local optima caused by global random sampling when pursuing asymptotic optimization, in order to quickly and stably generate a high-quality path that is strictly located within the vascular lumen and has an approximately optimal length.

[0009] [Technical Solution] To address the aforementioned technical problems, this invention proposes an improved RRT* algorithm for path planning of a lung vascular targeting robot. Its core lies in combining a strategy that integrates multiple sampling methods with accurate in vivo collision detection, specifically including the following steps: Step 1: Load the STL model file of the lung vessels and perform spatial initialization, obtain the spatial boundary of the model, and define the start and end points of the path; Step 2: Generate candidate sampling points using a multi-strategy fusion sampling method; Step 3: Perform in vivo collision detection on the candidate sampling points generated in Step 2 based on the parity principle of ray intersection counting; Step 4: For the sampling points that pass the collision detection, perform node expansion and path optimization based on the RRT* algorithm framework; Step 5: Post-process the path obtained in Step 4 using a redundant node removal strategy to delete unnecessary nodes and optimize the path.

[0010] Optionally, step 1 includes: Read the set of all vertex coordinates of the STL model ; Calculate the starting point and the length, width, and height of the space based on the vertex coordinates:

[0011]

[0012] in, The starting point in space; The dimensions are the length, width, and height of the space.

[0013] Optionally, the multi-strategy fusion sampling in step 2 is an adaptive fusion of a global sampling strategy, a cone-shaped region sampling strategy, and a circular sampling strategy, and its sampling method is as follows: According to the preset weighting coefficients , , ,in In each sampling, with probability Choose to implement a global sampling strategy, based on probability. Choose to implement a cone-shaped region sampling strategy, with probability Choose to implement the circular sampling strategy; The global sampling strategy is as follows: within the spatial boundary defined by axisStart and axisLWH, uniform random sampling is performed on each coordinate axis to generate sampling points. ; The sampling strategy for the cone-shaped region is as follows: using the latest node of the current random tree... Point to target point A cone-shaped sampling region is constructed using the direction vector d as the central axis; an adaptive radius factor is employed. Dynamically adjust the sampling range, where The number of consecutive sampling failures; sampling points are randomly generated within the conical sampling area; The circular sampling strategy is as follows: starting from the current optimal path... and the end point Midpoint of the line With the center of the ball; Let be the sampling radius, where This is the current optimal path length. The random perturbation factor is within the interval [0.8, 1.2]; sampling points are randomly generated within the spherical region.

[0014] Optionally, the weighting coefficient , , These correspond to 0.3, 0.4, and 0.3 respectively.

[0015] Optionally, the in-body collision detection in step 3 includes: for the point P to be detected, constructing a ray and using a ray-triangle intersection algorithm; if the number of times the ray intersects the surface of the STL model is odd, then the point P is determined to be inside the model.

[0016] Optionally, step 3 further includes line segment-model collision detection: multiple points are uniformly sampled on the line segment connecting the two points for internality detection; if all points are located inside the model, the line segment is determined to be safe.

[0017] Optionally, step 4 includes: S4.1: Find the distance sampling point in the random tree T The nearest node ; S4.2: From Towards Directional expansion step size Generate new nodes ; S4.3: In the context of Centered on, with radius Search for the set of neighboring nodes within the neighborhood ; S4.4: From Select the node that minimizes the cumulative cost of the path as... Optimal parent node The cumulative cost is the cost from the root node to that node plus the cost from that node to... The distance; S4.5: To Nodes in the process perform reconnection optimization: if via If the path to a neighboring node has a lower cost, then update the parent node of that neighboring node to [the newer one]. .

[0018] Optionally, the redundant node removal strategy in step 5 is a greedy pruning strategy, specifically including: Given an initial path point sequence ,in Starting from, The endpoint; From the starting point Begin by trying nodes that are not directly adjacent to each other. Establish a direct connection; from Begin by checking if it can be associated with a subsequent node in the sequence. To achieve collision-free direct connection, among which >Current index +1; If possible, delete. and All intermediate nodes between, and As the new current node; Repeat the above checking and deletion process until the current node becomes the endpoint. This yields the optimized path.

[0019] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the lung vascular targeting robot path planning method of the improved RRT* algorithm as described above.

[0020] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the improved RRT* algorithm for pulmonary vascular targeting robot path planning as described above.

[0021] [Beneficial Effects] 1. Significantly Improved Sampling Efficiency and Convergence Speed: This invention innovatively employs an adaptive sampling strategy, integrating global, cone-shaped, and circular adaptive sampling strategies to overcome the blind sampling defects of traditional RRT / RRT* algorithms. Cone-shaped region sampling guides the random tree to grow rapidly towards the target, significantly shortening the initial path discovery time. In particular, the circular sampling strategy dynamically adjusts the sampling region based on the current optimal path length, enabling the algorithm to automatically focus on searching the space near better solutions in the later stages of operation. Experimental results show that this mechanism effectively reduces invalid sampling and drives faster path length convergence.

[0022] 2. Guaranteeing absolute path safety and asymptotic optimization: This invention employs precise in vivo collision detection based on the ray intersection principle, ensuring that all nodes and path segments are strictly located within the blood vessel cavity, eliminating the risk of "passing through walls." Within the RRT* algorithm framework, through optimal parent node selection and reconnection optimization mechanisms, the overall path length can be continuously and progressively shortened.

[0023] 3. Further optimization of path smoothness and economy: This invention systematically eliminates unnecessary turning points in the path through a greedy redundant node pruning and processing strategy, further compressing the path length while ensuring safety, so that the final path is closer to the theoretical shortest centerline and better meets the actual needs of clinical operation.

[0024] In summary, through the synergistic effect of the above-mentioned innovative points, this invention effectively solves the contradiction between efficiency and optimality in the prior art, and can quickly plan a safe, smooth and approximately optimal intravascular navigation path, providing reliable technical support for precise navigation in vascular interventional surgery. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention, 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 The overall flowchart of the path planning method for lung vascular targeting robots based on an improved RRT* algorithm provided by this invention is shown in the present invention.

[0027] Figure 2 This is a schematic diagram of the global sampling strategy of RRT* provided by the present invention.

[0028] Figure 3 This is a schematic diagram of the cone-shaped region sampling strategy provided by the present invention, showing a cone-shaped sampling region based on the latest node and the target point.

[0029] Figure 4 This is a schematic diagram of the circular sampling strategy provided by the present invention, showing the sampling area centered on the current optimal path estimation point.

[0030] Figure 5 This is a schematic diagram illustrating the principle of the collision detection algorithm provided by the present invention.

[0031] Figure 6 This is a schematic diagram illustrating the process of expanding nodes, searching neighborhoods, and selecting the optimal parent node in the RRT* algorithm provided by this invention.

[0032] Figure 7 A schematic diagram illustrating the principle and process of the greedy redundancy pruning strategy provided by this invention.

[0033] Figure 8 The diagram shows the planning results of the traditional RRT algorithm, RRT* algorithm, and the algorithm of this invention in the same model.

[0034] Figure 9 A comparison diagram showing the original path and the final path generated by the method of the present invention in the model. Detailed Implementation

[0035] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. 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] First, the nature and origin of the three-dimensional blood vessel model in the embodiment will be explained: The core of this invention is path planning within the passable cavities (i.e., lumen) of pulmonary blood vessels. Therefore, the three-dimensional geometric model used must accurately describe the boundaries of this cavity. Typically, this model is derived from the processing of medical imaging data such as computed tomography angiography (CTA) or magnetic resonance angiography (MRA) of a patient's lungs. Using specialized medical imaging software such as Mimics and 3D Slicer, vascular segmentation, three-dimensional reconstruction, and surface meshing are performed to generate an STL format model representing the surface of the vascular lumen. This invention is not sensitive to the shape of triangles but requires a mesh manifold; a coarse mesh will affect collision results, so the mesh must be preprocessed, i.e., smoothed. The algorithm assumes that the vascular lumen is a connected three-dimensional space without isolated internal cavities; that is, starting from any point in the lumen, one can reach any other point within it without traversing the model surface. This model is a closed curved shell composed of numerous triangular facets, with no thickness of its own. The internal space enclosed by this surface is defined as the vascular lumen passable by the robot. All path points and path segments in this invention must be located within this "internal space".

[0037] During the algorithm development and testing phases, to simplify the process, 3D modeling software (such as 3D Max) can be used to manually construct a vascular cavity STL model with a similar topological structure for verification through geometric splicing. Regardless of the source, the STL model processed in this invention is mathematically considered as a closed triangular mesh surface whose interior is a feasible region.

[0038] Example 1 This embodiment provides an improved RRT* algorithm for path planning in a lung vascular targeting robot, such as... Figure 1 As shown, the overall process of this method is as follows: First, model loading and spatial initialization are performed, followed by the main planning loop. In the loop, a multi-strategy fusion sampling method is used to generate candidate sampling points, and precise in vivo collision detection is performed on these points. If the detection passes, new nodes are expanded on the random tree, and the neighborhood search, optimal parent node selection, and reconnection optimization of the RRT* algorithm are executed to update the tree structure and progressively optimize the path. When the tree first connects to the target point, an initial path is recorded, and optimization continues based on this path. After planning, a redundant node removal strategy is executed on the obtained path, finally outputting an optimal, collision-free path located within the blood vessel lumen. Specific steps are as follows: Step 1: Load the STL model file of the lung vessels and perform spatial initialization, obtain the spatial boundary of the model, and define the start and end points of the path; Specifically, read the coordinates of all vertices in the STL model to form a vertex set:

[0039] Based on this set of vertices, the spatial boundary of the model is calculated. The starting point of the space is defined as the minimum value of the coordinates of all vertices on each coordinate axis:

[0040] The length, width, and height of the space are defined as the maximum span along each coordinate axis:

[0041] Therefore, axisStart and axisLWH jointly define an outer cubic space that exactly surrounds the entire blood vessel model, restricting subsequent sampling operations to this effective space and improving space utilization.

[0042] Step 2: Generate candidate sampling points using a multi-strategy fusion sampling method; Traditional RRT* algorithms involve completely random sampling across a wide space, resulting in low efficiency. To address this, this invention proposes a multi-strategy adaptive fusion sampling method to dynamically balance global exploration and local optimization. This sampling method integrates three strategies: global sampling, cone-shaped region sampling, and circular sampling. The fusion mechanism involves assigning a set of weight coefficients to the three strategies (…). , , ,and In each iteration of the algorithm, one strategy is randomly selected based on these weights to generate candidate sampling points for the current iteration. In the long run, the frequency of selection for each strategy approximates its weight, thus achieving a fusion and balanced control of the advantages of the three strategies overall. In a preferred embodiment, based on the results of comparative experiments, [the algorithm] sets... =0.3, =0.4, A weighting of 0.3 achieves the best balance between global exploration, goal-oriented growth, and local optimization in pulmonary vascular pathway planning tasks, resulting in superior average performance. The selection mechanism is as follows: 1. Calculate the cumulative probability interval: Based on the weighting coefficients, define three consecutive cumulative probability intervals: Circular sampling strategy range:

[0043] Conical region sampling strategy interval:

[0044] Global sampling strategy range:

[0045] Substituting the preferred weight values, the specific range is: .

[0046] 2. Random selection strategy: Each time candidate sampling points need to be generated, a random number uniformly distributed in the range [0, 1] is first generated. .

[0047] 3. Perform the corresponding sampling: like Then, the circular sampling strategy is invoked to generate candidate sampling points. .

[0048] like Then, the cone-shaped region sampling strategy is invoked to generate candidate sampling points. .

[0049] like Then, the global sampling strategy is invoked to generate candidate sampling points. .

[0050] like Figure 2 As shown, the global sampling strategy aims to maintain the algorithm's global exploration capability. Within the spatial boundaries axisStart and axisLWH defined in step 1, it independently performs uniform random sampling on the X, Y, and Z coordinate axes to generate candidate sampling points. The specific formula is as follows:

[0051] in, This represents the position of a candidate node randomly generated during the path planning process. This is a random function that returns a random floating-point number that follows a uniform distribution over the interval [0,1]. and yes The maximum and minimum values ​​of the axis. for The total span of the axis, multiplied by a random number, gives a value in [0, ..., ... A random offset within the range, and add this random offset to the lower limit. Above, ultimately landing on [ One within the interval The random coordinates. Similarly, this strategy uses uniform random sampling within the entire initialization space boundary to ensure the algorithm's global exploration capability.

[0052] like Figure 3 As shown, the cone-shaped region sampling strategy aims to guide the random tree to grow efficiently in the target direction. Let the latest node of the current random tree be... The target point is First, calculate the growth direction vector d:

[0053] This vector serves as the central axis of the conical region, around which the entire cone is generated, and all sampling points are referenced to this direction. To dynamically adjust the exploration range in narrow and tortuous vascular environments, an adaptive radius factor is introduced:

[0054] in, This represents the number of consecutive sampling failures, i.e., the number of times the newly extended edge hits the blood vessel wall; 0.3 is the base scaling factor, which determines... The overall magnitude controls the size of the cone radius; 0.5 is the minimum offset, even if there are no obstacles during the algorithm's execution. , There is also a baseline value that maintains the most basic exploration breadth, preventing the cone from being too small and missing the optimal feasible path; 50 is the normalized denominator for the number of failures, meaning that after 50 consecutive failures, each It takes several attempts to reach 1, ensuring a certain tolerance for failure.

[0055] This design allows for an appropriate expansion of the sampling range to find alternative paths when encountering consecutive obstacles, while narrowing the range to accelerate progress when the path is clear. Subsequently, a normalized orthogonal basis is constructed based on the direction vector d. The formula for constructing the orthogonal basis is as follows:

[0056] in, according to The size is chosen to be [1,0,0] T Or [1,0,0] T To avoid numerical singularities, sampling points are randomly generated within the cone-shaped region:

[0057] in,

[0058]

[0059] in, The diagonal vector of the environment boundary. It is the preset maximum cone angle, which is the deviation of the sampling point from the principal axis. Angle; The azimuth angle is perpendicular to the principal axis. Within the plane, the rotation angle of the sampling point around the principal axis; Let be the radius. This strategy guides the algorithm to explore in a biased direction towards the target, accelerating convergence.

[0060] like Figure 4As shown, after the algorithm has found an initial path from the starting point to the ending point, a circular sampling strategy is used to perform a refined search in the vicinity of this path to further optimize the path. The core idea of ​​this strategy is to dynamically focus the sampling area on the spatial range where a shorter path is more likely to be found as the search progresses, thereby improving the efficiency of local optimization. The specific implementation steps are as follows: 1. Initialization: At the start of the algorithm, before any feasible path has been found, the sampling radius is set to a circle. Set a sufficiently large initial value, such as half the diagonal length of the entire planning space, to ensure global exploration capability.

[0061] 2. Dynamically Update Radius: Whenever the algorithm updates the current optimal path, i.e., when it finds a shorter feasible path than the existing records, it updates the radius based on the length of the current optimal path. Dynamically adjust sampling radius A preferred update method is to: Set as ,in Let be a random factor that fluctuates within the interval [0.8, 1.2].

[0062] In this design, This provides a reasonable benchmark for the sampling radius that is proportional to the current optimal path length. (Perturbation factor) The design aims to balance the need for fine-grained local search with the need to avoid getting trapped in local optima: when When the sampling range is in the interval [0.8, 1), the sampling region is appropriately shrunk to promote in-depth optimization near the current optimal path; when Within the range [1, 1.2], a moderate expansion of the sampling region helps to explore potentially better detour paths and prevents the omission of globally optimal solutions due to premature shrinkage of the region. This mechanism ensures that the sampling region can adaptively focus on promising spaces while avoiding the loss of necessary exploration capabilities due to over-focusing.

[0063] This means a better (shorter) path This corresponds to a smaller sampling area, guiding the algorithm to perform a finer search within a more compact space.

[0064] 3. Determine the sampling center: the center of the sphere for circular sampling. Defined as the starting point of the current optimal path and the end point The midpoint of the line, i.e. .

[0065] 4. Generate sampling points: as described above For the center of the ball, the updated Candidate sampling points are randomly generated within a spherical region with radius . The random points generated in spherical coordinates will be converted to rectangular coordinates:

[0066] in, It is the azimuth angle. Polar angle, Indicates the radius of the circular region. These represent the sampling points at... axis, axis, The distance along the axis, therefore the formula for generating circular sampling points is:

[0067] Through the above mechanism, the circular sampling region can adaptively shrink and anchor itself near the current optimal path. Since the sampling center is always located at the midpoint of the line connecting the start and end points, and the radius always covers this baseline region, the probability of finding a feasible path in the connected space is maintained. The shorter the path, the smaller and more focused the sampling region, thereby effectively reducing random sampling in vast invalid regions and driving the algorithm to converge efficiently towards the asymptotically optimal solution.

[0068] In each planning loop, the algorithm randomly selects a sampling strategy to execute based on the set weight coefficients, thereby combining the advantages of the three strategies to efficiently generate candidate sampling points.

[0069] Step 3: Perform in vivo collision detection on the candidate sampling points generated in Step 2 based on the parity principle of ray intersection counting; To ensure that the planned path is strictly located within the passable vascular cavity, this embodiment performs rigorous collision detection on the sampling points generated in step 2 and the subsequently generated path segments. The collision detection in this embodiment is based on the parity principle of ray intersection counting. Here and below, "inside the model" refers to the closed space enclosed by the STL triangular mesh surface loaded in step 1, i.e., the vascular cavity.

[0070] like Figure 5 As shown in the upper part, for the point P to be detected, it is necessary to determine whether it is inside the cavity defined by the STL model. Construct a ray R starting from point P. To avoid ambiguity when the point is exactly on the model surface, slightly offset the starting point of the ray: ,in It is a very small positive value.

[0071] For each triangular facet in the STL model The ray-triangle intersection algorithm is used to determine whether ray R intersects with it. This algorithm is implemented by solving a system of linear equations: make First, calculate the auxiliary vector: ,like If the ray is parallel or approximately parallel to the plane of the triangle, it is considered non-intersecting. Otherwise, calculate: ,if or If they are not intersecting, then they are considered disjoint. Calculate... ,if or The intersection points are determined to be non-intersecting. Finally, the parameters along the ray at the intersection point are calculated. ,if If the rays intersect from the front of the triangle, then the triangle is considered to be intersecting.

[0072] According to the point-in-Polyhedron test principle in computational geometry, if the total number of intersections between rays emanating from point P and the surface of the closed STL model is odd, then point P is determined to be inside the model (i.e., inside the blood vessel cavity); if the total number is even, then it is located outside (outside the blood vessel cavity or inside the blood vessel wall entity).

[0073] like Figure 5 As shown in the lower section, for a line segment connecting two points A and B, it is necessary to determine whether it is completely inside the model. To ensure the reliability of the detection, this embodiment uses a method of uniformly sampling a fixed number of points on the line segment for verification. Based on the parameters... Defined line segment Select at equal intervals above Number of sampling points, i.e., sampling interval is For each sampling point Perform the point inclusion detection described above. If all sampling points are determined to be inside the model, the line segment is considered safe and collision-free; if any sampling point is located outside the model, the line segment is determined to have collided with the blood vessel wall. In this embodiment, the preferred method is... The value is 10.

[0074] This fixed-number sampling strategy is an optimal solution that strikes a balance between detection reliability and computational efficiency. (Number of sampling points) The algorithm can be adaptively adjusted based on the geometric complexity of the vascular model and the planning step size. This is based on the fact that in vascular path planning, the step size for a single expansion is usually limited. For typical pulmonary vascular curvature characteristics and a preset step size, it has been verified that a certain number of sampling points is sufficient to reliably identify potential interference between line segments and the vessel wall in most cases. Using a fixed number of sampling points, rather than dynamically adjusting based on length, simplifies the computational logic and improves the overall efficiency of the algorithm. For regions with particularly large curvature or longer detection line segments, those skilled in the art can appropriately increase the number of sampling points. In exchange for higher collision detection confidence.

[0075] Step 4: For the sampling points that pass the collision detection, perform node expansion and path optimization based on the RRT* algorithm framework; For sampling points that pass collision detection This is then incorporated into a random tree for expansion and optimization. This process follows the basic framework of the RRT* algorithm, such as... Figure 6 As shown, the specific steps are as follows: S4.1: In the constructed random tree T, find the distance from the sampling point. The nearest node is denoted as :

[0076] S4.2: From Towards Extend the direction by a fixed step size r to obtain a new node. :

[0077] In generation Beforehand, the line segment collision detection method in step 3 needs to be used to ensure that the line segments ( , ) Safe and collision-free. The step size r is a preset empirical value, which is set to r=15 in this embodiment. Its value needs to be balanced between exploration efficiency and expansion granularity: too small a step size will lead to slow convergence and a surge in the number of iterations; too large a step size may cross narrow cavities, reducing feasibility and exploration ability in complex vascular topologies.

[0078] S4.3: In the context of Centered on, with radius Within the spherical neighborhood, search for all potential neighbor nodes in the random tree T to form a set. Neighborhood radius The selection of is related to the step size r, and is usually set to a multiple of r. In this embodiment, it is set as follows: This proportional relationship was determined through experimental optimization: if If the value is too large, too many irrelevant nodes will be included in the cost calculation, severely impacting real-time performance; if If the value is too small, the number of optimized paths that new nodes can connect to is limited, which weakens the incremental optimization capability of the RRT* algorithm.

[0079] S4.4: Traversal For each section q in the equation, calculate the distance from the starting point via q. Path cumulative cost: The goal is to obtain from neighboring nodes... Choose one that can make The parent node with the lowest path cost. The calculation method is as follows:

[0080] in, This represents the distance from the root node (starting point) to the node. The cumulative path length. Set as The parent node, and update .

[0081] S4.5: Traversal Each neighbor node in Consider if... As Calculate the new cost from the new parent node. If the cost is less than Current cost And the side If there is no collision, then parent node Updated to and update This step optimizes the connections between existing nodes in the tree, continuously reducing the overall path cost.

[0082] The above process is repeated until the random tree is first expanded to the target point. Within the neighborhood of the given path, an initial feasible path is obtained. The algorithm will continue to run more iterations, continuously sampling and optimizing, until the path length asymptotically converges to a better solution.

[0083] Step 5: Post-process the path obtained in Step 4 using a redundant node removal strategy to delete unnecessary nodes and optimize the path.

[0084] To further shorten the path and improve smoothness, this invention employs a greedy redundant node pruning strategy for post-processing.

[0085] Figure 7 This demonstrates the principle behind the strategy. Given an initial sequence of path points... ,in Starting from, The endpoint.

[0086] The pruning process starts from the beginning. start: The current node (Initial time) ) and its subsequent non-directly adjacent nodes ( ) Try to make direct connections one by one; Use the line segment collision detection method from step 3 to determine the line segment collision. Whether it is completely inside the blood vessel model and without collision; If we can find such a farthest, reachable node Then prove and All intermediate nodes between It is redundant; remove it from the path sequence. Will Set it as the new current node and repeat the above steps; Until the current node is the endpoint At that time, the pruning process ends.

[0087] The aforementioned greedy redundancy pruning strategy is an efficient post-processing optimization technique. Its core objective is to shorten the path length as much as possible while ensuring path safety (collision-free). It approximates a globally optimal solution by selecting the farthest safe node reachable from the current node at each step of the process. In geometrically complex connected cavities such as pulmonary vessels, this strategy effectively eliminates most unnecessary detours. This strategy is a heuristic approach aimed at obtaining significantly optimized and practical results. Its output path is close to the shortest in most cases, but it does not strictly guarantee a mathematically global optimum. Those skilled in the art will understand that under certain extreme geometric constraints, other more complex optimization algorithms may exist that can obtain shorter paths, but the method proposed in this invention achieves a good balance between optimization effectiveness and computational efficiency.

[0088] To objectively and intuitively verify the superior performance of the algorithm of this invention, this embodiment conducts a path planning comparison experiment on the same complex blood vessel STL model with a bifurcation structure, running the traditional RRT algorithm, the RRT* algorithm, and the improved RRT* algorithm proposed in this invention. The experimental results are as follows: Figure 8 As shown, the planning performance of the three algorithms exhibits a significant step-like difference. Figure 8 The algorithm clearly demonstrates the planning paths of the three algorithms under the same starting and ending point constraints.

[0089] The paths generated by the traditional RRT algorithm exhibit significant randomness and tortuosity (see...). Figure 8 (Top image). Due to its simple random sampling and nearest-point expansion mechanism, the path repeatedly meanders and turns sharply within the blood vessel lumen. Although it can eventually reach the target point, its path length is far from optimal, and the nodes are dense. In practical applications, such a path can lead to excessively long operation time, increased instrument wear, and even increased surgical risks.

[0090] The path of the traditional RRT* algorithm has been fundamentally improved (see...). Figure 8(The middle diagram). Thanks to its optimal parent node selection and reconnection optimization mechanism, the path is continuously optimized locally during the growth process, becoming more linear overall and significantly shorter. However, limitations of its sampling strategy still exist; the path still has unnecessary detours or bends in some areas, failing to fully utilize the geometric space of the blood vessel lumen, and still falling short of the theoretically shortest centerline.

[0091] In contrast, the improved RRT* algorithm of this invention demonstrates superior path planning performance (see...). Figure 8 (See the bottom image). Its path is the simplest and closest to the central axis of the blood vessel. This is directly attributed to the multi-strategy fusion sampling method proposed in this invention: global sampling ensures the completeness of the exploration; cone-shaped region sampling effectively guides the path to grow efficiently towards the target, avoiding blind exploration; while circular sampling, after finding the initial path, focuses on the region near the optimal path for refined searching, driving the path to continuously converge to a better solution. At the same time, accurate in vivo collision detection ensures that each step of the path expansion is strictly limited to the inside of the lumen, eliminating any risk of interference with the vessel wall.

[0092] Figure 9 This further reveals the powerful role of the post-processing optimization module within the algorithm of this invention. The figure compares the original path generated when the algorithm first connects to the target point with the final path obtained after post-processing steps such as redundant node removal in the same model. It can be clearly observed that although the original red path has a basic structure, it still contains minor, unnecessary nodes and turns in some straight sections. After post-processing using a greedy pruning strategy, the final path systematically removes these redundant nodes, further straightening the path by directly connecting distant nodes that are visually accessible and collision-free. This process significantly and considerably shortens the total length without compromising path safety, and further enhances the path's coherence and executability.

[0093] In summary, through a series of comparative experiments and visualization analysis, it has been fully demonstrated that the improved RRT* algorithm-based path planning method for pulmonary vascular targeting robots proposed in this invention successfully overcomes the inherent defects of traditional algorithms by innovatively integrating adaptive sampling strategies, performing rigorous in vivo collision detection, integrating the progressive optimization framework of RRT*, and introducing efficient path post-processing. It not only plans a safe and feasible path but also efficiently generates a high-quality path with approximately optimal length. This method provides an efficient, reliable, and practical technical solution for solving the key challenges of real-time and accurate navigation path planning in pulmonary vascular interventional surgery, and has significant clinical application value and promising prospects for wider application.

[0094] Example 2 This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the lung vascular targeting robot path planning method with the improved RRT* algorithm as described in Embodiment 1.

[0095] Example 3 This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the lung vascular targeting robot path planning method using the improved RRT* algorithm as described in Embodiment 1.

[0096] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A path planning method for lung vascular targeting robots based on an improved RRT* algorithm, characterized in that, Includes the following steps: Step 1: Load the STL model file of the lung vessels and perform spatial initialization, obtain the spatial boundary of the model, and define the start and end points of the path; Step 2: Generate candidate sampling points using a multi-strategy fusion sampling method; Step 3: Perform in vivo collision detection on the candidate sampling points generated in Step 2 based on the parity principle of ray intersection counting; Step 4: For the sampling points that pass the collision detection, perform node expansion and path optimization based on the RRT* algorithm framework; Step 5: Post-process the path obtained in Step 4 using a redundant node removal strategy to delete unnecessary nodes and optimize the path.

2. The method according to claim 1, characterized in that, Step 1 includes: Read the set of all vertex coordinates of the STL model ; Calculate the starting point and the length, width, and height of the space based on the vertex coordinates: in, The starting point in space; The dimensions are the length, width, and height of the space.

3. The method according to claim 1, characterized in that, The multi-strategy fusion sampling in step 2 is an adaptive fusion of the global sampling strategy, the cone-shaped region sampling strategy, and the circular sampling strategy. The sampling method is as follows: According to the preset weighting coefficients , , ,in In each sampling, with probability Choose to implement a global sampling strategy, based on probability. Choose to implement a cone-shaped region sampling strategy, with probability Choose to implement the circular sampling strategy; The global sampling strategy is as follows: within the spatial boundary defined by axisStart and axisLWH, uniform random sampling is performed on each coordinate axis to generate sampling points. ; The sampling strategy for the cone-shaped region is as follows: using the latest node of the current random tree... Point to target point A cone-shaped sampling region is constructed using the direction vector d as the central axis; an adaptive radius factor is employed. Dynamically adjust the sampling range, where The number of consecutive sampling failures; sampling points are randomly generated within the conical sampling area; The circular sampling strategy is as follows: starting from the current optimal path... and the end point Midpoint of the line With the center of the ball; Let be the sampling radius, where This is the current optimal path length. The random perturbation factor is within the interval [0.8, 1.2]; sampling points are randomly generated within the spherical region.

4. The method according to claim 3, characterized in that, The weighting coefficient , , These correspond to 0.3, 0.4, and 0.3 respectively.

5. The method according to claim 1, characterized in that, The in-body collision detection in step 3 includes: for the point P to be detected, constructing a ray and using a ray-triangle intersection algorithm; if the number of times the ray intersects the surface of the STL model is odd, then the point P is determined to be inside the model.

6. The method according to claim 5, characterized in that, Step 3 also includes line segment-model collision detection: multiple points are uniformly sampled on the line segment connecting the two points for internality detection. If all points are located inside the model, the line segment is determined to be safe.

7. The method according to claim 1, characterized in that, Step 4 includes: S4.1: Find the distance sampling point in the random tree T The nearest node ; S4.2: From Towards Directional expansion step size Generate new nodes ; S4.3: In the context of Centered on, with radius Search for the set of neighboring nodes within the neighborhood ; S4.4: From Select the node that minimizes the cumulative cost of the path as... Optimal parent node The cumulative cost is the cost from the root node to that node plus the cost from that node to... The distance; S4.5: To Nodes in the process perform reconnection optimization: if via If the path to a neighboring node has a lower cost, then update the parent node of that neighboring node to [the newer one]. .

8. The method according to claim 1, characterized in that, The redundant node removal strategy in step 5 is a greedy pruning strategy, specifically including: Given an initial path point sequence ,in Starting from, The endpoint; From the starting point Begin by trying nodes that are not directly adjacent to each other. Establish a direct connection; from Begin by checking if it can be associated with a subsequent node in the sequence. To achieve collision-free direct connection, among which >Current index +1; If possible, delete. and All intermediate nodes between, and As the new current node; Repeat the above checking and deletion process until the current node becomes the endpoint. This yields the optimized path.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the improved RRT* algorithm for lung vascular targeting robot path planning as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the improved RRT* algorithm for pulmonary vascular targeting robot path planning as described in any one of claims 1 to 8.

Images